AMD RyZen CPU Architecture for 2017

[DavidGraham]

Ryzen is an excellent CPU which just seems to be bottlenecked by both drivers as well as intel optimised games.

These don't elevate themselves to a the status of even preliminary results, testing conditions are loose, with variable factors and outright inaccurate conclusions.

Ryzen is an excellent CPU which just seems to be bottlenecked by both drivers

Except that AMD never implied of such things, AMD kept recommending testers to bench 1440p and 4K, turn off SMT, and test with better RAM. But they NEVER implied any issue related to NV's DX12, they never recommended testing CrossFire instead of a single Titan X/1080Ti/1080. In fact, they usually use one of the three GPUs each time they bench something. You would think that they would be the ones preaching about DX12 driver. But nope they didn't. Some Youtube hero with lackluster technical skills did, and people are inclined to believe him! I suggest waiting for a more reputable source to get to the bottom of this.

 

[ProspectorPete]

As I said, I understand your position, honestly.

Remember when reputable sources were publishing articles about RX480 being broken due to PCI-E power draw, which could never be fixed or something like that? I would not really hold my breath for them to report on this

 

[DavidGraham]

Remember when reputable sources were publishing articles about

HardwareUnboxed just did:

FuryX vs 1070 on DX12, Nothing out of the ordinary, the 1070 easily beat the FuryX especially in Ashes of The singularity. And decimates it in Tomb Raider (23% faster). However CF scaling on Ryzen is particularly massive in Tomb Raider, and that's what that AdoredTV guy noticed, he was testing CF scaling. Knowledgeable and educated people have been saying the same thing from the start.

 

[Malo]

And he basically found the same issue at this particular part and acknowledges there's an issue in the comments as well?

Seems overall to be more of an issue with Nvidia driver under DX12 and scaling beyond 4 cores than Ryzen in particular. Certainly a curious find.

 

[ProspectorPete]

And he basically found the same issue at this particular part and acknowledges there's an issue in the comments as well?

Seems overall to be more of an issue with Nvidia driver under DX12 and scaling beyond 4 cores than Ryzen in particular. Certainly a curious find.

This.

 

[DavidGraham]

Seems overall to be more of an issue with Nvidia driver under DX12 and scaling beyond 4 cores than Ryzen in particular. Certainly a curious find.

This is probably a CPU limitation (this area is heavily single threaded, Core1 is hammered the most). However it's not exclusive to Ryzen alone, it probably happens on Intel CPUs as well. Intel having a better single threaded performance handles it better. The testing revealed that TitanXP is barely faster than 1070 in that areas. Also the FuryX/1070 are barely faster than a single RX 480 in the same area as well.

Still the takeaway here is that Ryzen is NOT limited by the DX12 path on NV GPUs as evident here: (the 1070/FuryX have the same relative fps with 1800X/7700K)

 

[ProspectorPete]

This is probably a CPU limitation (this area is heavily single threaded, Core1 is hammered the most).

If it is a CPU limitation then the dual 480s would not outperform anything

 

[Jawed]

A compiler bug in Ashes of the Singularity has been fixed by compiling with a different compiler, so we can cross off yet another gigantic difference:

https://www.reddit.com/r/Amd/comments/62vp4k/patch_boosts_ryzens_perf_by_31_on_aots_by_fixing/

And, on top of that, AotS in gameplay on 4 cores or less, cuts out graphics features to make the game play faster.

https://www.reddit.com/r/Amd/comments/62mf0m/aots_performance_riddle_solved_aots_turns/

So that's yet another gigantic difference explained.

 

[DavidGraham]

If it is a CPU limitation then the dual 480s would not outperform anything

And it is NOT NV's DX12 as well. Something else is at works here. Probably the mGPU implementation itself in DX12.

 

[ProspectorPete]

Well, if it's a bug that crossfire setups have too many frames per second, then I wish more games had bugs like that

 

[Jawed]

HardwareUnboxed just did:

Nothing out of the ordinary

When you link something that directly contradicts what you're saying, I can only conclude that you didn't watch the video properly.

 

[DavidGraham]

When you link something that directly contradicts what you're saying, I can only conclude that you didn't watch the video properly.

Actually, that video reaffirmed exactly what I said in previous posts, CF scaling is handled differently in this game compared to single GPU. In single GPU mode, results are straightforward. Nothing out of the ordinary with the 1070/FuryX fps deltas. Only when CF is introduced do things get out of wack in this title. Among them is better scaling on Ryzen 1800X than i7 7700K.

 

[xEx]

The difference in MoBos is still huge, new BIOS show huge performance upgrade(avg fps) but for some reason(saw more than one source) the min FPS drops in some cases(?)

TR and other benchs comparative between BIOS:

 

[ProspectorPete]

Should AMD have waited 2 months? Or are these findings only possible through user test scenarios?

 

[Rodéric]

Can't they write articles instead of posting videos ?!
First I don't care about their faces and second I have little time to waste listening to someone...

 

[ProspectorPete]

Actually, that video reaffirmed exactly what I said in previous posts, CF scaling is handled differently in this game compared to single GPU. In single GPU mode, results are straightforward. Nothing out of the ordinary with the 1070/FuryX fps deltas. Only when CF is introduced do things get out of wack in this title. Among them is better scaling on Ryzen 1800X than i7 7700K.

I told you this was going to happen:

 

[ProspectorPete]

On Intel a 1060 would destroy a 470.

 

[3dilettante]

Can't they write articles instead of posting videos ?!
First I don't care about their faces and second I have little time to waste listening to someone...

I've debated bringing this up as the discussion has drifted to video embeds. I haven't weighed in much since that's become the dominant source. I don't like to ramble over data I haven't evaluated, but I haven't been convinced that there's enough worth investing the time.

An article with text and graphs can provide data that can be assimilated by the reader in seconds, and can be searched, quoted, and cross-referenced readily. The time to content ratio of a 10-20 minute video is such an inversion of time and effort needed on the part of the reader/viewer relative to the information gained, or what can be done with it.

I would normally expect editing a video would take more time, otherwise I'd almost feel like this was economizing the author's effort (is ad revenue easier/better?) at the expense of orders of magnitude more time on the part of the viewer.
There are certain items that can be done better with video or context if an interview, but otherwise really categorizing how much time it takes to express something that counts as Signal and the time it takes to get to those points shows how much dead air counts to the Noise in SNR.

If there's a digest or other static format that can condense all this down, it would really be great versus trying to scan through footage.

 

[xEx]

AMD new what they were doing, they new about Mobos being at the very edge of stability with the ram and new how much it would hurt performance they new about every single bottleneck and which games surfer more and sure they had to know about their vgas being better with ryzen than Nvidias.

If they didn't now about this issues being the one that design, test, and validate their cpu I don't know what they were doing then.

Why they chose to launch it knowing how much performance they were leaving under the table? Ask Lisa su maybe they though reviewers would be more forgiving and would interpret the data differently (most of them did, some didn't) either way would be interesting in a year from now to see a comparative review of ryzen with a mature platform compare to the launch reviews.

Enviado desde mi HTC One mediante Tapatalk

 

[xEx]

I told you this was going to happen:

This was one of the best video explaining why and how there are differences and that it doesn't mean one is better than the other just being utilized better in some situations

Enviado desde mi HTC One mediante Tapatalk