RX480 vs GTX 1060 at same clocks test?

[FordGT90Concept]

Most large studios have a variety of hardware for QA testing. Bulk of dev is done on NVIDA. In other words, they code for NVIDIA and patch for AMD.

id Software on DOOM:
http://www.dsogaming.com/interviews...n-mega-textures-pbr-global-illumination-more/

We choose Vulkan, because it allows us to support Windows 7 and 8, which still have significant market share and would be excluded with DirectX 12. On top of that Vulkan has an extension mechanism that allows us to work very closely with AMD, NVIDIA and Intel to do very specific optimizations for each hardware.

Vulkan reference:
https://www.khronos.org/registry/vulkan/specs/1.0/styleguide.html#extensions
Vendor docs:
https://github.com/KhronosGroup/Vulkan-Docs/tree/1.0/doc/specs/vulkan/appendices

GCN specific details:
http://gpuopen.com/gcn-shader-extensions-for-direct3d-and-vulkan/

The GCN architecture contains a lot of functionality in the shader cores which is not currently exposed in current APIs like Vulkan™ or Direct3D® 12. One of the mandates of GPUOpen is to give developers better access to the hardware, and today we’re releasing extensions for these APIs to expose additional GCN features to developers.

 

[RejZoR]

Does it matter? If game is "patched" for AMD when it goes gold, what does it matter?

 

[FordGT90Concept]

Patching is a lot like "feature creep." Anything that is added as an afterthought will have suboptimal implementation.

I edited the previous post giving examples of commands GCN supports that a lot of developers don't use. GCN performance therefore suffers.

Optimizing software is hard work. Most developers only target a minimum framerate and, instead, rely on continuous improvements in overall hardware capability as well as in-game settings to reduce hardware load.

 

[BiggieShady]

I can't find any good articles that discuss how optimization for NVIDIA and AMD works for developers.

Siggraph slides are the best source.
IMO the biggest optimization a game needs (think doom opengl on nvidia) and the one slide that sums it up for OpenGL on nvidia:

Yeah, good old feeble OpenGL 3 years ago ... funny how same optimizations principles are now used in Vulkan and Dx12 ... semi manually

 

[RejZoR]

Patching is a lot like "feature creep." Anything that is added as an afterthought will have suboptimal implementation.

I edited the previous post giving examples of commands GCN supports that a lot of developers don't use. GCN performance therefore suffers.

Optimizing software is hard work. Most developers only target a minimum framerate and, instead, rely on continuous improvements in overall hardware capability as well as in-game settings to reduce hardware load.

DX and OGL had many commands that no one used. So?

 

[FordGT90Concept]

Feature set, sure, but the commands we're talking about drastically improve performance for their given architectures. If you click the link, some of those are memory swapping routines which saves many clockcycles of time.

 

[Vayra86]

Lets just say here that NVIDIA carefully crafts their stuff for each generation to give best benchmarking results for just that specific time and as a result, easier sales. AMD on the other hand always goes POWEEEEEER. It may not be the best for specific time, but it's very universal long term. It's also why AMD cards generally age better than NVIDIA cards, they just have more raw grunt as whole and they perform well even when workloads change when games evolve and demand different design of hardware for optimal operation. GeForce on the other hand may start to struggle faster when this happens...

Correct, but in real world performance the loss 'over time' on Nvidia is below 5% and even in extreme cases remains in the single digit range. Meanwhile the cards do use a lot less power and the performance bar is still raised by 30% year over year almost, whereas AMD is now still populating midrange only. And on CPU, compared to Intel, we've seen the exact same thing, of which Ryzen is simply a continuation. Also, considering DX11 AMD performance, you could easily look back and say GCN was not what it should be for DX11. Throughout all these years AMD never managed to extract the performance you'd expect from 'fat' GPUs like they had. Look at the 7970 for a great example - back then Nvidia was not even rocking delta compression but still had much leaner GPUs that went toe to toe with it, while the clocks were much closer than they are today.

It is the same philosophy as we see everywhere with AMD product: MOAR CORES. In the end this is the way they always do it, and from that point they start optimizing. So far, this philosophy has not resulted in products that top the competition, so as much as you can herald AMD's future proofing, the end result still is inferior. I'm still finding it difficult to praise AMD for releasing 'fat' products, I am much more convinced that this is out of pure necessity to keep themselves in the race with limited resources. Just copy/paste your solution X number of times to get the performance you need is a lot cheaper R&D wise. Therefore I also strongly doubt AMD's goal is primarily 'to build stuff that keeps doing well', I think that is just a result of a design philosophy based on low R&D cost, with a bit of marketing spin applied to it.

 

[RejZoR]

Bulldozer had other problems. Ryzen does not (despite all the drama around CCX units). It's because no one was taking the MOAR CORES as benefit, we're still stuck with god damn quad cores in year 2017. Please make them die already, for the love of god.

 

[Vayra86]

Bulldozer had other problems. Ryzen does not (despite all the drama around CCX units). It's because no one was taking the MOAR CORES as benefit, we're still stuck with god damn quad cores in year 2017. Please make them die already, for the love of god.

But that's not true. There ARE lots of products on CPU with more cores. The reason we haven't embraced them on the consumer end is because the vast majority of consumers doesn't require them at all, even a quad core, and especially a 4c/8t is already heavy overkill for at least 75% of all consumer systems worldwide. And even today the comparison of the 7700K to any other CPU for gaming speaks volumes. Again: efficiency and lean product wins over 'fat' and wide product for these kinds of workloads - and for almost every consumer workload. Also consider mobile: this is where the growth is, and fat product is power hungry.

Just because we enthusiasts ask for it, doesn't mean there is a business case.

 

[EarthDog]

Bulldozer had other problems. Ryzen does not (despite all the drama around CCX units). It's because no one was taking the MOAR CORES as benefit, we're still stuck with god damn quad cores in year 2017. Please make them die already, for the love of god.

ehh... its not so much the cpus are still here its that there is such few pieces of software iut there which makes use of more than 4c8t...

... people were saying go quad when q6600 was out for petes sake... and just within the last two uears was a a quad(no other threads) was considered a 'minimum' for gaming. It will be years more when the minimum is more than 4c8t... so again, amd looking too far into the future imo.

 

[Kanan]

@Vayra86 comparing Ryzen to GCN and come to the conclusion it's as bad received is just that, bad. Ryzen has nothing to do with GCN and is a great thing praised by anyone but you.

GCN on the other hand is much more than "moar cores" you keep repeating this nonsense for a while now and it's still wrong. Aside from that GCN works perfectly fine for DX11 aside from the fact that Nvidia is ahead in technology which then again has nothing to do with "GCN being bad at DX11" which is not true anyway. The fact Nvidia is ahead has to do with tons of other mistakes AMD did, GCN is superior compared to Kepler, but after Nvidia released Maxwell they lost the competition on all accounts despite Fiji and HBM, which was a nice try and technology demonstration but nothing more.

RX 480 has always been faster than GTX 970. It was also faster than R9 290/390 vanilla. But generally not faster than R9 390X and GTX 980. Today, it's usually beating both.

Not really. It still has a hard time to come out ahead of 970, loses against 980, 1060 OC, 390X and 780 Ti as well. My Superposition thread is a great example for that.

 

[Vayra86]

@Vayra86 comparing Ryzen to GCN and come to the conclusion it's as bad received is just that, bad. Ryzen has nothing to do with GCN and is a great thing praised by anyone but you.

GCN on the other hand is much more than "moar cores" you keep repeating this nonsense for a while now and it's still wrong. Aside from that GCN works perfectly fine for DX11 aside from the fact that Nvidia is ahead in technology which then again has nothing to do with "GCN being bad at DX11" which is not true anyway. The fact Nvidia is ahead has to do with tons of other mistakes AMD did, GCN is superior compared to Kepler, but after Nvidia released Maxwell they lost the competition on all accounts despite Fiji and HBM, which was a nice try and technology demonstration but nothing more.

Not really. It still has a hard time to come out ahead of 970, loses against 980, 1060 OC, 390X and 780 Ti as well. My Superposition thread is a great example for that.

Ehhh. don't put words in my mouth. I did not say Ryzen was badly received and I've been on the fence buying a 6c/12t myself... I said it does not have the performance crown, specifically in consumer grade workloads and most notably gaming, where a 'leaner' product takes that crown just like the leaner Nvidia GPUs take the crown every time. We were talking about different GPU arch and design philosophy, and it is clear to see that philosophy applied not just to GPU but also to CPU.

GCN does NOT work perfectly fine for DX11. It delivers the performance that suited its price points on DX11, but the efficiency is out the window entirely and even under DX12 the perf/watt is not what it could be, once again perfectly demonstrated by the leaner products taking the efficiency crown.

If you can't see that parallel in design philosophy, well...

 

[Kanan]

Ehhh. don't put words in my mouth. I did not say Ryzen was badly received and I've been on the fence buying a 6c/12t myself... I said it does not have the performance crown, specifically in consumer grade workloads and most notably gaming, where a 'leaner' product takes that crown just like the leaner Nvidia GPUs take the crown every time. We were talking about different GPU arch and design philosophy, and it is clear to see that philosophy applied not just to GPU but also to CPU.

GCN does NOT work perfectly fine for DX11. It delivers the performance that suited its price points on DX11, but the efficiency is out the window entirely and even under DX12 the perf/watt is not what it could be, once again perfectly demonstrated by the leaner products taking the efficiency crown.

If you can't see that parallel in design philosophy, well...

Not really wanted to put words in your mouth but why buy a cpu you're talking bad about, get a Intel then and be done with it. You'll just get angry about it anyway if it misses your stellar expectations you always have.

GCN did perfectly fine for DX11, it's a matter of definition when to say that and when not. Just because it had efficiency-wise a hard time with Maxwell doesn't mean it was bad. Again compared to Kepler it did very well. I pretty much took care to explain the whole picture and not sound like someone who thinks GCN is perfect. Care to understand me at least.

Ps. A good example why it did great would be 390 and 390X, that despite being "old" encountered GTX 970 and 980.

 

[Vayra86]

Not really wanted to put words in your mouth but why buy a cpu you're talking bad about, get a Intel then and be done with it. You'll just get angry about it anyway if it misses your stellar expectations you always have.

GCN did perfectly fine for DX11, it's a matter of definition when to say that and when not. Just because it had efficiency-wise a hard time with Maxwell doesn't mean it was bad. Again compared to Kepler it did very well. I pretty much took care to explain the whole picture and not sound like someone who thinks GCN is perfect. Care to understand me at least.

Ps. A good example why it did great would be 390 and 390X, that despite being "old" encountered GTX 970 and 980.

But I don't think we are in disagreement on GCN. You and I both see the efficiency problem, and that problem has not been solved with Polaris. We're saying the same thing essentially: GCN had the performance that suited the price points, but it never had the efficiency to keep scaling over time, and it does not have that efficiency today. Look at the RX580 and how its perf/watt plummeted due to the higher clocks. The gap has gotten wider than it was during Kepler.

BTW about my Ryzen stance: I just can't justify the performance gain versus the price for my own situation, but I don't think Ryzen is a bad product. It's doing pretty damn well, but I'd have to clock it to 4.0 to match my current Ivy.

 

[Kanan]

But I don't think we are in disagreement on GCN. You and I both see the efficiency problem, and that problem has not been solved with Polaris. We're saying the same thing essentially: GCN had the performance that suited the price points, but it never had the efficiency to keep scaling over time, and it does not have that efficiency today. Look at the RX580 and how its perf/watt plummeted due to the higher clocks. The gap has gotten wider than it was during Kepler.

Agreed RX 580 is just bad efficiency wise, honestly I would never buy it. Would straight go for 1060 or 1070 then. RX 480 at least was on Maxwell level efficiency wise but RX 580 is a step back to Hawaii levels of efficiency. I called it a "desperate move by AMD in order to counter 1060" two days ago, you see why.

 

[FordGT90Concept]

Tile-based rasterization in Vega should improve power efficiency.

 

[Vayra86]

Agreed RX 580 is just bad efficiency wise, honestly I would never buy it. Would straight go for 1060 or 1070 then. RX 480 at least was on Maxwell level efficiency wise but RX 580 is a step back to Hawaii levels of efficiency. I called it a "desperate move by AMD in order to counter 1060" two days ago, you see why.

Precisely, and at the same time, the only way AMD can make faster GPUs is either to go wider, or to clock higher, on their current architecture. If they clock higher, they lose massively on efficiency, and if they go wider once again, the performance gain will be limited while the die becomes much more expensive to create, yields will suffer and with that, AMD's margin. And there is your link to Ryzen: Ryzen, just like GCN, does not like high clocks, and is a 'wide' CPU. Even the ones with just 4c/8t. Intel Core is 'lean' : drop 8 cores in it and it just clocks a bit lower, while Core also has a 7700K that clocks extremely high.

 

[Kanan]

And there is your link to Ryzen: Ryzen, just like GCN, does not like high clocks, and is a 'wide' CPU.

Well I still have hopes for Ryzen that it will clock higher in future. Ryzen II will anyway, I'm talking about better yields at Ryzen I. Phenom II was a bad overclocker as well when it was first released (well at least if you call over 20% bad - but compared to Intel it was). My own 940 only did 3.7GHz reliably. But later revisions clocked higher. In the end you can compare Ryzen to Phenom but not to FX because FX followed the path of Pentium 4 to be specifically made for highest clocks.

PS. I think what Intel did was combining their experience from Pentium 4 with Pentium 3 to engineer the optimal CPU that not only clocks high but also has great IPC.

 

[cdawall]

You could even go as far as to say Nvidias tflop numbers hold true while AMDs doesn't. The 1070 has high tflop and high perf while the RX 480 misses out on the latter, only coming out on par with the 1060. So I wouldn't hold my breath for Vega being fast just because of high tflop (theoretical) numbers. AMD since GCN 1 and higher generations had always underutilisation problems, if Vega could fix this, it would be great, if not, I dont think so. It depends too much on DX12 optimized games and Vulkan then. But the fact is, DX11 is still way more important so you have to hope it will be nice in DX11 as well to hold its ground. I hope you don't mind me talking about Vega.

Start bitmining with the amd cards and it shows easily that those numbers ring true. They just aren't as usable for a standard end user as one would hope.

Not really. It still has a hard time to come out ahead of 970, loses against 980, 1060 OC, 390X and 780 Ti as well. My Superposition thread is a great example for that.

Superposition also shows the 980ti competing dead even with the 1080 and demolishing the 1070. It isn't exactly the go to for proof of concept. If coded in Vulkan instead of DX11/OGL you would have seen the same graph lean red. All it shows is nvidias brute force gpus hold up well in dx11. This is something we already knew.

 

[Tatty_Two]

Ryzen 1 - Good, new architecture...... toe in the water
Ryzen 2 - Very good/outstanding, fine tuning, improved process & efficiency, better clocks .......... Full dive from 300M into the Niagara falls.

Well that's the kind of philosophy being spoken by some, it of course remains to be seen.

 

[cdawall]

Ryzen 1 - Good, new architecture...... toe in the water
Ryzen 2 - Very good/outstanding, fine tuning, improved process & efficiency, better clocks .......... Full dive from 300M into the Niagara falls.

Well that's the kind of philosophy being spoken by some, it of course remains to be seen.

Phenom 1 vs phenom 2 showed a similar approach.

 

[Kanan]

Superposition also shows the 980ti competing dead even with the 1080 and demolishing the 1070. It isn't exactly the go to for proof of concept. If coded in Vulkan instead of DX11/OGL you would have seen the same graph lean red. All it shows is nvidias brute force gpus hold up well in dx11. This is something we already knew.

Not really. As seen in my thread 980 Ti can only mess with 1080 at retarded clocks at highest voltage and is still clearly behind even then. But this is nothing new and was discovered at 1080 release when PCGH even managed to *best* a 1080 with a Titan X overclocked to 1520 MHz in *games*. 3072 Shaders of the almost same type simply won against 2560 shaders at higher clocks.

 

[cdawall]

Not really. As seen in my thread 980 Ti can only mess with 1080 at retarded clocks at highest voltage and is still clearly behind even then. But this is nothing new and was discovered at 1080 release when PCGH even managed to *best* a 1080 with a Titan X overclocked to 1520 MHz in *games*.

My card is a throw away on the stock bios with a minor amount of voltage added... It is one of the ones up there.

 

[Kanan]

My card is a throw away on the stock bios with a minor amount of voltage added... It is one of the ones up there.

You're a lottery winner, buddy. Nothing more or less. Most 980 Ti aren't doing so great.

 

[cdawall]

You're a lottery winner, buddy. Nothing more or less. Most 980 Ti aren't doing so great.

There are three cards in the shop that hit that same batch of numbers two of which were reference cards. Quite a few cards clock higher than those. With a bios mod for tdp that gb card I posted can go further, but that's more effort than I was will to put forth.