CPU sales, sure. The value per core is through the roof. Nvidia is just too entrenched with gamers and CUDA users for Navi to have a chance. It would take a silicon miracle for Navi to be good enough to convert team green fanboys.
No reason to think that. Ryzen showed that despite years of neglecting gamers (FX really was horrible for games) they come back as soon as the offer is better. Maybe less in the first generation, but in the second of just having the better offer they do (with the Ryzen 2000 series).
AMD is already still popular in the RX 570/580 segment, if Navi were competitive AMD would be fine there.
CUDA might be a different story, that depends on the software and if tensorflow etc is even still relevant then.
I'm collecting benchmarks and creating a global one in https://www.pc-kombo.com/benchmark/games/cpu. The best Intel cpus are faster because they have a higher clock. Where the clock is equal Ryzen is minimally slower, and they are cheaper then while having more threads. Regardless, the Ryzen processors reach very good FPS in current games, look at the FPS of the Ryzen 5 2600 for example (https://www.pc-kombo.com/us/benchmark/games/cpu/compare?ids%...).
Well, that's kind of true. 2600X vs i5-8400 is the goto comparison for that, since they both are at 4 GHz (4.1 vs 4.0), see https://www.pc-kombo.com/us/benchmark/games/cpu/compare?ids%... for that. Yes, the i5-8400 is a bit faster while having a lower tdp and being at the same price, depending on the game of course.
But the 2600X is unlocked and comes with a better cooler. And the Ryzen 5 2600 is cheaper than both while also being unlocked, and the cooler is still better than that of the i5. The 2600 especially is really attractive, even more so if you factor in that AM4 is the better supported platform (it will run on the board you bought for the Ryzen 3 1200 when that series was released, and it is supposed to run the Ryzen 3000 series).
There now is the i5-9400F which is again a little bit faster than the i5-8400 while being cheaper. If that one had a cheaper board than the Z390 chipsets supporting it out of the box it would be a real reason for AMD to lower the price of the Ryzen 5 2600(X).
"Popular" is a very strong word for the 580 and Polaris in general; according to the steam hardware survey[0], the 1060 has a bit over 15% of the users while the 580, 480, 570 and 470 don't even reach 3% combined.
Good point. Maybe what I'm seeing is already included in the jump from 0.68% to 1.20%. Dunno, I really get the impression that the RX 580 is a more popular option than the 1060 in the communities I look at, though many there now pick more expensive RTX cards. Obviously I don't see the whole picture.
I get the feeling that OpenCL is going to become a second-class citizen in favour of ROCm (https://rocm.github.io/), which looks very promising for HPC/deep learning on AMD GPUs.
I haven't got stuck into it yet, but when I was doing some research it seemed like the AMD team had invested a lot of effort in minimizing the work needed to either support/port CUDA-specific code.
I'm optimistic about AMD fighting their way back into contention as far as deep learning goes. I don't know what that means in the gaming world, but I assume a win in the former will help them in the latter.
As to why you should care, it depends at what level you work. I'm not a graphics/HPC programmer, so I can't really comment on what you'd be using it for in those areas.
For DL applications, though, CUDA/CUDNN used to be the only GPU libraries that offered some of the common/optimized operations you'd encounter in deep learning (e.g. 2d convolutions). That meant you were limited to NVIDIA GPUs within all the higher level frameworks (Tensorflow/Pytorch/etc).
With ROCm more developed, it can supplant CUDA as the low-level library for those frameworks. This means you can start using AMD GPUs for training/inference, which are considerably cheaper.
In fact, in the few months that have passed since I last checked, it now appears that ROCm supports TF - big step.
it means you're no longer locked to NVIDIA GPUs
With CUDA users silicon is secondary. Nvidia has huge edge on the software side that is not going away soon. Even if AMD matches the raw performance, the actual performance and development tooling that allows it is what makes the difference.
AMD just don't seem to push the software side hard enough.
CUDA is fairly entrenched, but there is extraordinary little "loyalty" or stickiness among gamers (nor should there be).
Right now nvidia reigns supreme because they have the best peak performance and value proposition. If AMD presents an option that upends this the ratios would change dramatically.
Just as we saw when ATI/AMD share halved over just two years when their designs were no longer competitive.
The reason I argue this is in part because of what happened with Nvidia's GeForce 400 series (Fermi). That generation was an absolute dog (heat, power, driver issues, you name it) versus AMD's absolutely fantastic Radeon 5000 architecture.
Despite the faults in Nvidia's cards, they still outsold AMD something like 10:1. Jim from AdordedTV goes into great depth on this strange era of GPU competition and dives deep into the press at the time and retrospective statistics.
"Competitive" is more than just raw performance, it seems.
What you're describing is called market stability. The more stable a market is, the harder the momentum a brand has shifts towards competitors. It doesn't just take a single inferior generation of products to tank a brand/company and swing the whole market around, you need multiple sequential generations like that for that to happen.
With next-gen xbox, playstation, and Google Stadia all running on AMD they will continue to ship plenty of units on the GPU front. Small market share on discrete GPUs translates to better growth potential, which is the whole point of the article. AMD is already shipping 7nm GPUs, they are just too expensive to be competitive.
As 7nm continues to mature there is significant potential for AMD to grow sales if they are releasing a 2nd gen 7nm product at the same time as Nvidia is releasing 1st gen. Small market share means greater growth potential.
AMD is also in a better position than ever before because of having CPU, GPU, and next-gen custom SOCs consolidated on a single process, which means they are a much higher volume customer for TSMC than ever before, which means priority treatment, and they are manufacturing the broadest range of designs on that node, which means their internal knowledge of it is the best.
3Q is the earliest that they will see any revenue from these products, which is why the are forecasting 2Q decline. 3Q will likely be partial ramp, but 4Q should see a big pop.
I have a GTX 970 and would love to replace it with an AMD card.
But that has more to do with Apple's anti-Nvidia vendetta and me wanting to consolidate my gaming computer into an eGPU than any particular preference in GPU manufacturers.
There has been an increase in submission from poor quality sources in the recent months. I don't mind Rumours, but it needs to be backed up with evidence that make sense. Like the recent Servethehome pieces on EYPC 2 PCI-E count. ( Servethehome is a great news sources anyway )
Can't wait to get my hands on a 64 core Threadripper, but I am not sure about Navi; I'd rather pair it with Titan RTX or whatever is next from NVidia on 7nm. Intel is now interesting only for NUCs to me (their quad core little ones are as fast as 4790k, which is unbelievable, and perfect for eGPU setups).
I'm very bullish on AMD continuing to make gains in server land and workstations but I do wonder what the laptop situation is going to be like in the next generation. Intel put a lot of work into power efficiency not just at a chip level but at the chipset and driver level too. This has brought large gains for battery life and while I expect AMD to catch up eventually I'm not sure that it'll happen in Ryzen 2.
I just wished AMD invested in truly supporting DL in their GPUs by writing software (at least get one big player like pytorch or keras working 100% there).
I am very happy with Ryzen in my Linux box and would certainly use a GPU from them as well.
I feel like every year for the last decade I've been reading about how AMD is just about to make major strides in the CPU and GPU markets. It feels like it never comes. I would welcome it, but it doesn't feel like enough to catch up to perceived Intel performance; at this point, it seems like they need to really exceed it.
41 comments
[ 3.5 ms ] story [ 67.0 ms ] threadThis piece is fluff.
No reason to think that. Ryzen showed that despite years of neglecting gamers (FX really was horrible for games) they come back as soon as the offer is better. Maybe less in the first generation, but in the second of just having the better offer they do (with the Ryzen 2000 series).
AMD is already still popular in the RX 570/580 segment, if Navi were competitive AMD would be fine there.
CUDA might be a different story, that depends on the software and if tensorflow etc is even still relevant then.
Thank you :)
But the 2600X is unlocked and comes with a better cooler. And the Ryzen 5 2600 is cheaper than both while also being unlocked, and the cooler is still better than that of the i5. The 2600 especially is really attractive, even more so if you factor in that AM4 is the better supported platform (it will run on the board you bought for the Ryzen 3 1200 when that series was released, and it is supposed to run the Ryzen 3000 series).
There now is the i5-9400F which is again a little bit faster than the i5-8400 while being cheaper. If that one had a cheaper board than the Z390 chipsets supporting it out of the box it would be a real reason for AMD to lower the price of the Ryzen 5 2600(X).
[0] https://store.steampowered.com/hwsurvey/videocard/
Right now AMD is kinda usable-ish on DX and so and so in OpenCL if one spends a lot of time optimizing specifically for them.
Their OpenGL support is atrocious. Their Vulkan implementation is full of bugs.
AMD has had issues with drivers for ages. And there is no improvement in sight.
I haven't got stuck into it yet, but when I was doing some research it seemed like the AMD team had invested a lot of effort in minimizing the work needed to either support/port CUDA-specific code.
I'm optimistic about AMD fighting their way back into contention as far as deep learning goes. I don't know what that means in the gaming world, but I assume a win in the former will help them in the latter.
https://gpuopen.com/ported-caffe-hip-heres-happened/
As to why you should care, it depends at what level you work. I'm not a graphics/HPC programmer, so I can't really comment on what you'd be using it for in those areas.
For DL applications, though, CUDA/CUDNN used to be the only GPU libraries that offered some of the common/optimized operations you'd encounter in deep learning (e.g. 2d convolutions). That meant you were limited to NVIDIA GPUs within all the higher level frameworks (Tensorflow/Pytorch/etc).
With ROCm more developed, it can supplant CUDA as the low-level library for those frameworks. This means you can start using AMD GPUs for training/inference, which are considerably cheaper.
In fact, in the few months that have passed since I last checked, it now appears that ROCm supports TF - big step. it means you're no longer locked to NVIDIA GPUs
AMD just don't seem to push the software side hard enough.
Right now nvidia reigns supreme because they have the best peak performance and value proposition. If AMD presents an option that upends this the ratios would change dramatically.
Just as we saw when ATI/AMD share halved over just two years when their designs were no longer competitive.
Despite the faults in Nvidia's cards, they still outsold AMD something like 10:1. Jim from AdordedTV goes into great depth on this strange era of GPU competition and dives deep into the press at the time and retrospective statistics.
"Competitive" is more than just raw performance, it seems.
As 7nm continues to mature there is significant potential for AMD to grow sales if they are releasing a 2nd gen 7nm product at the same time as Nvidia is releasing 1st gen. Small market share means greater growth potential.
AMD is also in a better position than ever before because of having CPU, GPU, and next-gen custom SOCs consolidated on a single process, which means they are a much higher volume customer for TSMC than ever before, which means priority treatment, and they are manufacturing the broadest range of designs on that node, which means their internal knowledge of it is the best.
3Q is the earliest that they will see any revenue from these products, which is why the are forecasting 2Q decline. 3Q will likely be partial ramp, but 4Q should see a big pop.
But that has more to do with Apple's anti-Nvidia vendetta and me wanting to consolidate my gaming computer into an eGPU than any particular preference in GPU manufacturers.
| If the next-gen CPUs live up to the hype ...
| if Navi ends up being a worthwhile solution ...
| [a] 24 percent year-over-year [decline] ...
| single-digit percentage revenue growth [...] could end up being conservative ...
| depending on how things play out ...
| It's all speculation at this point ...
| though that bit is less clear ...
| There are plenty of "ifs" ...
... there is nothing BUT ifs in this piece.
Thanks!
I am very happy with Ryzen in my Linux box and would certainly use a GPU from them as well.