Ryzen possibly held back by Nvidia's drivers

[Mothballs]

Adored means well, But they're one of the ones that unless multiple other sources agree with (Not Joker) I'll take with a grain of salt.

They're very good for putting out a theory and looking through other peoples data, they just lack the resources to prove things conclusively themselves. Which means if they say something, someone else is going to have to test it for me to 100% Believe it.

That said, they have put a ton of foot work into this. Interesting, and worth looking into.

[TOMPPIX]

well... this changes my view on dx12. 

[DocSwag]

2 hours ago, Terryv said:

Isn't Pascal also just Maxwell with a node shrink?

 

I've long suspected that Nvidia doesn't actually do asynchronous compute on a hardware level. This is further evidence to that.

 

 

Also, DX12 being windows 10 exclusive is crap, I do hope Vulkan becomes the API of choice. I don't like being forced into Microsoft.

Mostly, yes, most of the actual compute stuff (ROPs, Cores, etc.) weren't really touched, though memory compression was improved and some extra features were added, including SMP (though that didn't go anywhere).

 

EDIT: Pascal is sort of like Maxwell refined. The basic stuff is the same but some stuff including the memory controller (for memory compression and to support HBM2 and G5x on some Pascal cards), the display controller (for DP 1.4 support), and a bit of the ROPs (Pascal does tile based rasterization in a bit of a different pattern than Maxwell). The relationship between Pascal and Maxwell, IMO, is more similar to that between Fury X and 290x.

1 hour ago, Terryv said:

Agreed, I've been watching his videos (AdoredTV) for a while. I like the way he does things and his effort to understand why certain things happen.

 

Unfortunately his results don't always match up to popular opinion (witch is why he's getting so much hate ATM).

 

I've yet to see bias in his reviews, always shows good and bad from both sides and uses sound logic to come to conclusions.

 

 

I still think he tilts stuff a bit in some reviews. As well, it's quite obvious he's a bit biased, as evidenced by his masterplan and gameworks videos, though in general his reviews are pretty good, though I make sure to keep my mind open when watching since he is a bit biased IMO.

[Majestic]

16 hours ago, Mothballs said:

They're very good for putting out a theory and looking through other peoples data, they just lack the resources to prove things conclusively themselves. Which means if they say something, someone else is going to have to test it for me to 100% Believe it.

Except that the data shown is entirely his own. Not sure how creating your own data is somehow more credible though. He can completely fabricate results and you'd be none the wiser. This is the first set of data with these parameters, so ofcourse you should withold judgement until more tests roll out under the same conditions. But being the first set of data, there is no way you can flatout claim they're false or inaccurate.

 

The only data you shouldn't really trust is the data that is done under the same parameters and yields completely different results than is the current consensus. Or at the very least scrutinize heavily. 

[xAcid9]

20 hours ago, Humbug said:

in legacy APIs, I.e. Upto directx11 or any version of openGL traditionally Nvidia had lower CPU overhead and better multithreading.

 

There is a claim however that Nvidia GPU architecture is less suitable for modern APIs like Vulkan/DX12 and that compared to AMD, NVidia is attempting to emulate more functions in software. It looks like that is the accusation that adoredtv is trying to expand on. He is saying that this newer part of NVidia's software suite is not yet well optimized or multithreaded.

That make sense. Even PaxwellPascal is kinda crippled in DX12/Vulkan.

[RagnarokDel]

21 hours ago, EminentSun said:

This could simply be due to the fact that amd gpus gets far more out of vulkan and dx12 as the rx 480s saw massive gains with either cpu.

yeah but that's not about the GPU per say. The results with the nvidia card would lead you to believe that Ryzen is bottlenecking the 1070 when it is clearly not the case on AMD cards, the processor is still able to push more. If it wasnt, it would have the same kind of results as it did with the Nvidia card.

17 hours ago, Mothballs said:

Adored means well, But they're one of the ones that unless multiple other sources agree with (Not Joker) I'll take with a grain of salt.

They're very good for putting out a theory and looking through other peoples data, they just lack the resources to prove things conclusively themselves. Which means if they say something, someone else is going to have to test it for me to 100% Believe it.

That said, they have put a ton of foot work into this. Interesting, and worth looking into.

AdoredTV is a "newcomer", he doesnt have the outreach that other youtubers do. The whole reason he used rx 480s in Crossfire is because his contact(s) refused to send him a Fury X lol

[ivan134]

This guy(s) also carried out this test after seeing Adored's test and the same thing is happening in The Division.

 

 

[Carclis]

Looks like German magazine "C't" was able to observe similar behaviour in comparing Titan XP and Fury X performance when paired with the R7 1800x. Look at the 720p benchmarks.

 

[Mihle]

 

[Majestic]

Seen it, AdoredTV's results were valid. But at the moment an outlier. That said, Steve did mention he rushed it which means shortcuts were made.

Like how two Hawaii chips do not make 2 Polaris chips and the Fury X was being fully maxed out whereas the 1070 wasn't. Meaning the CPU results on the FuryX were bottlenecked.

 

Guess we'll need to wait for Vega and more DX12 games to have this verified some more. 

 

I dont know why he said Tomb Raider was poorly optimized though, it is definitely not poorly optimized. Nixxes did a great job.

[David89]

On 31.3.2017 at 6:19 PM, ivan134 said:

This is what has been infuriating me when people say DX 12 is crap so far and they only want Vulkan. Most benchmarks are down with Nvidia's cards which are garbage at low level APIs. Fact is DX 12 has a better track record so far than Vulkan when using AMD GPUs. About 80% DX 12 games work properly and give better performance than DX 11 using and cards. Vulkan is only 50% (everyone seems to forget the Talos principle).

 

This isn't just AMD catching up to Nvidia's DX 11 performance. There 480 beats the 1060 no matter the API the 1060 uses in all DX 12 games except BF1 and quantum break. The 480 is also faster in DX 12 than DX 11 in all these games, including games that had bad DX 12 launches like deus ex MD and the division. Reviewers aren't retesting these games after patches and we have to rely on Reddit and tech forums to get this info.

 

Also, for people who still don't believe that Pascal is only slightly less garbage at low level APIs than Maxwell, there you go. Software emulation will never match up having proper hardware.

DX12 IS crap - but not really from a performance standpoint. You will be completely locked in to Windows 10. If you want to use DX12, you must use Windows 10. Vulkan may have only 50% - so be it, i'll take 50% and cross platform compatibility any day for those (rationally thinking pretty irrelevant) 20%. For me personally, that's also an instant game over for DX12.

 

Vulkan is also OpenSource. Nvidias Linux drivers are IMHO absolute garbage (although i grudgingly admit, they have a lot better OpenGL performance), Nvidia doesn't really care about OpenSource and will lock you in to their eco system no matter what - GameWorks, G-Sync and so on.

 

About AdordedTV: Yes, he's a bit biased, he's admitted to being an AMD Fan, but he does darn good work and still the Intel CPUs won. He could have manipulated those last few percent in favour of AMD, but he didn't. Also, it has been proven by others, that his numbers are "the real deal", since they are pretty much the same.

 

 

[ivan134]

36 minutes ago, David89 said:

DX12 IS crap - but not really from a performance standpoint. You will be completely locked in to Windows 10. If you want to use DX12, you must use Windows 10. Vulkan may have only 50% - so be it, i'll take 50% and cross platform compatibility any day for those (rationally thinking pretty irrelevant) 20%. For me personally, that's also an instant game over for DX12.

 

Vulkan is also OpenSource. Nvidias Linux drivers are IMHO absolute garbage (although i grudgingly admit, they have a lot better OpenGL performance), Nvidia doesn't really care about OpenSource and will lock you in to their eco system no matter what - GameWorks, G-Sync and so on.

 

About AdordedTV: Yes, he's a bit biased, he's admitted to being an AMD Fan, but he does darn good work and still the Intel CPUs won. He could have manipulated those last few percent in favour of AMD, but he didn't. Also, it has been proven by others, that his numbers are "the real deal", since they are pretty much the same.

 

 

Cool story

[Majestic]

3 minutes ago, ivan134 said:

Cool story

Indeed. 

[PopReference]

1 hour ago, Mihle said:

 

Watched this and made me wonder if SLI would get similar gains that crossfire does going from DX11 to DX12.

 

And ROTR is not a consistent benchmark

[Nena Trinity]

Hardware Unboxed tested this to but he do not think Nvidia are to blame tough... o3o

[roylapoutre]

Am I the only one thinking the comparison between Crossfire GPUs and single GPU (ie 2x480 vs 1x1070) is kinda wrong to look at CPU performance ? I understand the reasoning behind the video, but multi-GPU setups are supposed to dramatically increase CPU overhead, no matter the API/GPU/CPU, thus creating a CPU bottleneck, making this test less than ideal. I might be completely wrong though, but it seems really really "un-scientific" to make this kind of test with multi-GPU configuration, especially if you compare the results to a single GPU config. Not saying the conclusions are wrong, but it feels weird 

[xAcid9]

2 hours ago, Mihle said:

-snip-

boy that Fury-X working hard in DX12, GPU usage almost 100% all the times but it probably render the scene inefficiently with it inferior geometry engine. Polaris improved that quite significantly and i hope Vega will improve that even more.

 

 

[Majestic]

1 hour ago, roylapoutre said:

Am I the only one thinking the comparison between Crossfire GPUs and single GPU (ie 2x480 vs 1x1070) is kinda wrong to look at CPU performance ? I understand the reasoning behind the video, but multi-GPU setups are supposed to dramatically increase CPU overhead, no matter the API/GPU/CPU, thus creating a CPU bottleneck, making this test less than ideal. I might be completely wrong though, but it seems really really "un-scientific" to make this kind of test with multi-GPU configuration, especially if you compare the results to a single GPU config. Not saying the conclusions are wrong, but it feels weird 

Yes, you're the only one who doesn't get it. Not sure what's wrong with comparing them both under the same conditions. CPU overhead should only affect absolute numbers, not relative numbers.

[SpaceGhostC2C]

2 hours ago, Nena360 said:

Hardware Unboxed tested this to but he do not think Nvidia are to blame tough... o3o

Unfortunately, he only made a few useful tests (295x2 vs Titan X).

The Fury / 1070 tests were kind of useless, since he kept settings constant, as if it was a GPU review. As a consequence, there were circumstances in which one system was CPU-bound, while the other was GPU-bound. Overall, I think he missed the main point, that is, that the relative changes between DX11 and DX12 depended on the GPU used in CPU-bound scenarios, which may imply different settings for different GPUs. By chance, the 295 vs Tittan comparison suffered less from this problem, but he didn't use these cards with as many games.

[Mothballs]

It's looking more like Ryzen works better with Crossfire, or possibly a glitch? We're not sure yet, there's a ton of inconsistency.

[Majestic]

Just now, Mothballs said:

It's looking more like Ryzen works better with Crossfire, or possibly a glitch? We're not sure yet, there's a ton of inconsistency.

Was already posted mate, I'll add it to the OP i guess.

[Mothballs]

Just now, Majestic said:

Was already posted mate, I'll add it to the OP i guess.

Ah, sorry, Missed it by going too fast. Whoops.

[roylapoutre]

16 hours ago, Majestic said:

Yes, you're the only one who doesn't get it. Not sure what's wrong with comparing them both under the same conditions. CPU overhead should only affect absolute numbers, not relative numbers.

I know we're not looking at relative numbers here, and that was not the point I tried to make. My point was that CPU workload is pretty different from single-GPU to dual-GPU setups. And it seems possible that with a dual-GPU setup, one would see limitations on both Intel and AMD CPUs, closing the performance gap between them when using 2xRX480.

[LAwLz]

Oh look, another bollocks video from AdoredTV where he posts outrageous claims.

AMD not performing as expected? Gotta be Nvidia crippling AMD! Surely it can't be that AMD might simply not be as good as the hype train, which AdoredTV certainly played a big role in, got out of hand and set their expectations too high. Nope, gotta be a conspiracy theory that Nvidia is behind it all!

 

One gaming that is known to perform poorly shows a bit of inconsistent results, actually showing Nvidia cards performing worse than they should? Yep, I am sure this is Nvidia crippling their own cards, making AMD cards look better, just so that Intel can get more sales!

Because crippling your own performance, so that both Intel CPUs and AMD GPUs appear to be better is a great strategy...

 

I wonder how much time AdoredTV spent on finding this one game where this happens, and how many games he had to ignore because it didn't align with the theory he wanted to push.

[done12many2]

3 hours ago, LAwLz said:

Oh look, another bollocks video from AdoredTV where he posts outrageous claims.

AMD not performing as expected? Gotta be Nvidia crippling AMD! Surely it can't be that AMD might simply not be as good as the hype train, which AdoredTV certainly played a big role in, got out of hand and set their expectations too high. Nope, gotta be a conspiracy theory that Nvidia is behind it all!

 

One gaming that is known to perform poorly shows a bit of inconsistent results, actually showing Nvidia cards performing worse than they should? Yep, I am sure this is Nvidia crippling their own cards, making AMD cards look better, just so that Intel can get more sales!

Because crippling your own performance, so that both Intel CPUs and AMD GPUs appear to be better is a great strategy...

 

I wonder how much time AdoredTV spent on finding this one game where this happens, and how many games he had to ignore because it didn't align with the theory he wanted to push.

 

I didn't even know who the hell AdoredTV was until seeing this thread, but after looking at his home page, I can understand your post a bit more.  Seems like a very biased reviewer.