I'm not sure who this is really for. I mean that honestly.
At least on the workstation side - i do a lot of solidworks-based 3d modeling. The A6000 can easily, for example, raytrace in real time anything i can even find to throw at it.
Like models that Solidworks still has trouble opening and rendering normally can raytrace instantly.
What part of workstation software is still GPU bound at this point on high end GPU's?
I guess if CAM was taking more advantage of GPU's, i could see it useful there, but on the modeling side, i honestly don't get it.
Access to more memory opens up higher resolutions in ML video and image workflows. I am pushing close to these numbers for GPU memory on the 64G M1 Max.
All of the responses are about game development, and that sort of thing which I totally understand. but all of their quotes are from like cad developers.
so I definitely feel like they are mistargeting at this point
I totally buy the game developer need here, as well as some of the live video processing folks. but that convinces me even more that these are totally mistargeted given their sales points
Photorealistic renders for catalogs and architectural visualization can still take many hours on the best GPUs available (if the scene even fits into VRAM). Even a slight performance improvement can be worth a lot of money for companies doing this kind of work.
Color correction uses as much GPU memory as it can get. I used to maintain DaVinci Resolve systems with expansion PCIe chassis to handle all of the exapnsion cards. Each year, we'd upgrade the GPUs with more cores and memory. Each upgrade allowed the system to be much faster. However, this used to be on Macs, so an Nvidia card is no longer useful on that platform. Might still be a cool thing to play with under WinX/Linux
First off, throwing a LUT in there is not color grading!!!!!!!! People making this claim are appreciated as it lets me know immediately that they are not familiar with the subject at all. It's like a shibboleth. Edit: I mean this in the nicest way. It's just a pet peeve of mine. You already stipulated you didn't fully grok the topic.
In a full color grading session, you can have power windows. You can have those power windows tracking. You can have noise reduction. You can have other filters. You can do so so much. All of that takes up memory. You can be dealing with 8k footage. You can have many many many layers of that footage. You can be grading in >16bit color space.
You can do all of that with very little VRAM at a snails pace, or you can do it in real-time with all of that VRAM and keep the client in a supervised session happy. With all of that VRAM, you can export at faster than real-time for delivering the content to the client at the end of the session rather than the next day after an overnight render.
The A6000s have become very popular for ML engineering, especially in the NLP space since those models are enormous. 48GB VRAM and 2x faster training time is tempting for those applications!
Although an small audience, this is exactly the kind of card that live video processing (IMAG) needs. Software such as Hippotizer, Pandora's Box, Resolume, and vMix rely heavily on GPU.
I feel like this is the card for the desktop deep learning scientist. The A100 is much more expensive, and are better for the datacenter and don't really suit a home rig. The recent 4090's looked nice but don't support NVLink. For me, it's all about the RAM. 48GB is a dream.
Me as well, but with 48GB of RAM what is the point? They should sell at least 96GB one. There is no pressure/enticement to upgrade from A6000 to be honest.
Agreed (especially since I just got a third A6000 this week), but by the time they come out the faster training/inference time might be enticing to some. I would love to see an 80GB workstation card though...
Yeah obviously better to go with 4090's if your model fits in the VRAM. But if the models you want to work with are larger than 24GB you'll need to spend the big money.
Display Support: Maximum Digital Resolution (1) 7680x4320
Standard Display Connectors HDMI(2), 3x DisplayPort(3)
Multi Monitor 4
HDCP 2.3
Card Dimensions: Length 304 mm
Width 137 mm
Slots 3-Slot (61mm)
Thermal and Power Specs: Maximum GPU Temperature (in C) 90
Graphics Card Power (W) 450 W
Minimum System Power (W) (4) 850 W
Supplementary Power Connectors 3x PCIe 8-pin cables (adapter in box) OR
450 W or greater PCIe Gen 5 cable
I see it now, thank you for pointing at it. That is not a great decision by Nvidia, IMO.
Personally, I would buy a couple of 3090's with NVLink and call it day if it was for personal use. I would wait and see actual performance before even thinking about Lovelace cards. Nvidia makes all kinds of unsubstantiated claims during these events.
NVLink is important if (1) if your model doesn't fit in the RAM on a single card or (2) you want to increase batch size past the point of each batch fitting in the RAM of a single card or (3) to increase training speed by spreading the workload.
I don't think this supports NVLink anymore. The product page for the RTX A6000 mentions NVLink but it is gone from the product page of the new RTX 6000.
$4800 is what the previous RTX A6000 costs in the US after taxes. That is a BIG cost for anyone outside of a professional setting.
You could buy 2 x 3090's with NVLINK for half that. I recently bought a datacenter M40 with 24GB of RAM for $150 on Ebay then added water cooling. The Dell Precision 5810 I bought second hand can now run some really interesting stuff on those 24GB of RAM albeit slower than a 3090. For less than $500 I can have two watercooled M40's that will run almost anything I can throw at them.
My understanding may be flawed, but I am under the impression that 2 x GPUs with NVLink will pool memory. It does not change the underlying 2 x GPU architecture in a way that would represent 2x as a single when looking at the system config. Furthermore, I believe the application code needs to have NVLink enabled to take advantage of the pooled memory. Pytorch and Tensorflow have knobs to turn in order to take advantage of NVLink... where programs like DaVinci Resolve may need patching, etc.
That's different, NCCL manages traffic between GPUs, but each GPU (with its memory) is still treated as a separate entity by Pytorch or TF. Basically if your model does not fit in a single GPU memory, Pytorch will not automatically distribute it across two GPUs even if the GPUs are connected with NVLink. You can use model/tensor parallel methods to do this, but it's not going to be automatic (will require writing extra code).
I've been thinking of doing the same with the M40. I'm more interested in learning and time isn't really an issue, mainly just running big models. Is water cooling that important? Is it great for a prosumer looking to run models and research the field with?
Nvidia smells a lot of profit and maximise it by forcing prosumers to buy trashy value "quadro" or even worse buying data center GPU's for working on state of the art problems (6-20k usd minimum investment to do normal sized Deep Learning in next 2/3 years incoming and probably 3/4x that for training).
Big fuck you for nvidia with this release prosumer market is dying right now and there is no alternative cause AMD software is so bad you are forced to use CUDA in for example DL/DS.
I don't understand why the ML community went so hard in on CUDA. Maybe using nvidia cards made sense back around pascal, but I don't get why they would choose to lock themselves into nvidia's proprietary APIs.
Alright, in that case I suppose the trade-off was clear: use a nice API and lock yourself in, or use a less nice API and be able to use any hardware. Nvidia cards becoming less competitive over time is in any case a very predictable outcome. Sounds like the ML community made a trade-off, and this is the downside of that trade-off.
@mort96 I've reached the thread depth and can't reply directly - but it's not just a case of a nice API. I don't even think it was possible to use alternative hardware and SDKs for consumer GPGPUs for deep learning when the field was taking off. It was not so much a choice but the only viable option.
Has Nvidia taken advantage of their position? Absolutely. But don't fault the ML community when there was nothing else available.
HN limits how fast you can reply. I think it's to reduce flame wars and force people to take some time to cool off if things get heated :) Just wait a few minutes and it lets you respond.
OpenCL has existed for about as long as CUDA, and can be used on GPUs from any of the major manufacturers. What makes OpenCL so unsuitable for ML that the ML community just had to use CUDA?
CUDA was just plain better back in the day, more feature robust, and more features in general. Much simpler to pick up as well, especially at the time with what tools and libraries we were forced to use
NVIDIA also partnered up very quickly with many big players in the game, the sales people went to work, but they had the technological feats to back it up.
AMD's OpenCL implementation and tooling were really, really rough. Lots of hard lockups, memory leaks, and forum threads ending in the person with the question giving up and switching to team green. I followed this path eventually, too. The breaking point happened when I had spent an entire day trying to get a bit of OpenCL working, thinking that I was at fault, but then I tried running it on a NVidia box, hoping to get a more descriptive error, but the code Just Worked. NVidia's OpenCL implementation was better than AMD's, and not by a small amount. CUDA was better still. I realized I had thrown away $n0000 of my own time chasing a $n00 discount on the AMD card (controlling for perf). Never again.
Now that AMD has money, hopefully they have fixed their stack, but I'm still in "once bitten, twice shy" mode. I want to see someone else in my field using AMD on tasks I care about before I try team red again.
Same experience here, as recently as with AMD's 5000 series GPUs. As far as I'm aware their OpenCL stack is still terrible, they're focusing mainly on HIP and ROCm, but it too has terrible consumer side support and an update model that makes it essentially useless as a long term investment. They include device code for every supported GPU, so they have to drop support for older cards to keep the binaries reasonably sized.
In contrast, CUDA compiles down to a device agnostic intermediate language so they can support the same code on several generations of hardware, limited only by the CUDA feature level available on the hardware.
AMD still has a long way to go for now. Both in terms of 'primary' functionality like supported hardware and compilers and 'secondary' functionality like detailed profiling and debugging tools.
It's amazing to me that AMD didn't put more resources behind better GPGPU and DL software tooling like NVIDIA did. This was an excusable error in 2015, but not in 2018. This probably cost AMD tens of billions of dollars in lost market cap.
There are other options, including TPUs. Besides, cloud computing is what any sensible practitioner uses, despite how building a home rig is a fun hobby.
I agree that most teams just use cloud compute. But if you're a startup that wants to not burn crazy amounts of money, then a home/office rig is definitely the way to go. A month or two worth of training in the cloud will cost the same as a V100.
Yea, large companies that don't really care about operational spending will happily use GPUs. But for anyone else, spending a few thousand on GPUs will save a lot of money very quickly.
I think that they've given up on competing directly for consumer level compute. It makes up a small (albeit important, since it's how you get developers in the first place) part of the market and their main userbase seems more than willing to overlook the lack of productivity features because the (predominantly gamer) users misguidedly think those are only for workstation/server grade hardware.
They may be hoping to live off gamers and big server contracts (eg Frontier) until they get their software stack in a decent position and outgrow their current reputation of having buggy software.
There are literally nothing reaching the knees of all you can do with CUDA. Peoples forgot how Nvidia can be good by creating and expanding it's own market just by providing quality softwares/API.
On a side note, AMD did a different (yet interesting) move by embracing the FOSS culture (working with uptreams).
Nvidia basically created the whole AI-on-GPU market, so it's not like there was any other option for like 10 years... OpenCL came as a hindered afterthought that is still strictly inferior to CUDA.
This card is expected to be around $6,000. Maybe it's just me, but I don't consider it that unreasonable. A lot of ML scientists are (were?) earning around 7-figures. Companies paying these salaries can afford a a $6,000 GPU. Of course, it'd be nice if there was a little price discrimination for the rest of us. Still, $6,000 doesn't seem too bad for cutting edge tech.
Sorry here in EU we are making a lot less past 2 years we have big increase in salary but still avg for EU is probably about ~60-80k USD if you have 4/5+ years of exp and if you are very good you can get 100-120k USD+ (not counting Meta/Google and few other but only in few countries) and ofc it is pre-tax as taxation is a lot higher than US so in some countries you'll get about 40-60% total salary (counting VAT on basic products and other taxation).
Not even counting Eastern Europe and Asia where you get fraction of that.
Then again, I've been places that are stingy about printing, which didn't make much sense to me, so I guess they have some... other way of looking at this stuff.
Imagine you have to spend 15% of your yearly salary every 2 years cause nvidia tax (ofc in western developer countries and more like 50% excluding cost of living in less fortunate areas)
Not everyone is living the dream in the US of A, y'know?
For those privileged enough, this might be reasonable. For the rest of the world (bar some exceptions), not so much.
Could they have picked a more confusing name? They already had a Turing based Nvidia RTX 6000 from 2018 [1]. Then the RTX A6000 came out based on Ampere. Now this new one has to be called the the "NVIDIA RTX 6000 Ada Generation". This is going to be such a headache for anyone who has to requisition things through a purchasing department at work: "No I don't want the RTX 6000 or the RTX A6000. I want the RTX 6000 Ada."
They're also going to have a whole "RTX 60xx" series in like 4-5 years, aren't they? Seems like a really dumb choice to use the RTX branding with the same number format for a totally separate product line, but maybe they're banking on that confusion to have people accidentally purchasing an 8 year old RTX 6000 when what they really wanted was an RTX 6050.
Or since Nvidia is historically stupid with naming systems and can't stick with them, maybe the RTX 4-digit numbering goes away before 6000 series.
They had a chance to keep a numbering system going when they did the GTX 1xx, GTX 2xx, GTX 3xx.... GTX 9xx, GTX 10xx, and could have gone to GTX 11xx and been set for as long as they wanted (well over a hundred years) counting up toward GTX 99xx), but instead they went to GTX 20xx and RTX 30xx so now they're going to run out of numbers again.
Bets on what's next? Do we hit RTX 10,000 and then jump back down to XTX 100?
PTX (Path Tracing) is the suspected next paradigm. Will it they move to 5 digits or 6? Do the tiers become hundreds or thousands or do they stay in the same place?
Just to think that this all started with the GeForce 1, 2, 3, 4, and then oh shit, now we’re up to 5900? It’s probably Radeon’s fault because no one knew how it iterated all the way to a Radeon 9800 all of a sudden.
Numbers don’t lie people, stop fucking with it. We’re up to the GeForce 20 basically. Marketing is a helluva a drug.
I used to joke with a friend when they asked why I didn't upgrade my graphics card (a 9800 GT), that it was at least 9 times better than their 1080 Ti.
Somehow I hope Intel really does successfully break into the GPU game with more rationally named products. And yes I know how absurd their names can get.
How would you suggest rationally naming GPU products? Tflops, shaders, means nothing to the average consumer. Can this one run the newest call of shooty on the “best” settings? GPU performance is usually presented as relative to another piece of hardware to demonstrate the improvements.
It doesn't have to be based on a particular metric, they should just pick a numerical sequence and stick to it. I would suggest:
G-M, where G is the generation and M is the market segment. They could even make M a letter (A, B, C) to avoid confusion (rather than using 50, 60, 70, 80, which was the closest thing they got to a convention). A=awesome, B=better, C=consumer. They can stick a W on the front for workstation cards, and give them another letter at the end for special high-memory runs or whatever.
Do they have an internal product codes separate from marketing names, e.g. Model A2649, Model 1900, 0F10383, 11A511, 3270, 993, etc., or are their internal systems all like "RTX6000ADA"?
I think they do it on purpose to keep the average consumer from being able to understand the price/performance to expect between the SKUs in each generation. If they’re selling what should be a 4060Ti as a 4080 [1] they need to “reset” the naming scheme every once in a while when they run out of names at the top end.
IMO it’s shrinkflation for GPUs and the naming confusion is used as an attempt to hide it. My prediction is they change the entire naming scheme for the next generation because the current scheme invites historical comparison and criticism like the Reddit post I linked.
Watch them switch to pure, meaningless names. It wouldn’t surprise me.
I think they're intentionally going for confusion, at this point. I can't pinpoint which gen really pushed things in a bad direction, but it existed in the 3000 series, and an LTT video from yesterday even put it on full display, claiming a laptop has hardware that it doesn't have.
Disorientation marketing is a technique to prevent customers from developing their own mental model of how a product should be contextualised (better or worse than competitor/predecessor?). In place of such a model, the vendor substitutes their own messaging.
Based on the response of PC gamers on reddit to the 40 series cards, that's probably a good thing. People with 1440p@60Hz monitors complaining about cards meant to run 4k@144Hz being too expensive.
They have also created confusion (intentional or not, who knows) with the RTX 4080 12GB and RTX 4080 16GB having significant differences other than just the memory size. Some conjecture I've read around the internet is that it was the product the RTX 4070 should have been but was given the 4080 designation due to it's price bracket being significantly higher than previous x070 cards.
I just want to add that they use the same numbering scheme on the laptop versions of the RTX cards which is just awful. Some of the laptop RTX cards are really equivalents to prior generation desktop cards, but it is not indicated on any level because a desktop RTX 20xxx is named the same as the laptop 20xxx.
They stopped giving a shit. They want you to upgrade to the next series ala Apple.
In what world is that not fraud? Imagine if Ford were to advertise a new F150 for an (apparently) good price and then offer two models:
"F150 V8": an expensive truck with a V8 motor.
"F150 V6" a cheaper V6 sedan.
Clearly, these are two different products, even though they can do the same thing. You're sneakily using a known naming convention in your official product name to deceive customers on the price perception of the 4080 and the performance of the 4080 12GB.
Why can't the marketing people who come up with this crap pick decent, incrementing product model numbers?
There's a "renaissance" of this bullshit in the entertainment space, too, where apparently these loons think that resetting the numbers is cool. For example, "Doom" (2016) instead of the more sensible "Doom 4", because fuck that game that came out in 1994!
Or how Nvidia simply reset all their product numbers a while ago. My first graphics card was a geforce TI 4200! And now they are back to the 4000s!
I think the Doom thing is different as some of the customers werent eating solid food when the original came out.
The NVidia thing is a reflection of them having a hard time segmenting the market. The old “enterprisy engineers” vs “gamers” hasn’t worked for a long time, and they feel they deserve a bigger piece of the action on the server/cloud side.
They have an insatiable lust for higher numbers, that requires going out of sequence. My suggestion: just add another circle for each generation! So now we can be on DOOOOOM, and NVIDIA could have the GTX 10000000000000000000. I guess they'll want to use calculator notation, so like GTX 1E20, wait, dang it, my joke suggestion has become better than their current naming scheme.
RTX 40 series doesn't support NVLink, so sadly it's not an option. Pretty sure this was a strategic exclusion by Nvidia to make sure you couldnt get cheaper cards and link them together for parallel compute, and instead push these more expensive prosumer cards.
Yup. They started this transition away from NVLINK in the same generation in which they introduced their tensor cores.
It's part of their ML profit strategy.
There's really nothing you can do outside of soldering on higher capacity RAM chips onto a 4090. There are some 20GB 3080 cards floating around out there.
Yeah I can’t replace the 1080 Ti I have with any of the 40xx cards in my DAN A4.1 case. They’re all quoted as a little bit too big. And with EVGA out, who is gonna make a quiet card for me?
No, on the contrary, it is the fully enabled chip used in 4090, with all the 18k shaders a.k.a. CUDA cores available.
In RTX 4090, 2k of the 18k are disabled, so only 16k are available.
In RTX 6000, presumably both the clock frequency and the supply voltage are lower, for better reliability, and that lowers the power consumption from 450 W to 300 W.
This is probably not targeted at gamers. I'd bet the RTX 4090 will have far superior gaming performance at a lower price, while this will likely have superior CUDA/tensor performance. I can't see how ECC RAM would be beneficial to gaming, but for drafting, scientific applications, machine learning, etc, it'd be pretty valuable.
So the chances are pretty high this would go into a machine with ECC RAM for the system as well.
Is this the third naming convention in 3 generation or is it a return to the previous naming convention? Because it looks like this is intended to be deliberately confusing.
So we had in the last 3 generations;
Quadro RTX 6000, RTX A6000 and now RTX 6000.
I wonder if this is a lower powered dual slot card, simply because the buyers of workstation models from HP, Dell etc would be unreasonably hamstrung by a three slot model - traditionally these systems have a full powered x16 lane in the location that would be taken by a three slot cooler.
Is Nvidia getting in on the same joke USB and HDMI are?
Serious question.
They have a RTX 4090, RTX 4080, and RTX 4080 but less good. And now RTX 6000?
What happens in 2 generations when they want to make a RTX 6090 and so on? Why are there 2 very different RTX 4080?
I mean, really, this started at the 3000 series. I have a laptop with a dock that contains a RTX 3080, full stop. But you have to dig into the specs to learn it's not a full RTX 3080. But it's also not a RTX 3080 Max Q? It's a "RTX 3080 Laptop" whatever that's supposed to mean. But that's not in the marketing. It's fully implying that it's a full RTX 3080.
This is how users get confused and angry when their hardware isn't working as expected.
136 comments
[ 3.7 ms ] story [ 216 ms ] threadAt least on the workstation side - i do a lot of solidworks-based 3d modeling. The A6000 can easily, for example, raytrace in real time anything i can even find to throw at it. Like models that Solidworks still has trouble opening and rendering normally can raytrace instantly.
What part of workstation software is still GPU bound at this point on high end GPU's?
I guess if CAM was taking more advantage of GPU's, i could see it useful there, but on the modeling side, i honestly don't get it.
That's working on Hero assets for current-tech games. UE5 Nanite assets and film assets are much heavier.
I’ve done color correction in games and it’s computationally very cheap to throw a LUT in there, so clearly something else is going on?
In a full color grading session, you can have power windows. You can have those power windows tracking. You can have noise reduction. You can have other filters. You can do so so much. All of that takes up memory. You can be dealing with 8k footage. You can have many many many layers of that footage. You can be grading in >16bit color space.
You can do all of that with very little VRAM at a snails pace, or you can do it in real-time with all of that VRAM and keep the client in a supervised session happy. With all of that VRAM, you can export at faster than real-time for delivering the content to the client at the end of the session rather than the next day after an overnight render.
Here's the full table pasted:
GPU Engine Specs: NVIDIA CUDA® Cores 16384 Boost Clock (GHz) 2.52 Base Clock (GHz) 2.23
Memory Specs: Standard Memory Config 24 GB GDDR6X Memory Interface Width 384-bit
Technology Support: Ray Tracing Cores 3rd Generation Tensor Cores 4th Generation NVIDIA Architecture Ada Lovelace NVIDIA DLSS 3 NVIDIA Reflex Yes NVIDIA Broadcast Yes PCI Express Gen 4 Yes Resizable BAR Yes NVIDIA® GeForce Experience™ Yes NVIDIA Ansel Yes NVIDIA FreeStyle Yes NVIDIA ShadowPlay Yes NVIDIA Highlights Yes NVIDIA G-SYNC® Yes Game Ready Drivers Yes NVIDIA Studio Drivers Yes NVIDIA Omniverse Yes Microsoft DirectX® 12 Ultimate Yes NVIDIA GPU Boost™ Yes NVIDIA NVLink™ (SLI-Ready) No Vulkan RT API, OpenGL 4.6 Yes NVIDIA Encoder (NVENC) 2x 8th Generation NVIDIA Decoder (NVDEC) 5th Generation AV1 Encode Yes AV1 Decode Yes CUDA Capability 8.9 VR Ready Yes
Display Support: Maximum Digital Resolution (1) 7680x4320 Standard Display Connectors HDMI(2), 3x DisplayPort(3) Multi Monitor 4 HDCP 2.3
Card Dimensions: Length 304 mm Width 137 mm Slots 3-Slot (61mm)
Thermal and Power Specs: Maximum GPU Temperature (in C) 90 Graphics Card Power (W) 450 W Minimum System Power (W) (4) 850 W Supplementary Power Connectors 3x PCIe 8-pin cables (adapter in box) OR 450 W or greater PCIe Gen 5 cable
This is not an oversight on NVIDIA's part, this is intentional product segmentation.
Personally, I would buy a couple of 3090's with NVLink and call it day if it was for personal use. I would wait and see actual performance before even thinking about Lovelace cards. Nvidia makes all kinds of unsubstantiated claims during these events.
NVLink is important if (1) if your model doesn't fit in the RAM on a single card or (2) you want to increase batch size past the point of each batch fitting in the RAM of a single card or (3) to increase training speed by spreading the workload.
https://www.nvidia.com/en-us/design-visualization/rtx-a6000/
https://www.nvidia.com/en-us/design-visualization/rtx-6000/
You could buy 2 x 3090's with NVLINK for half that. I recently bought a datacenter M40 with 24GB of RAM for $150 on Ebay then added water cooling. The Dell Precision 5810 I bought second hand can now run some really interesting stuff on those 24GB of RAM albeit slower than a 3090. For less than $500 I can have two watercooled M40's that will run almost anything I can throw at them.
None of major ML frameworks such as Pytorch of TF support that. I’m not sure why.
Big fuck you for nvidia with this release prosumer market is dying right now and there is no alternative cause AMD software is so bad you are forced to use CUDA in for example DL/DS.
You reap what you sow.
I remember writing CUDA parallelized C++ code in 2010 and it was a piece of cake.
Nope, AMD never gave them a choice.
Has Nvidia taken advantage of their position? Absolutely. But don't fault the ML community when there was nothing else available.
OpenCL has existed for about as long as CUDA, and can be used on GPUs from any of the major manufacturers. What makes OpenCL so unsuitable for ML that the ML community just had to use CUDA?
NVIDIA also partnered up very quickly with many big players in the game, the sales people went to work, but they had the technological feats to back it up.
After that, it's the network effect.
Now that AMD has money, hopefully they have fixed their stack, but I'm still in "once bitten, twice shy" mode. I want to see someone else in my field using AMD on tasks I care about before I try team red again.
In contrast, CUDA compiles down to a device agnostic intermediate language so they can support the same code on several generations of hardware, limited only by the CUDA feature level available on the hardware.
AMD still has a long way to go for now. Both in terms of 'primary' functionality like supported hardware and compilers and 'secondary' functionality like detailed profiling and debugging tools.
They may be hoping to live off gamers and big server contracts (eg Frontier) until they get their software stack in a decent position and outgrow their current reputation of having buggy software.
On a side note, AMD did a different (yet interesting) move by embracing the FOSS culture (working with uptreams).
Not even counting Eastern Europe and Asia where you get fraction of that.
Then again, I've been places that are stingy about printing, which didn't make much sense to me, so I guess they have some... other way of looking at this stuff.
[1]https://www.nvidia.com/content/dam/en-zz/Solutions/design-vi...
Or since Nvidia is historically stupid with naming systems and can't stick with them, maybe the RTX 4-digit numbering goes away before 6000 series.
They had a chance to keep a numbering system going when they did the GTX 1xx, GTX 2xx, GTX 3xx.... GTX 9xx, GTX 10xx, and could have gone to GTX 11xx and been set for as long as they wanted (well over a hundred years) counting up toward GTX 99xx), but instead they went to GTX 20xx and RTX 30xx so now they're going to run out of numbers again.
Bets on what's next? Do we hit RTX 10,000 and then jump back down to XTX 100?
Numbers don’t lie people, stop fucking with it. We’re up to the GeForce 20 basically. Marketing is a helluva a drug.
G-M, where G is the generation and M is the market segment. They could even make M a letter (A, B, C) to avoid confusion (rather than using 50, 60, 70, 80, which was the closest thing they got to a convention). A=awesome, B=better, C=consumer. They can stick a W on the front for workstation cards, and give them another letter at the end for special high-memory runs or whatever.
So this could be something like the: w20a-h.
Few examples: there was GeForce MX 4000 (in 2003), GeForce GTX 480 (in 2010) and now GeForce RTX 4080.
Intel at least had the decency to go into five figure numbers with their CPUs.
IMO it’s shrinkflation for GPUs and the naming confusion is used as an attempt to hide it. My prediction is they change the entire naming scheme for the next generation because the current scheme invites historical comparison and criticism like the Reddit post I linked.
Watch them switch to pure, meaningless names. It wouldn’t surprise me.
1. https://old.reddit.com/r/pcmasterrace/comments/xk2nsf/the_40...
Is that really a thing? That strikes me as the same level of shady practices as things like Planned Obsolescence.
[1]https://www.nvidia.com/en-us/geforce/graphics-cards/40-serie... (you'll have to manually scroll down and click the Specs tab)
They stopped giving a shit. They want you to upgrade to the next series ala Apple.
"F150 V8": an expensive truck with a V8 motor.
"F150 V6" a cheaper V6 sedan.
Clearly, these are two different products, even though they can do the same thing. You're sneakily using a known naming convention in your official product name to deceive customers on the price perception of the 4080 and the performance of the 4080 12GB.
Maybe the EU can whip them into shape.
Yes, it’s a quadro card.
There's a "renaissance" of this bullshit in the entertainment space, too, where apparently these loons think that resetting the numbers is cool. For example, "Doom" (2016) instead of the more sensible "Doom 4", because fuck that game that came out in 1994!
Or how Nvidia simply reset all their product numbers a while ago. My first graphics card was a geforce TI 4200! And now they are back to the 4000s!
Get a few large channels on board and hopefully they won't try it again
Unfortunately, it is probably also a way to get your early review sample privilege revoked.
The NVidia thing is a reflection of them having a hard time segmenting the market. The old “enterprisy engineers” vs “gamers” hasn’t worked for a long time, and they feel they deserve a bigger piece of the action on the server/cloud side.
Maybe NVIDIA will help people slowly start to realize that employees should not have to play the telephone game to make purchases.
It's part of their ML profit strategy.
There's really nothing you can do outside of soldering on higher capacity RAM chips onto a 4090. There are some 20GB 3080 cards floating around out there.
I guess even for $6k on a graphics card you still can't effectively future proof anymore.
Which confuses me, is this one of the 4080 chips with a lot of memory?
In RTX 4090, 2k of the 18k are disabled, so only 16k are available.
In RTX 6000, presumably both the clock frequency and the supply voltage are lower, for better reliability, and that lowers the power consumption from 450 W to 300 W.
So the chances are pretty high this would go into a machine with ECC RAM for the system as well.
They have proven to be the best at this discipline with no competitor in sight.
Consumers of the world are in awe.
Catch a grip nvidia - this is a joke.
Serious question.
They have a RTX 4090, RTX 4080, and RTX 4080 but less good. And now RTX 6000?
What happens in 2 generations when they want to make a RTX 6090 and so on? Why are there 2 very different RTX 4080?
I mean, really, this started at the 3000 series. I have a laptop with a dock that contains a RTX 3080, full stop. But you have to dig into the specs to learn it's not a full RTX 3080. But it's also not a RTX 3080 Max Q? It's a "RTX 3080 Laptop" whatever that's supposed to mean. But that's not in the marketing. It's fully implying that it's a full RTX 3080.
This is how users get confused and angry when their hardware isn't working as expected.