Let's address a certain AdoredTV video on Ryzen and Tomb Raider...

[othertomperson]

So I'm sure you've seen this video doing the rounds, particularly among AMD fanboys looking for any and every possible reason that someone else who isn't AMD is responsible for Ryzen not being the fantastic gaming CPU they were promised -- and bonus points if you can claim that it's somehow Nvidia's doing (let's choose to ignore the fact that the only Nvidia CPU you're likely to be gaming on is inside the Nintendo Switch)

 

As a preface I quite like Jim, I like the testing that he does and he's always a fun and interesting person to have a discussion with. However, his conclusion in this oftly cited gotcha video is just flat out wrong. This isn't the first time I've actually brought up on this forum how astonishingly multithreaded this game is in Directx 12, so to see the Internet plastered with this video claiming that this game is limited to mainly one core on Nvidia drivers is a little odd.

 

I have, however, never looked at this game at 1080p, so I decided to have a little look at that, both with SLI on, and off.

 

Just a warning now, apologies for "screenshots" being literal photographs of my monitor. GeForce Experience is borked for me atm and I cba to fix it, and steam overlay doesn't work with RotTR.

 

 

So moving on, I tested Geothermal valley on my ageing 6-core both with SLI on (2xGTX 1080) and SLI off, in the same place as both AdoredTV and DigitalFoundry. Gaming with only one GTX 1080 even at 1080p it is difficult to say whether I am CPU or GPU bound. They are both running pretty close to their limit, but the 1080 undoubtedly has nothing left to give, while the 3960X has a little. This is actually why I still don't consider 1080p a "low resolution" when it comes to benchmarking CPUs. Even at 1080p, with a Sandy Bridge CPU an overclocked GTX 1080 is still at 98% utilisation.

 

 

The picture shows at that point in Geothermal Valley I am getting 102.3 FPS, while at 98% GPU utilisation. It gets interesting when you look at CPU utilisation. Not too shabby, 75% utilisation across all 12 logical threads. I find it hard to look at this CPU utilisation and conclude that the Nvidia driver is loading one core particularly more than the others as AdoredTV claimed.

 

Nevertheless, I had a look at what the numbers look like with Multi GPU enabled.

 

 

In the same position GPU utilisation has plummeted all the way down to 56%, either this game has crap multi GPU support or I am truly CPU limited here. FPS has increased to 110.5 fps, which coupled with the complete lack of stuttering implies the latter to me. So let's have a look at what the CPU looks like here.

 

 

See when I first saw utilisation like this in a Dx12 game my reaction was "holy shit, at last". 98% utilisation across 12 logical cores. Evidently, I am indeed CPU limited here. Whether I was GPU bottlenecked before, or whether the introduction of SLI has also resulted in additional CPU load, I can't say, but it is clear that my CPU has very little left to give here, while the 1080s do.

 

What is clear, however, is that the claim being circulated that Nvidia's Dx12 driver for Rise of the Tomb Raider cannot split the first core's workload across the CPU, and therefore this is the reason for Ryzen's lacklustre performance is simply not true. This game loves cores. This game should love Ryzen much more than the 7700K, regardless of whose GPU it is paired with.

 

[Jito463]

I do agree that ATV's testing methodology left something to be desired.  To be conclusive, he should have at least tested the 2 x RX 480's on an Intel system (assuming he had one handy).  However, your testing methodology is equally flawed, as you're testing an SLI setup but only on Intel.  While - as I understood it - his claim was that the alleged bug primarily affected Ryzen.

 

In order to be thorough, someone would need to test single GPU AMD/Nvidia cards on both an AMD and an Intel system on both DX11 and DX12.  Then repeat the tests with both a Crossfire setup and then an SLI one, then aggregate the results.

[othertomperson]

11 minutes ago, Jito463 said:

I do agree that ATV's testing methodology left something to be desired.  To be conclusive, he should have at least tested the 2 x RX 480's on an Intel system (assuming he had one handy).  However, your testing methodology is equally flawed, as you're testing an SLI setup but only on Intel.  While - as I understood it - his claim was that the alleged bug primarily affected Ryzen.

 

In order to be thorough, someone would need to test single GPU AMD/Nvidia cards on both an AMD and an Intel system on both DX11 and DX12.  Then repeat the tests with both a Crossfire setup and then an SLI one, then aggregate the results.

His claim wasn't Ryzen specific. He claimed that Nvidia's driver wasn't multithreaded enough to benefit an 8-core over a 4-core, and that this is why the 7700K seemed to do better.  Clearly something isn't working properly with Ryzen in this game. I have just dismissed it being this.

[Enderman]

I have no idea why people even take this guy seriously...

When he started off he literally had no hardware and did no tests, just talked about other people's tests and comparing them even though they were ran in completely different situations.

Now his tests show the opposite of most other real tests done by reputable reviewers, just to please AMD kids.

[Jito463]

7 minutes ago, othertomperson said:

His claim wasn't Ryzen specific. He claimed that Nvidia's driver wasn't multithreaded enough to benefit an 8-core over a 4-core, and that this is why the 7700K seemed to do better. However else it may interact with Ryzen specifically, I think that particular claim that's going around can be dismissed.

You may be correct, I'll have to watch the video again.  I may have misinterpreted what he was saying.

[LabRat]

how come there's no "thumbs down" button on this site?

[othertomperson]

Just now, LabRat said:

how come there's no "thumbs down" button on this site?

Because expressing your dissatisfaction with words and valid reasoning is more productive.

[LabRat]

you're so smug. but you should look a little harder. there is valid reasoning to the tests being done and results being made that don't coincide with other reviewers. maybe that fanboyism should be set aside before calling out other people.

and maybe a 1070 or a 1080 or a 1080ti don't fit in the same benchmark with a 480 either.

[othertomperson]

22 minutes ago, LabRat said:

you're so smug. but you should look a little harder. there is valid reasoning to the tests being done and results being made that don't coincide with other reviewers. maybe that fanboyism should be set aside before calling out other people.

and maybe a 1070 or a 1080 or a 1080ti don't fit in the same benchmark with a 480 either.

And if "valid reasoning" escapes you, there's always name-calling to fall back on

 

Not sure how taking screenshots of his claims being demonstrably falsifiable makes me a fanboy, but sure whatever.

 

And I agree that the 1070, 1080, 1080 Ti shouldn't be in a CPU benchmark with a 480. I agree that the 480 shouldn't be used to benchmark CPUs at all (apart from maybe at 800x600 resolution). And I certainly don't think that SLI and Crossfire should be introduced into CPU benchmarks. However, AdoredTV chose to do all of these. If you are going to look at multi-GPU on AMD and assert that there's a driver issue on Nvidia you should at least also look at multi-GPU on Nvidia to be sure that they don't make up the same ground by introducing this additional variable. This is what I did. Not that it mattered, his point was already disproven in the single GPU result I demonstrated.

 

If I'm a fanboy to anything it's to the scientific method, and that's all about "calling out other people".

 

I want to reiterate that I don't have a problem with Jim or his videos. He came to a conclusion in this video that I disagree with and I think is demonstrably incorrect. I think the degree to which videos like this are passed around uncritically with forums and comments sections full of "Nvidia is sabotaging Ryzen" type comments is highly unhelpful. It's obvious that Nvidia's Dx 12 driver is not limited to 1 or 4 cores as some have claimed, so evidence against the claims made here should be presented.

[LabRat]

what name calling is that?

somebody asked this guy to do a test. he did it. it falls out of the realm of "business as usual". people don't like it so they gripe.

maybe Linus can do the tests and see if there is an agreement..?? I'm not holding my breath.

after the new cards come ( not the re-brands ), I'd like to see this test run again.

[othertomperson]

1 hour ago, LabRat said:

what name calling is that?

somebody asked this guy to do a test. he did it. it falls out of the realm of "business as usual". people don't like it so they gripe.

maybe Linus can do the tests and see if there is an agreement..?? I'm not holding my breath.

after the new cards come ( not the re-brands ), I'd like to see this test run again.

Linus (or more precisely, Luke) has been benchmarking CPUs at 4K, and benchmarking GPUs overclocked as far as his sample will allow (which means his results are not indicative of the product in general) for some time. I don't know if LTT still does this, I unsubbed some time ago because their methodology made their results meaningless.

 

And the part where you called me smug and a fanboy for having the audacity to present data which contradicted the new AMD-follower orthodoxy.

[LabRat]

i won't start a pissing match for anybody's benefit. lets call a truce.