There's not? Both AMD and Intel have GPUs that are competitive in certain market segments. AMD in particular is pretty much even in terms of rasterization performance on the high end. Intel is just new (again) to the game of discrete graphics.
Nvidia, on the other hand, invested tons and tons of money to make CUDA good over the last like 15 years, while AMD has not really put any serious effort into AI/ML workloads. It's mostly a software support problem, not a hardware problem to my understanding.
AMD is even on performance and significantly more affordable for consumers. but people wont stop buying nvidia's ripoff products like the 4060. really frustrating
Even more frustrating is on the Linux side. We've had a couple of threads about how bad Wayland is, when most of the problems are due to Nvidia binary drivers.
It's been 11 years since Linus gave Nvidia the finger. Nobody who is thinking of possibly running Linux should ever buy an Nvidia product. Yet Nvidia is still very commonly spec'd into Linux builds even today.
AMD cards are _very bad_ at running CUDA. If you're running linux on a desktop, there's a decent chance you're the type who wants to play with ML as well...
AMD has a translation layer called HIP as part of the ROCm toolset. This doesn't get us 100% of the way there, but it is an alternative that is gaining traction as they work to improve it.
KDE's Wayland session is fine on Nvidia, at least with my 3070. There used to be more visual artifacts, but everything I'm using works, from CUDA to Mesa to DLSS. The largest remaining bug I'm aware of has to do with how blurred windows are presented. It's pretty functionally solid, at least with newer drivers.
For ML workloads, AMD wasn't competitive due to lack of software support for PyTorch. Nvidia isn't making most of their profit from selling individual GPUs to gamers, they're making money from data center GPUs that are bought by the thousands at five figure prices each for ML training.
I made the same argument to myself when I bought an AMD GPU for gaming 4 years ago. What doesn’t show up in a side by side comparison is that AMD’s drivers (or Windows’ treatment of them) fucking suck. Until I bought an NVIDIA replacement recently I was very frequently having to reinstall the drivers and also intermittently encountering driver crashes in games and browsers.
That’s not to mention having to fiddle with the drivers/games more than I was accustomed to, which IIUC is more a consequence of having less market share and thus not being a priority for game developers to work out of the box. I can excuse that, but not crashes or reinstalling every couple months.
Completely agree. I bought an AMD GPU like 6 years ago. The software support fucking sucked. The drivers were bad, games weren't optimised for AMD. AMD looks great on paper in benchmarks. As a product, overall, it is bad. That's why people keep buying nvidia. I will gladly pay double for a product that works.
I am using a Radeon RX 6900 XT GPU since two years and never ever had any stability problem or crash running my PC ~10 hours/day, playing a some (various, demanding) games almost every day.
AMD might've gotten more traction if it wasn't schizophrenic with regards to compute support on its hardware. GCN was such a great architecture for heavy compute, yet it was mostly wasted by shit software.
Crypto really only managed to use it because there was more money to be made using AMD, justifying the wrangling needed to get compute to work on it.
Now they continue to waste their potential with ROCm's design choices which make its hardware support range very limited.
In my eyes, as someone who ran 150k+ AMD gpus for crypto... it really boiled down to being able to heavily tune various aspects of the GPUs. Each GPU is a snowflake and being able to tweak the knobs individually, really made a huge difference on performance. Super low level performance enhancements in code helped a lot too. I worked with some of the popular mining apps to improve things there.
AMD is already backporting ROCm to "lesser" hardware [0]. I wouldn't expect them to spend too many resources on going back further than that though. I'd rather see them improve their newer stuff. The larger issue is that it is impossible to rent time on the super high end GPUs that they use to build the super computers with... nobody offers that and it is something I'm working on fixing.
The thing I hear a lot from people who skew towards AMD is that raytracing and DLSS and all the software solutions that nvidia has are gimmicks, or somehow actually not a big deal.
They’re stuck on the concept of rasterized graphics being the ground truth in gpu value.
Then they’re shocked when the wider consumer market doesn’t hold their perspective.
AMD hasn’t had the big picture vision that nvidia has had. For example, with Vega AMD had a gpu architecture with exceptional async compute performance a generation before nvidia was competitive.
AMD then abandoned Vega, switched to a leaner and less compute friendly rDNA architecture while nvidia went in the opposite direction, pushed compute heavy features like raytracing and dlss and AMD hasn’t had an exceptional response since then.
Creating reflections using rasterization requires shortcuts and sacrifices. And once you see them, you can't unsee them, and then you find them visually distracting.
There are basically 3 ways to create reflections without ray-tracing:
1. Double the geometry and render it in the reflection. This is rarely the approach taken because it's incredibly resource intensive. Oddly, Duke Nukem 3D back in what, 1996? did this, but that had pretty simple graphics. Genuinely not feasible for any sort of reflect surface that is curved.
2. Screen-space reflection: Just take the rendered image, invert it, and overlay it onto the reflective surface. This is probably the most common way to do it. It's fast and looks decent. The problem is that it requires the reflected image to be visible on the screen. When it's not, there's no reflection. It creates a bizarre effect when, for example, you're looking over a body of water at something in the distance, and you see the reflection in the water, but as you move the camera down, the reflection gets cut off and eventually disappears entirely.
3. Using a static image as an environment mapped texture. Apex Legends does this for shiny surfaces. It creates a great illusion of a reflection, but taking even a second to actually look at it, you'll notice that the reflection doesn't actually show what it should unless you're standing in the exact right place.
I will concede that gamers typically don't notice the issues created using methods 2 and 3, especially in a game like Apex Legends where you're moving around too much to notice.
But me? I dunno. I notice and get distracted.
EDIT: DLSS, on the other hand, is basically a crutch. It's not surprise that DLSS became a big thing at the same time RTX did. RTX requires more juice to run at full resolution than the GPU has, so they introduced DLSS to render at a low resolution and upscale it.
4060 is one of the better-value options in the stack, actually. The RX 7600 is still on 6nm and is no faster, much less efficient, and skips many of the microarchitectural improvements to rdna3 to try and keep cost down.
$250-270 is fine for this type of product even today.
Similarly, the 4070 is also fairly attractively priced even compared to 6800XT/6950XT or 7800XT. More efficient, better feature set, same performance tier, similar perf/$. But people still get upset about intangibles like memory bus or die size (despite this already being measured in the performance!), or just don’t want to admit that wafer costs and R&D costs are rising in the post-Moores law era. Operating margins were actually down relative to the last few gens (until AI hit), and were broadly similar to 2012 levels. This is just what it costs now, and smaller die take the worst of the cost increases because memory controllers and pcie don’t shrink anywhere near as much as logic (the gpu cores). There is a “minimum cost floor” that has been rising substantially due to this.
CUDA’s definitely the “moat”, in the Warren Buffet sense - it’s not impossible for someone like AMD or Intel to make something CUDA compatible, but it might cost enough to wreck the economics of trying to compete in the space.
I don't think you will need something coda-compatible. Apple has relatively solid pytorch support without copying CUDA, but PyTorch+ROCm is still not working on windows (and I think there's no ROCm support for the consumer GPUs from AMD).
I think Apple proves that it is possible, and possible relatively quickly.
Pytorch supports directML on windows - is that not sufficient (or performant) for your usecase? It feels like AMD are less willing to step on Microsoft's toes, possibly as they just don't have the market share clout.
And Rocm supports the higher RDNA2 and the highest-end RDNA3 consumer GPUS - [0], though only the top RNDA3 cards in linux [1] (officially - other cards often do "work" but YMMV)
One thing Nvidia are much better at is supporting the whole stack - while I doubt many people are actually running much ML on 8gb mid-low end cards, it seems a hook for sales if people "could" do that, and a starting point for personal experimentation. I know multiple people who buy low end Nvidia GPUs because they "might" experiment with CUDA some time in the future.
> One thing Nvidia are much better at is supporting the whole stack - while I doubt many people are actually running much ML on 8gb mid-low end cards, it seems a hook for sales if people "could" do that, and a starting point for personal experimentation. I know multiple people who buy low end Nvidia GPUs because they "might" experiment with CUDA some time in the future.
but this is very important. Many develop locally before sending it to a cluster, especially if you only have limited credits available.
It’s important to have a vision of what you want to do as a company.
For as long as I can remember, Jensen has been telling us during all hands meetings that Nvidia is a visual computing solutions company, not a semiconductor company. (The “visual” part has only been dropped in the past few years.)
Nvidia sells hardware, but they also sell complete solutions, and everything in between.
I do care that Geforce experience works flawlessly. I mean, it doesn't, and my grandkids will probably still use DDU, but when shadowplay works it's indispensable as a gamer (or even for recording video calls.) I'm sure AMD has an alternative or I could use OBS, but I'd rather just pay a premium and rarely have to think about it. And also CUDA.
AMD is coming out with the MI300x on Dec 6th, which has 192GB and runs on the same hardware platform as H100s (OAM-UBB). They've doubled down on ROCm/Pytorch and other projects. Don't expect them to stay behind for much longer.
They also own a high end networking company and will sell you an entire rack or datacenter integrated for you. Also yeah, the decade or so they’re had thousands of in-house engineers and people around the world working on CUDA is HUGE. Also the crypto boom encouraged them to buy up tons of fab capacity. Nvidia had foresight on multiple levels here.
Nvidia’s problem is that every company is a credible threat because in the ongoing gold rush they can’t make enough shovels to meet demand and damn there is gold in them thar hills. People are going to do literally whatever it takes to make their competition’s products work. If you are selling a shovel, even a shitty shovel, you are a competitive threat to Nvidia right now.
There's also AMD - but it seems like a hard industry to get into. Intel keeps trying, most recently with Intel Arc[1], but they've never been able to (or perhaps wanted to) crack the high end.
Saying they've "never been able to" is a kind of weird perspective. They're really only just starting on their dedicated GPUs. The Alchemist architecture is their very first attempt, and they've been putting a lot of effort into repairing and refining the drivers. The hardware is actually pretty decent when it's firing on all cylinders.
But Nvidia and AMD have a huge time advantage on Intel. It's going to be a few generations and billions of dollars before Intel can catch up, assuming they keep with it.
Intel has been in the integrated GPU market for awhile but with little pressure to innovate and very different design goals.
AMD and Intel are competitive in gaming. Where Nvidia excels is other types of acceleration. AMD and Intel can't hold a candle to Nvidia when it comes the raytracing renderers used in every 3D and CGI workflow, for example.
AMD's Radeon group has, by any definition, kept up. In gaming, in the tier of cards that people actually buy, they trade blows with NVidia.
But making a GPU is hard. Making a good GPU is even harder. This becomes a chicken and egg problem. If NVidia's hardware has quirks, and you're a game developer, you optimize your game for Nvidia's hardware. If someone else tries and builds a GPU, they discover that many, popular, games run like sh*t. Because of tens of years of strange workarounds to ship a game, drivers rewrite commands for specific games, game devs abuse the directx spec or do it wrong.
However, Nvidia set out 10-15 years ago to corner the professional general purpose GPU compute market with a software layer called CUDA. Software was written and optmised for CUDA, and not any other generic graphics library for reasons I don't know (I'm not a graphics developer, just a gamer). So now Nvidia enjoys a moat in gpGPU (as it was called).
I do wonder how much of an impact the PS5 and Xbox Series have on this though since they both have RDNA2 GPUs. Consoles tend to face higher optimization pressure, especially as the generation wears on. It might be why AMD has kept so competitive in gaming as of late.
If you mean AI/ML/Cuda? I fully agree with you. AMD is suffering the consequences of their half hearted efforts with ROCm right now.
If you mean gaming? AMD has had better mid range gpus since the rx 480 was released in '16. Nvidia has been resting on their laurels and AMD has recently seemed to be fine playing the 'price is right' game where they price their gpus ~ 10% less than nvidia and don't even attempt to take market share.
AMD is committed to actively improving ROCm and directly supporting projects like Pytorch. They were caught with their pants down, but working hard to catch back up.
Nvidia got out in front with CUDA. They got a huge first-mover and vendor lock-in advantage with that one. AMD dropped the ball and is playing catch-up.
Gaming isn't the lucrative part of the business — it's selling data center GPUs for AI/ML. That's where Nvidia is raking in the cash.
It's in the article:
"...of the company's $18.12 billion in revenue, $14.51 billion was generated by its data center division."
They spent over a decade building up CUDA into the industry standard for GPU compute programming. That's why it's so hard for anyone to compete in this domain. AMD is doing better in gaming, but that's practically a sideshow business for Nvidia now.
I know, but my point is there were heavy incentives to compete with Nvidia. But nobody did, at least not with the same intent. And now they're (Nvidia) raking it in due to the AI/ML wave.
If others had comparable GPU capacity and know how, the market wouldn't be so skewed in favour of Nvidia.
Hardware development has much higher inertia than software development, and so you can't just chase the wind and see market-disrupting results in a year or two.
Nvidia's success is rooted in design and business choices made a long time ago. They were bets at that time, as ambitious hardware initiatives need to be, and other vendors like AMD made different bets. Nvidia won their bet more spectacularly than others, and now gets to benefit from their lead as others slowly work to reorient their ships in its direction.
Absolutely, positively nothing stops AMD from shipping affordable GPUs with more VRAM. It might not be a popular device - most people are buying $100-200 GPUs. But I would buy a 64GB 7900XTX, not a 24GB one.
The simple answer is, like many markets with few competitors, they like prices high. It's like a point of no return if they undercut their crappy workstation graphics.
Well there’s the looming threat of the fact that Nvidia is making margins that are way bigger so they can easily match any AMD undercutting and then undercut AMD if they want to.
The products that are red hot right now also use HBM not VRAM.
Right now, there's a big gap in nvidia's product offering: nvidia is happy to let their gaming/affordable cards languish, to stop them competing with the server products at 4x the price.
And the ML community isn't very motivated to port their code from nvidia to amd because the nvidia tax isn't too bad.
If AMD could take one of their existing 24GB consumer cards and quadruple the RAM, though? Open source ML efforts would have support within days.
And nvidia couldn't respond to undercut them without also undercutting their cash cow server GPUs.
AMD is kind of ironically hamstring by x86. They make a lot more profit off their CPUs than they do their GPUs and they only have so many TSMC wafers for both products. If they try to buy their way into more GPU market share they're losing margin on the GPUs they do sell and denying themselves more profitable CPU sales. And Nvidia has historically been pretty good about aggressively back filling any holes in their product stack AMD figures out how to exploit with a late product introduction. AMD seems like they've mostly given up on gaining in the discrete market.
Intel is making some better moves lately but the impression I get is they're losing money on every sale and need to buy their wafers from TSMC like AMD. I'm not sure how long they'll stomach that situation or if they'll eventually get their GPUs onto their own process.
> If AMD could take one of their existing 24GB consumer cards and quadruple the RAM, though? Open source ML efforts would have support within days.
They ironically could do exactly this because the MCD disambiguates the memory config from the gpu config. They would need a new MCD die but they could put four or 8 memory controllers on a MCD instead of 2.
Operating margins are actually down compared to previous gens, roughly similar to 2012 levels, until AI comes into the picture. And gaming cards haven’t increased their margins equivalently.
Nvidia isn’t going to sell gaming cards at an operating loss out of AI money. If the consumer market is soft then it’s soft, and with the general slowdown in Moores law cost progression it doesn’t make sense to have “firesale gens” that set an impossible standard for future gens to beat.
~20% perf/$ progression per generation is going to be about what’s sustainable, because that’s about what tsmc is giving the chip companies. You can “cheat” by using older worse nodes like ampere and turing did, and like AMD did with the RX 7600, to keep costs down, but that’s a one-time boost and consumers grow to expect that pricing going forward. It’s been a market-strategy mistake and has resulted in mis-calibrated consumer expectations around how much costs are actually increasing. Just make the products you can make and sell them at the same margins you’ve always operated at.
Super refresh should recalibrate a couple of the worst misses, 4080 MSRP is indefensible and 4070 ti is just way too expensive for a 12gb card, but, 4060 and 4070 and 4090 are all pretty on-target and are not particularly high-margin products especially compared to dirt cheap Turing and ampere chips.
AMD typically loses out in certain features. Video encoding/decoding is worse on AMD, if I'm not mistaken. They've also had a reputation for bad drivers, at least in the past. Not sure how true that is now.
AMD has real good hardware to compete at least in the gaming sector. But, the hardware is only as good as the software/driver supporting it. In the ML and Cryptor sector, Nvidia has built a moat around their hardware called CUDA, a pretty much standard if you want your GPU specific code to be “portable”. I’m afraid OpenCL won’t matter anymore unless Nvidia supports it. Anyone with more industrial domain knowledge wants to chime in?
I think Nvidia has built a firmware and library support moat around their hardware that is difficult for others to keep up with. They invest time to work with studios to make sure they have game ready drivers available the same day as new AAA games are launched.
Nvidia has been the pick axe vendor for 2 gold rushes in a row now. First crypto, then AI. It's an extremely nice position to be in. Regardless of how something like OpenAI shakes out, Nvidia will keep making money.
Seriously. I can't think of any other company that has managed to "luck out" this much.
GPU's were designed for graphics. But then they just happened to be the right tool for crypto, and then the right foundation for ML.
I can't think of any other company that has managed to benefit so much from a product designed for one thing, that turned out to be the solution to other things as well.
(Edit: obviously they've done a ton of work to capitalize on these two trends and plan for them as they saw them coming, but that was always leveraging their pre-existing massive investment in GPU design and expertise.)
Luck is the combination of preparation and opportunity. Cuda has played a huge role in Nvidia's ML success. They were working on this for years before the AI wave took off.
As much as I hate praising big corp, NVDA played the game right. They saw the value of GPGPUs and invested heavily in CUDA, sponsored universities and research labs when the DL boom happened, helped all ML frameworks into incorporating CUDA (dl or otherwise), has done a lot of work on cudNN, sparsity, and has published countless papers.
It didn't simply "turn out", and claiming such is doing a disservice to all the people who made that reality.
Intel could have very well built accelerators and GPUs, but they didn't. AMD also didn't even though they had a GPU division as well.
Hardly , this is not about nvidia doing the foundation to make their own luck .
It is about those industries taking off in a big way.
Nvidia has zero contribution towards tranformers which was the driver behind the rise of AI or towards all the money that flowed in crypto during the low interest era.
The point has all to do with how they have been part of two cycles of boom unrelated to their product.
They are not AWS to have driven the cycle by their product itself , they have only capitalized on it with their good work .
I recall that all the way back during the early days of their GPUs they were already talking about their compute capabilities, so I rather than luck, it's really just a lot of foresight, with crypto and AI emerging because of GPUs becoming good enough for the task, rather than happening to fit the problem.
I don't think it's exactly "luck", the trends are being driven by the technology, which they have done a good job of inventing. For better or worse, low-watt parallel compute is the primary area of innovation in compute hardware, so naturally the areas of innovation in software are more likely to leverage this innovative hardware. In other words, when all you have is a ton of tiny processors, every problem looks like a massively parallel process. It does raise the question of what we are missing because we don't have e.g. super-fast single-threaded processors or whatever new tech exists in an alternate universe.
> I can't think of any other company that has managed to benefit so much from a product designed for one thing, that turned out to be the solution to other things as well.
> In 2003, a team of researchers led by Ian Buck unveiled Brook, the first widely adopted programming model to extend C with data-parallel constructs. Ian Buck later joined NVIDIA and led the launch of CUDA in 2006, the world's first solution for general-computing on GPUs.
In the early 2000s it was a common theme in the CS literature that the Von Newman architecture was about to run out of steam and some kind of parallel processor was going to become mainstream, out of all that work the GPGPU became the most famous. If NVIDIA hadn’t done it, someone else would have.
The thing is, Nvidia has "done it" and still no one else is "doing it". As the gp points out, Nvidia launched CUDA a while and this made Nvidia's products general purpose parallel processors.
They have succeeded as that and yet still no one is selling chips that either have an equivalent interface or (better) are drop-in Cuda compatible (since we know interfaces and programming language copyrightable).
The situation seems a bit like early PC days when companies sold (roughly) PC compatible computers were appropriately shoved aside by drop-in compatible clones/white-boxes. A company that makes a thing "kinda like" the leader but with it's own sauce is just trying to carve a piece of an existing market. The company that makes something exactly compatible with the leader has an incentive to go further. I hate to say "embrace and extend" but there you have it.
My own personal history has to do with it, because when I was doing my PhD work they defunded our supercomputing center. Parallel computing went from being “the future” to a boring backwater dominated by the military industrial complex and a handful of ‘boondoggle’ projects like figuring out how proteins fold in water which turned out to be a waste of time because we found out proteins don’t “fold” but instead they get folded. I diagonalized a 65 spin cluster on an IBM AIX machine if I used a supercomputer I could have diagonalized a spin 90 maybe and it wouldn’t have made any difference in the results.
Turned out having a powerful computer on your desktop was revolutionary and having to fill out paperwork to run your job was just… a tired old throwback.
I got a lot more excited about commercial “big data” developments in the 2000s as these completely escaped the gravity of the national labs and the handful of problems they do over and over again.
A different lens is that GPUs are designed for parallel computing, of which graphics is the first primary use case. Crypto and AI/ML are perfect suits for parallel computing, and not the last ones. And nvidia is in the perfect spot there for the future of parallel computing as a whole.
If OpenAI doesn’t achieve AGI, which it won’t, demand probably won’t continue to grow. LLM’s will find their limit and massive training budgets won’t be needed.
Since I'm being rate limited for one downvote:
> All science is brute force of guess and test, and OpenAI does a lot of that.
Nope. Science looks to answer a critical question. It doesn't conduct trillions of experiments at one time.
> You don't think there's any possibility of emergent properties? Given that we understand physically how the human brain works, but not how consciousness emerges from that, I'm pretty damn unconfident.
Possibility? Why? Because Altman & co are hyping things up? They have no proof of a breakthrough. It's just posturing.
If you accuse me of hubris, are you willing to take a $10,000 bet? It's clear that there is no science happening at OpenAI. It's pure brute force. You won't learn anything with that approach.
You don't think there's any possibility of emergent properties? Given that we understand physically how the human brain works, but not how consciousness emerges from that, I'm pretty damn unconfident.
That article says that gaming cash has been eclipsed by A.I. in the short term but you could say the gaming market is more of a sure thing in the long term.
It could let them push for higher gfx card pricing since the cards would have a higher duty cycle and deliver more value in a cloud environment but unfortunately I am in the wrong side of a DSL connection.
I am disappointed that nobody in cloud gaming has made a game that using the monsterous GPU instances in the cloud to synthesize a super-complex world and generate video for multiple players simultaneously but I think a system like that would not really beat highly optimized single players and furthermore optimizing it to the point where they could really deliver a “skip three generations” kind of experience would be expensive and risky.
I don't think ML is driving consumer GPU prices, although I could be wrong. They are selling 3080s and below primarily for gaming. Maybe an argument could be made for the 3090/4090, but even then that is extreme solo enthusiast territory.
It's clearly the higher end chips that are bringing in the AI revenue.
That’s a great business model, selling pickaxes and dynamite on credit though can backfire if your are the one loaning out the money. Nvidia is a bit at risk it this gold rush ends suddenly
While the current over marketing of AI is a fad, this person is making a dumb claim. AI is genuinely revolutionary and the next step in modern computing.
It is very unfortunate for people like you understood that crypto was a scam and then assumed that the next thing that all the same people pivoted to was also a scam.
There's also a lot of substance, though. AI tools are useful for real, shipping products today. Sure there's a lot of hype and people overselling their capabilities, but under the hood there's honest-to-god useful shit.
While I think the 'AI' trends are waaaaaaaay overinflated right now and somewhat grifty, even if it completely collapsed Nvidia would be fine. They provide lots and lots of compute power to those who need it, and in the commercial and public sectors that's always needed - far more as technology advances.
Someone somewhere will always need to run a bigger, better simulation and Nvidia is more than ready to provide the hardware for that.
I think you can pretty much bank on the fact that large scale vector and matrix calculations aren't going away, regardless of which hype-wave they're a part of.
The threat to NVIDIA is probably from startups also working in the space of machine learning acceleration, not from industry demand itself.
If anything, ML will just move more and more from "tech"[bro] type companies and into more and more broader industry. Aerospace & defense, manufacturing automation, etc. etc.
It's not about chatbots and AGI, it's about dealing with huge quantities of data, and that's not going away.
Their PE ratio was over 100 before the earnings announcement last week, so it has fallen. 18.12B revenue, 9.24B net this last quarter - 6.05B revenue, 1.41B net three quarters ago.
With growth in revenue and net this quick, the 12 month trailing P/E ratio will be high.
I agree people may have missed the boat but for that PE ratio i wouldn't focus too much on it as PE is calculated using 1 year of past data. It fails spectacularly in providing a valuation when there's extreme YoY growth which there is here.
See the PE chart of Nvidia here and notice the large step changes everytime a quarterly report comes out: https://ycharts.com/companies/NVDA/pe_ratio . As each quarter where they had <1/10th the earnings is pushed out of the PE calculation they have a big step change.
Their forward PE is 25.32x right now. They have an expected 8% quarterly growth forecast. The FAANG have PE's higher than that and much lower growth. Really on fundamentals of PEG (PE and growth) Nvidia is pretty good value. But... There's huge risk here. Will they continue or will a competitor emerge? Can they rely on chips from Taiwan? etc.
So the investor story isn't that they have a crazy PE. They honestly don't. The investor story is that fundamentals for Nvidia look quite good but there's also risks looming.
I guess PE is better for steady-state businesses like utilities. But even with PEG, $NVDA is quite high - I get 16, but I much prefer this to be below 1.
None of this means they're not a good buy, just that it would've been really fantastic to already be an owner.
I couldn't find a consistent definition for it so I used the past five years. That's another factor that would weigh against fast-growing tech companies.
Maybe the best portfolio would be one part proven, stable value investments picked using measures like PE and PEG, and one part tech companies that are chosen for future profit potential, which you can't really measure or forecast so much as make an educated guess.
119 comments
[ 3.1 ms ] story [ 188 ms ] threadNvidia, on the other hand, invested tons and tons of money to make CUDA good over the last like 15 years, while AMD has not really put any serious effort into AI/ML workloads. It's mostly a software support problem, not a hardware problem to my understanding.
It's been 11 years since Linus gave Nvidia the finger. Nobody who is thinking of possibly running Linux should ever buy an Nvidia product. Yet Nvidia is still very commonly spec'd into Linux builds even today.
https://www.gamingonlinux.com/index.php?module=statistics&vi...
In the Windows land it's indeed mostly Nvidia.
That’s not to mention having to fiddle with the drivers/games more than I was accustomed to, which IIUC is more a consequence of having less market share and thus not being a priority for game developers to work out of the box. I can excuse that, but not crashes or reinstalling every couple months.
Crypto really only managed to use it because there was more money to be made using AMD, justifying the wrangling needed to get compute to work on it.
Now they continue to waste their potential with ROCm's design choices which make its hardware support range very limited.
AMD is already backporting ROCm to "lesser" hardware [0]. I wouldn't expect them to spend too many resources on going back further than that though. I'd rather see them improve their newer stuff. The larger issue is that it is impossible to rent time on the super high end GPUs that they use to build the super computers with... nobody offers that and it is something I'm working on fixing.
[0] https://www.tomshardware.com/news/amd-enables-rocm-and-pytor...
They’re stuck on the concept of rasterized graphics being the ground truth in gpu value.
Then they’re shocked when the wider consumer market doesn’t hold their perspective.
AMD hasn’t had the big picture vision that nvidia has had. For example, with Vega AMD had a gpu architecture with exceptional async compute performance a generation before nvidia was competitive.
AMD then abandoned Vega, switched to a leaner and less compute friendly rDNA architecture while nvidia went in the opposite direction, pushed compute heavy features like raytracing and dlss and AMD hasn’t had an exceptional response since then.
Creating reflections using rasterization requires shortcuts and sacrifices. And once you see them, you can't unsee them, and then you find them visually distracting.
There are basically 3 ways to create reflections without ray-tracing:
1. Double the geometry and render it in the reflection. This is rarely the approach taken because it's incredibly resource intensive. Oddly, Duke Nukem 3D back in what, 1996? did this, but that had pretty simple graphics. Genuinely not feasible for any sort of reflect surface that is curved.
2. Screen-space reflection: Just take the rendered image, invert it, and overlay it onto the reflective surface. This is probably the most common way to do it. It's fast and looks decent. The problem is that it requires the reflected image to be visible on the screen. When it's not, there's no reflection. It creates a bizarre effect when, for example, you're looking over a body of water at something in the distance, and you see the reflection in the water, but as you move the camera down, the reflection gets cut off and eventually disappears entirely.
3. Using a static image as an environment mapped texture. Apex Legends does this for shiny surfaces. It creates a great illusion of a reflection, but taking even a second to actually look at it, you'll notice that the reflection doesn't actually show what it should unless you're standing in the exact right place.
I will concede that gamers typically don't notice the issues created using methods 2 and 3, especially in a game like Apex Legends where you're moving around too much to notice.
But me? I dunno. I notice and get distracted.
EDIT: DLSS, on the other hand, is basically a crutch. It's not surprise that DLSS became a big thing at the same time RTX did. RTX requires more juice to run at full resolution than the GPU has, so they introduced DLSS to render at a low resolution and upscale it.
$250-270 is fine for this type of product even today.
Similarly, the 4070 is also fairly attractively priced even compared to 6800XT/6950XT or 7800XT. More efficient, better feature set, same performance tier, similar perf/$. But people still get upset about intangibles like memory bus or die size (despite this already being measured in the performance!), or just don’t want to admit that wafer costs and R&D costs are rising in the post-Moores law era. Operating margins were actually down relative to the last few gens (until AI hit), and were broadly similar to 2012 levels. This is just what it costs now, and smaller die take the worst of the cost increases because memory controllers and pcie don’t shrink anywhere near as much as logic (the gpu cores). There is a “minimum cost floor” that has been rising substantially due to this.
I think Apple proves that it is possible, and possible relatively quickly.
And Rocm supports the higher RDNA2 and the highest-end RDNA3 consumer GPUS - [0], though only the top RNDA3 cards in linux [1] (officially - other cards often do "work" but YMMV)
One thing Nvidia are much better at is supporting the whole stack - while I doubt many people are actually running much ML on 8gb mid-low end cards, it seems a hook for sales if people "could" do that, and a starting point for personal experimentation. I know multiple people who buy low end Nvidia GPUs because they "might" experiment with CUDA some time in the future.
[0] https://rocm.docs.amd.com/en/latest/release/windows_support.... [1] https://rocm.docs.amd.com/en/latest/release/gpu_os_support.h...
but this is very important. Many develop locally before sending it to a cluster, especially if you only have limited credits available.
For as long as I can remember, Jensen has been telling us during all hands meetings that Nvidia is a visual computing solutions company, not a semiconductor company. (The “visual” part has only been dropped in the past few years.)
Nvidia sells hardware, but they also sell complete solutions, and everything in between.
I do care that Geforce experience works flawlessly. I mean, it doesn't, and my grandkids will probably still use DDU, but when shadowplay works it's indispensable as a gamer (or even for recording video calls.) I'm sure AMD has an alternative or I could use OBS, but I'd rather just pay a premium and rarely have to think about it. And also CUDA.
Like are companies like this a credible threat? https://www.cerebras.net/
I hear a big part of Nvidia's lock-in is that they provide software that runs these AI projects?
Nvidia’s problem is that every company is a credible threat because in the ongoing gold rush they can’t make enough shovels to meet demand and damn there is gold in them thar hills. People are going to do literally whatever it takes to make their competition’s products work. If you are selling a shovel, even a shitty shovel, you are a competitive threat to Nvidia right now.
[1] https://www.tomshardware.com/news/intel-arc-graphics-cards-a...
But Nvidia and AMD have a huge time advantage on Intel. It's going to be a few generations and billions of dollars before Intel can catch up, assuming they keep with it.
Intel has been in the integrated GPU market for awhile but with little pressure to innovate and very different design goals.
[1] https://www.computer.org/publications/tech-news/chasing-pixe...
Nvidia has been making dedicated GPUs that entire time, while Intel basically only started. Intel has a LOT of catching up to do.
Intel and AMD just keep burning developers over and over.
But making a GPU is hard. Making a good GPU is even harder. This becomes a chicken and egg problem. If NVidia's hardware has quirks, and you're a game developer, you optimize your game for Nvidia's hardware. If someone else tries and builds a GPU, they discover that many, popular, games run like sh*t. Because of tens of years of strange workarounds to ship a game, drivers rewrite commands for specific games, game devs abuse the directx spec or do it wrong.
However, Nvidia set out 10-15 years ago to corner the professional general purpose GPU compute market with a software layer called CUDA. Software was written and optmised for CUDA, and not any other generic graphics library for reasons I don't know (I'm not a graphics developer, just a gamer). So now Nvidia enjoys a moat in gpGPU (as it was called).
If you mean gaming? AMD has had better mid range gpus since the rx 480 was released in '16. Nvidia has been resting on their laurels and AMD has recently seemed to be fine playing the 'price is right' game where they price their gpus ~ 10% less than nvidia and don't even attempt to take market share.
It's in the article: "...of the company's $18.12 billion in revenue, $14.51 billion was generated by its data center division."
They spent over a decade building up CUDA into the industry standard for GPU compute programming. That's why it's so hard for anyone to compete in this domain. AMD is doing better in gaming, but that's practically a sideshow business for Nvidia now.
I know, but my point is there were heavy incentives to compete with Nvidia. But nobody did, at least not with the same intent. And now they're (Nvidia) raking it in due to the AI/ML wave.
If others had comparable GPU capacity and know how, the market wouldn't be so skewed in favour of Nvidia.
Nvidia's success is rooted in design and business choices made a long time ago. They were bets at that time, as ambitious hardware initiatives need to be, and other vendors like AMD made different bets. Nvidia won their bet more spectacularly than others, and now gets to benefit from their lead as others slowly work to reorient their ships in its direction.
The simple answer is, like many markets with few competitors, they like prices high. It's like a point of no return if they undercut their crappy workstation graphics.
The products that are red hot right now also use HBM not VRAM.
And the ML community isn't very motivated to port their code from nvidia to amd because the nvidia tax isn't too bad.
If AMD could take one of their existing 24GB consumer cards and quadruple the RAM, though? Open source ML efforts would have support within days.
And nvidia couldn't respond to undercut them without also undercutting their cash cow server GPUs.
Intel is making some better moves lately but the impression I get is they're losing money on every sale and need to buy their wafers from TSMC like AMD. I'm not sure how long they'll stomach that situation or if they'll eventually get their GPUs onto their own process.
They ironically could do exactly this because the MCD disambiguates the memory config from the gpu config. They would need a new MCD die but they could put four or 8 memory controllers on a MCD instead of 2.
Nvidia isn’t going to sell gaming cards at an operating loss out of AI money. If the consumer market is soft then it’s soft, and with the general slowdown in Moores law cost progression it doesn’t make sense to have “firesale gens” that set an impossible standard for future gens to beat.
~20% perf/$ progression per generation is going to be about what’s sustainable, because that’s about what tsmc is giving the chip companies. You can “cheat” by using older worse nodes like ampere and turing did, and like AMD did with the RX 7600, to keep costs down, but that’s a one-time boost and consumers grow to expect that pricing going forward. It’s been a market-strategy mistake and has resulted in mis-calibrated consumer expectations around how much costs are actually increasing. Just make the products you can make and sell them at the same margins you’ve always operated at.
Super refresh should recalibrate a couple of the worst misses, 4080 MSRP is indefensible and 4070 ti is just way too expensive for a 12gb card, but, 4060 and 4070 and 4090 are all pretty on-target and are not particularly high-margin products especially compared to dirt cheap Turing and ampere chips.
GPU's were designed for graphics. But then they just happened to be the right tool for crypto, and then the right foundation for ML.
I can't think of any other company that has managed to benefit so much from a product designed for one thing, that turned out to be the solution to other things as well.
(Edit: obviously they've done a ton of work to capitalize on these two trends and plan for them as they saw them coming, but that was always leveraging their pre-existing massive investment in GPU design and expertise.)
As much as I hate praising big corp, NVDA played the game right. They saw the value of GPGPUs and invested heavily in CUDA, sponsored universities and research labs when the DL boom happened, helped all ML frameworks into incorporating CUDA (dl or otherwise), has done a lot of work on cudNN, sparsity, and has published countless papers.
It didn't simply "turn out", and claiming such is doing a disservice to all the people who made that reality.
Intel could have very well built accelerators and GPUs, but they didn't. AMD also didn't even though they had a GPU division as well.
It is about those industries taking off in a big way.
Nvidia has zero contribution towards tranformers which was the driver behind the rise of AI or towards all the money that flowed in crypto during the low interest era.
The point has all to do with how they have been part of two cycles of boom unrelated to their product.
They are not AWS to have driven the cycle by their product itself , they have only capitalized on it with their good work .
CUDA was first released in 2007:
* https://en.wikipedia.org/wiki/CUDA
* https://developer.download.nvidia.com/compute/cuda/1.0/NVIDI...
Two years before the Bitcoin paper (2009):
* https://en.wikipedia.org/wiki/Bitcoin
They had a presentation called "The Era of the Personal Supercomputing" at SIGGRAPH 2007:
* https://dl.acm.org/doi/10.1145/1281500.1281647
* https://www.nvidia.com/content/events/siggraph_2007/supercom...
Ian Buck (co-?)creator of CUDA speaking in 2008:
> Ian Buck talks about his background developing Brook for GPUs at Stanford university and what paths were taken for developing a C platform for GPUs.
* https://www.youtube.com/watch?v=Cmh1EHXjJsk
> In 2003, a team of researchers led by Ian Buck unveiled Brook, the first widely adopted programming model to extend C with data-parallel constructs. Ian Buck later joined NVIDIA and led the launch of CUDA in 2006, the world's first solution for general-computing on GPUs.
* https://developer.nvidia.com/cuda-zone
* http://graphics.stanford.edu/~ianbuck/
Nvidia purposefully went after parallel computing. Specific applications (cryptocurrency, ML/AI) appeared later.
The thing is, Nvidia has "done it" and still no one else is "doing it". As the gp points out, Nvidia launched CUDA a while and this made Nvidia's products general purpose parallel processors.
They have succeeded as that and yet still no one is selling chips that either have an equivalent interface or (better) are drop-in Cuda compatible (since we know interfaces and programming language copyrightable).
The situation seems a bit like early PC days when companies sold (roughly) PC compatible computers were appropriately shoved aside by drop-in compatible clones/white-boxes. A company that makes a thing "kinda like" the leader but with it's own sauce is just trying to carve a piece of an existing market. The company that makes something exactly compatible with the leader has an incentive to go further. I hate to say "embrace and extend" but there you have it.
Ethereum made a concerted effort to sabotage ASIC mining, and remain GPU minable. Such is the case for all other GPU minable coins.
So no, I don't think crypto needed GPUs.
Related to that, I feel like AI should start moving to FGPAs/ASICs soon.
Turned out having a powerful computer on your desktop was revolutionary and having to fill out paperwork to run your job was just… a tired old throwback.
I got a lot more excited about commercial “big data” developments in the 2000s as these completely escaped the gravity of the national labs and the handful of problems they do over and over again.
Since I'm being rate limited for one downvote:
> All science is brute force of guess and test, and OpenAI does a lot of that.
Nope. Science looks to answer a critical question. It doesn't conduct trillions of experiments at one time.
> You don't think there's any possibility of emergent properties? Given that we understand physically how the human brain works, but not how consciousness emerges from that, I'm pretty damn unconfident.
Possibility? Why? Because Altman & co are hyping things up? They have no proof of a breakthrough. It's just posturing.
All science is brute force of guess and test, and OpenAI does a lot of that.
AI is a little different for now, we'll see how it holds up.
I am disappointed that nobody in cloud gaming has made a game that using the monsterous GPU instances in the cloud to synthesize a super-complex world and generate video for multiple players simultaneously but I think a system like that would not really beat highly optimized single players and furthermore optimizing it to the point where they could really deliver a “skip three generations” kind of experience would be expensive and risky.
It's clearly the higher end chips that are bringing in the AI revenue.
Give it a few years, and it'll be pared down to a couple of real use cases with some very strong limitations.
Unlike crypto, which was all hype, no substance.
Maybe Apple’s iPhone grift will end soon too!!
Someone somewhere will always need to run a bigger, better simulation and Nvidia is more than ready to provide the hardware for that.
The threat to NVIDIA is probably from startups also working in the space of machine learning acceleration, not from industry demand itself.
If anything, ML will just move more and more from "tech"[bro] type companies and into more and more broader industry. Aerospace & defense, manufacturing automation, etc. etc.
It's not about chatbots and AGI, it's about dealing with huge quantities of data, and that's not going away.
Their revenue is up 206% but they have massive margins on the datacenter side meaning their earnings are up 1276% YoY
Can we fix the title to either say revenue up 206% or earnings up 1276% to be correct?
See https://nvidianews.nvidia.com/news/nvidia-announces-financia... or even just look at the linked article. They do not say earnigns up 206%. They say revenue up 206%
https://valustox.com/NVDA
A financial performance for the ages, and they look set to keep it up.
Unfortunately for new investors, their market cap is even more insane, giving them a crazy PE ratio of 73.
Their PE ratio was over 100 before the earnings announcement last week, so it has fallen. 18.12B revenue, 9.24B net this last quarter - 6.05B revenue, 1.41B net three quarters ago.
With growth in revenue and net this quick, the 12 month trailing P/E ratio will be high.
See the PE chart of Nvidia here and notice the large step changes everytime a quarterly report comes out: https://ycharts.com/companies/NVDA/pe_ratio . As each quarter where they had <1/10th the earnings is pushed out of the PE calculation they have a big step change.
Their forward PE is 25.32x right now. They have an expected 8% quarterly growth forecast. The FAANG have PE's higher than that and much lower growth. Really on fundamentals of PEG (PE and growth) Nvidia is pretty good value. But... There's huge risk here. Will they continue or will a competitor emerge? Can they rely on chips from Taiwan? etc.
So the investor story isn't that they have a crazy PE. They honestly don't. The investor story is that fundamentals for Nvidia look quite good but there's also risks looming.
I guess PE is better for steady-state businesses like utilities. But even with PEG, $NVDA is quite high - I get 16, but I much prefer this to be below 1.
None of this means they're not a good buy, just that it would've been really fantastic to already be an owner.
You should get a PEG <1. (25/36)
Maybe the best portfolio would be one part proven, stable value investments picked using measures like PE and PEG, and one part tech companies that are chosen for future profit potential, which you can't really measure or forecast so much as make an educated guess.