The future of AMD in graphics

[Innokentij]

Now that AMD is beaten in every metric possible from low end to the moon and back + power consumption and new technologoy. What can they do but lower prices and bend the knee to nvidia? Will Amd ever come back from this? Nvidia has total domination now, the hardcore crowd for AMD is basically dead and most switched to nvidia by looking at steam hardware use.

 

[Markfw]

For DP computational tasks, the Radeon VII is king, and this is one area AMD is up on.

 

[insertcarehere]

Navi needs to be a slam dunk architecture-wise for them at this point, something like 50% more performance/Watt to be possibly competive with whatever Nvidia pushes out at 7nm.

 

[Markfw]

Navi needs to be a slam dunk architecture-wise for them at this point, something like 50% more performance/Watt to be possibly competive with whatever Nvidia pushes out at 7nm.

I did not say they were competitive, I was just saying the DC people, and others that care about compute ability still get the Radeon VII for example, as it is the best. They have at least one area they are ahead on.

 

[ThatBuzzkiller]

For AMD, it's all about the deep learning and semi-custom contracts ...

For there to be any hope of AMD ever becoming competitive in PC graphics performance again is for more developers to adopt DX12 and there's still a couple of major developers out there who still have yet to do so despite their games being commonly used as benchmarks for many reviewers ...

 

[Guru]

They are winning major companies over and selling more and more workstation and professional graphic cards. In fact quarter on quarter and year on year they are significantly up. They cards are general compute powerhouses that can run a huge variety of computational work really fast. Even Nvidia's Volta is too weak across a wide range of computational workloads. It might beat AMD offerings in few workloads by a big margin, but it losses across a wide range.

Furthermore AMD can lower prices and stay competitive that way. I would definitely buy a Vega 56 over the 1070 or 1660ti at the same price. RX 580 8GB has been as low as $200, RX 570 as low as $160, these are tremendous value. GTX 1060 6GB has not moved bellow $250, apart from very few specific sales and promotions or low cost single slot cards from partners like Evga.

But AMD actually have the lead over Nvidia, they will release 7nm graphic cards somewhere around June/July and will have the process lead for over 6 months, likely 9 months before Nvidia can come with a new architecture.

In the meantime they are competitive with their price drops, again who would buy a GTX 1060 or 1660ti over a RX 580 or Vega 56, especially since the RX 580 can be had for $200-220 Heck with sales and promotions you can end up buying a RX 580 for as low as $180, that is insane. RIGHT NOW there are 2 sales on newegg from powercolor and XFX selling their 580 8GB for $180 and $190.

RX 570 goes for as low as $140 on newegg and comes with 2 new games with it. Right now there are 3 sales on newegg.

 

[SPBHM]

Navi is being worked on for a while and with close ties to the next gen consoles from Sony/MS
so I expect it to be fairly good...

 

[happy medium]

My opinion,?
Next 2 or 3 years....

In 2 years Intel will release a CPU that will beat AMD by 25%.
Ryzen will run out of gas.

On the APU front, Intel will start to be very competitive also with their new IGP.

On the GPU front Nvidia will have a incredible 7nm GPU and continue the one year advantage they have.

Intel will start to be a player in the mid range GPU war IF THEY CAN GET THEIR DRIVERS RIGHT.

All tech companies will feel the squeeze of moving smaller than 7nm.

 

[VirtualLarry]

In 2 years Intel will release a CPU that will beat AMD by 25%.
Ryzen will run out of gas.

All tech companies will feel the squeeze of moving smaller than 7nm.

I don't buy the first part for a second. Intel's CPU strengths, rely on their process technology, and as of now, they're actually BEHIND AMD. If they can catch up on process technology, I question if they will ever be able to beat the Zen architecture. In short, I don't see it.

Intel will start to be a player in the mid range GPU war IF THEY CAN GET THEIR DRIVERS RIGHT.

Third time's a charm? I won't hold my breath.

 

[insertcarehere]

They are winning major companies over and selling more and more workstation and professional graphic cards. In fact quarter on quarter and year on year they are significantly up. They cards are general compute powerhouses that can run a huge variety of computational work really fast. Even Nvidia's Volta is too weak across a wide range of computational workloads. It might beat AMD offerings in few workloads by a big margin, but it losses across a wide range.

Starting from a very low baseline in workstation and compute it is far more easy to gain marketshare than lose it, yes.

Furthermore AMD can lower prices and stay competitive that way. I would definitely buy a Vega 56 over the 1070 or 1660ti at the same price. RX 580 8GB has been as low as $200, RX 570 as low as $160, these are tremendous value. GTX 1060 6GB has not moved bellow $250, apart from very few specific sales and promotions or low cost single slot cards from partners like Evga.

Constant sales, bundles and price slashing is great for consumers but do nothing but hurt AMD's margins. Vega 56 for example, has a graphics core that has nearly twice the transistor count, and incorporates expensive HBM, vs the Nvidia counterparts its selling at similar prices. Given that no company is benevolent, I doubt AMD wanted to do these price cuts and sales all the time, they just thought it was necessary to be competitive in the broader marketplace.

But AMD actually have the lead over Nvidia, they will release 7nm graphic cards somewhere around June/July and will have the process lead for over 6 months, likely 9 months before Nvidia can come with a new architecture.

They have a process lead right now, Radeon VII uses 7nm and the only thing it has proven is that the Vega architecture needs to go to the rubbish bin.

 

[Markfw]

My opinion,?
Next 2 or 3 years....

In 2 years Intel will release a CPU that will beat AMD by 25%.
Ryzen will run out of gas.

On the APU front, Intel will start to be very competitive also with their new IGP.

On the GPU front Nvidia will have a incredible 7nm GPU and continue the one year advantage they have.

Intel will start to be a player in the mid range GPU war IF THEY CAN GET THEIR DRIVERS RIGHT.

All tech companies will feel the squeeze of moving smaller than 7nm.

I know its opinion, but WOW, you sure have it in for Intel.Nvidia, you think they will always be on top, and can do no wrong.

Please wake up and smell the roses, the world is changing.

 

[tamz_msc]

I find it likelier that Valve will come up with a Half Life 3 which lives up to the hype than AMD coming up with a competitive GPU on all metrics.

 

[SPBHM]

I find it likelier that Valve will come up with a Half Life 3 which lives up to the hype than AMD coming up with a competitive GPU on all metrics.

I thought the same about their CPUs before Zen

 

[Ajay]

Navi needs to be a slam dunk architecture-wise for them at this point, something like 50% more performance/Watt to be possibly competive with whatever Nvidia pushes out at 7nm.

It won’t be a 'slam dunk' AFAIK. Still CGN and just a mid-range GPU. Next Gen Radeon architecture is their next chance to get on a more equal footing with Nvidia.

 

[tamz_msc]

I thought the same about their CPUs before Zen

Zen happened because they had a long-term plan with concrete outcomes following from on-time execution. RTG has been lacking a plan ever since they were formed in 2015.

 

[IntelUser2000]

I thought the same about their CPUs before Zen

I think if AMD ever does get competitive, it might be a few years away. Yes, like tamz_msc says, if they might become a viable player in the future, the plans must form now.

Not too positive about Intel's plans in the discrete graphics part either. Raj's leadership didn't amount to much at AMD. Yea, maybe he was indeed constrained by something there, but he'll have to prove himself just like Intel as a whole needs to prove themselves in graphics.

Gen 11 iGPU will probably slightly beat Vega iGPUs. That's not really a surprise. They've historically been able to be competitive with last-gen AMD iGPUs. Xe will need better-than-normal jump to be competitive with Nvidia and AMD current gen. Nvidia as a company is also better run then Intel.

 

[sandorski]

Zen happened because they had a long-term plan with concrete outcomes following from on-time execution. RTG has been lacking a plan ever since they were formed in 2015.

No one outside of AMD knew that Zen was coming. To say that RTG has "no plan" is the same as someone years ago saying "AMD has no plan".

 

[SpaceBeer]

I think the main reason why Intel's Gen 11 IGP is faster than Vega IGP (acc. to one benchmark we have seen) is due to faster memory (+ eDRAM). I mean, it seems it is not faster than 2400G (with fast RAM I suppose). Plus, we are comparing Intel's next generation with AMD's current generation graphics

Spoiler: bench

I don't think Intel is big threat to AMD. It will be harder for them to gain new contracts and customers, but they won't lose many either. I think nVidia is the one that should be more concerned of Intel's GPU and AI/ML plans. Cause AMD customers will buy AMD products as long as they are competitive. And they are, even Vega (56, 64, VII). And we know AMD customers don't like neither Intel nor nVidia. On the other hand, most nVidia buyers already use Intel's products (both consumers and enterprises). And if Intel offers them competitive product for gaming, HPC, ML... many of them will go for it. No one was ever fired for buying Intel :d

 

[tamz_msc]

To say that RTG has "no plan" is the same as someone years ago saying "AMD has no plan".

This is what is meant by having a plan:

This slide dates from 2015. It had a simple, bold statement regarding what AMD expected from Zen and the final delivery surpassed these expectations.

Try as you might you won't find something similar from RTG. Hence I say they don't have any plans. They've inherited the legacy of an aging GPU architecture and have no idea what should be the next logical step from it. At this point they're throwing stuff at the wall hoping that it'd stick. The Radeon VII is the most glaring example of their lack of plans.

 

[IntelUser2000]

I think the main reason why Intel's Gen 11 IGP is faster than Vega IGP (acc. to one benchmark we have seen) is due to faster memory (+ eDRAM).

No.

GFXBench is basically for mobile. Intel was in mobile for quite a bit, and they spent resources optimizing for it.

Their current Gen 9 graphics(HD 620/630) are already quite competitive in GFXBench with the Vega iGPU. Where it really matters is how it does in PC games.

None of the leaks matters in terms of performance. Only the release date numbers really matter.

 

[beginner99]

Raj's leadership didn't amount to much at AMD. Yea, maybe he was indeed constrained by something there, but he'll have to prove himself just like Intel as a whole needs to prove themselves in graphics.

Raja focused on playing stupid power games. Obviously RTG was formed due to his idea and his obvious plan was to sell it to intel. AMD GPUs went downhill from the moment he was head and never recovered since. It's a good thing he left for AMD.

 

[IntelUser2000]

AMD and Intel will likely go nowhere in the Datacenter GPU space unless Nvidia falls over flat on its face. Past execution suggests Nvidia typically outmaneuvers both, because its a better run company. Oh, and Nvidia is the incumbent with CUDA.

Not sure about AMD but I don't think AI/ML will end up much for Intel. They tend to pursue side projects they abandon in 5-7 years at most. The projects usually start because they are too late and are following a trend.

Big companies like Intel have too much resources and don't know what to do with it outside their core competence. Too much veering from it, and the efforts are tossed in the garbage because its financially motivated, not for technological advancement or being a pioneer at it.

 

[AtenRa]

Since both AMD and NVIDIA are allocating more and more transistors in their GPUs for Compute, AMD will get closer and closer in perf/mm2 and perf/watt.

Lets wait and see where NAVI will go and then we can talk again if AMD can catch up NVIDIA the coming years.

 

[AtenRa]

This is what is meant by having a plan:

This slide dates from 2015. It had a simple, bold statement regarding what AMD expected from Zen and the final delivery surpassed these expectations.

Try as you might you won't find something similar from RTG. Hence I say they don't have any plans. They've inherited the legacy of an aging GPU architecture and have no idea what should be the next logical step from it. At this point they're throwing stuff at the wall hoping that it'd stick. The Radeon VII is the most glaring example of their lack of plans.

There is the 25x20 efficiency initiative in AMD for both CPUs and GPUs.

 

[IntelUser2000]

The difference between Zen and AMD's datacenter GPU efforts are that Zen uses the same ISA as Intel chips, but different ones from Nvidia GPUs.

That's why its much harder. AMD and Intel has much better chance on the consumer side. Actually for AMD taking even a few % of market away from Nvidia can be profitable. For Intel it'll be nowhere enough.