Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[Arachnotronic]

There are options in Blender to run without AVX2 for a more level playing field.

Why would you "level" it in that way? Both CPUs can run AVX2 code and the Intel CPUs are better at it. If you're going to use Blender, you're not going to NOT use AVX2 if you have an Intel CPU because it's "unfair" to the AMD CPU, are you?

 

[tamz_msc]

Why would you "level" it in that way? Both CPUs can run AVX2 code and the Intel CPUs are better at it. If you're going to use Blender, you're not going to NOT use AVX2 if you have an Intel CPU because it's "unfair" to the AMD CPU, are you?

The same reason why you would disable Async compute in a DX12 benchmark while testing Polaris and Pascal.

 

[Arachnotronic]

The same reason why you would disable Async compute in a DX12 benchmark while testing Polaris and Pascal.

If the benchmark is written to support Async compute, then why would you disable it?

 

[tamz_msc]

If the benchmark is written to support Async compute, then why would you disable it?

So that one group of fanboys don't scream bias.

Trolling is not allowed
Markfw
Anandtech Moderator

 

[CentroX]

When i get ny 1800X, do i need to reinstall win10? I rather not. But i know there can be issues switching brand and chipset.

 

[french toast]

So that one group of fanboys don't scream bias.

If it is a productivity software or an actual game then the best api/feature should be switched on.

 

[lolfail9001]

The same reason why you would disable Async compute in a DX12 benchmark while testing Polaris and Pascal.

Why would you? It's not like it is any performance impact worth caring about. More like performance penalty in Doom, for example.

So that one group of fanboys don't scream bias.

AMD fanboys, i guess.

Point stands: you use settings that produce best results if you want to compare what CPU does what software better.

Stop with the fanboys comments
Markfw
Anandtech Moderator

 

[tamz_msc]

If it is a productivity software or an actual game then the best api/feature should be switched on.

Yes, but how you report those results is also important. In case of Async compute, you should point out the differences between results when it is turned on vs the results when turned off. Same for any other benchmark.

 

[poohbear]

i dont get it, the 1800x and the 1700 are only 300mhz apart (all other specs seem the same), yet are $150 apart??? I'm guessing the 1800x is unlocked and that's why the price premium is there?

 

[Magic Hate Ball]

i dont get it, the 1800x and the 1700 are only 300mhz apart (all other specs seem the same), yet are $150 apart??? I'm guessing the 1800x is unlocked and that's why the price premium is there?

Nope, all are unlocked. There is speculation you're paying for a binned part however (that will overclock better).

 

[tamz_msc]

AMD fanboys, i guess.

Actually it's NVIDIA fanboys in case of DOOM Vulkan, but I won't digress any more.

Threadcrapping, trolling and flamebaiting is not allowed.
Markfw
Anandtech Moderator

 

[french toast]

Yes, but how you report those results is also important. In case of Async compute, you should point out the differences between results when it is turned on vs the results when turned off. Same for any other benchmark.

Use settings that provide best case performance for a uarch, as long as quality is the same.
So avx2/HSA/GPUcompute/opencl/cuda/dx12/vulkan/async compute/primitive shaders/intinsics etc etc.
Use the best settings that extract the best performance whilst giving equal quality for each cpu/gpu/APU.
This works both ways, if nvidia gets better performance with dx11 and async turned off, use those settings for nvidia vs what ever api add ons AMD uarch supports.

Only caveat is gameworks/gaming evolved quality enhancements such as hair works/tress fx and similar.

 

[tamz_msc]

Use settings that provide best case performance for a uarch, as long as quality is the same.
So avx2/HSA/GPUcompute/opencl/cuda/dx12/vulkan/async compute/primitive shaders/intinsics etc etc.
Use the best settings that extract the best performance whilst giving equal quality for each cpu/gpu/APU.
This works both ways, if nvidia gets better performance with dx11 and async turned off, use those settings for nvidia vs what ever api add ons AMD uarch supports.

Only caveat is gameworks/gaming evolved quality enhancements such as hair works/tress fx and similar.

I agree with everything that you said, but if there are features that favor one architecture over the another, point out how much that architecture gains from said feature. This helps the reader to decide what features he wants to have in the short term vs the long term. That way you can avoid the negative perceptions like 'oh DX12 is useless because it reduces performance with my NVIDIA card'. I hope you get my point.

 

[lolfail9001]

Actually it's NVIDIA fanboys in case of DOOM Vulkan, but I won't digress any more.

DOOM Vulkan loses performance from Async compute, last time i checked, so you really should not digress anymore.

This helps the reader to decide what features he wants to have in the short term vs the long term.

Except that these features are technical and obscure enough in terms of their impact that telling what features do what will hardly help.

 

[PotatoWithEarsOnSide]

'oh DX12 is useless because it reduces performance with my NVIDIA card'.

You mean...'Nvidia are so crap that they can't get their cards to properly use DX12'.

Threadcrapping, trolling and flamebaiting is not allowed.
And this is not the video forum
Markfw
Anandtech Moderator

 

[Glo.]

Lets keep out the Brand Fanboyism from technical discussion, please.

 

[Arachnotronic]

Lets keep out the Brand Fanboyism from technical discussion, please.

Hear, hear.

 

[tamz_msc]

DOOM Vulkan loses performance from Async compute, last time i checked, so you really should not digress anymore.

Except that these features are technical and obscure enough in terms of their impact that telling what features do what will hardly help.

Have you played DOOM? If you did you would know that there is no separate option to turn Async compute on or off. It is enabled by default when you choose to run it on Vulkan. So you cannot measure its impact if you can't test it in the first place.

I thought some sites, especially this one, are technically oriented when it comes to product reviews. I expect the readerbase, at least part of it, to be technically oriented as well.

 

[IEC]

Nitpick: The 6850K/6900K/6950X CB 15 ST score should be 164-165... that's what you would get with TBM enabled.

Also I am suspicious that AMD swapped out Blender for Cinebench. TBF I don't think it's a big deal since Blender supports GPUs now.

Cinebench R15 is one of the benchmarks recommended by Intel in the 6950X Reviewer's Guide for tech press.

 

[CentroX]

anyone calculated what single thread score should be for an overclocked 1800X to 4.3ghz or so? Can it reach 180?.

 

[PrivateCeralion]

Now, there are 4 updates

https://www.asus.com/us/Motherboards/ROG-CROSSHAIR-VI-HERO/HelpDesk_Download/

 

[french toast]

I agree with everything that you said, but if there are features that favor one architecture over the another, point out how much that architecture gains from said feature. This helps the reader to decide what features he wants to have in the short term vs the long term. That way you can avoid the negative perceptions like 'oh DX12 is useless because it reduces performance with my NVIDIA card'. I hope you get my point.

Agreed, point out the performance improvements each feature gives so consumes know if there is a benefit or a negative hit so they turn on or off.

The way i think of it is this, if you had a shiny new APU and you use blender software for work, you would enable every feature/api extension that gives you the best performance, whether that be AVX2/gpgpu compute/HSA etc.
So when purchasing the APU you are going to buy that product based on the apps you would be using, therefore reviews should show products in that best case scenario that a customer would use, reviewers should enable OR disable settings that give best performance to each competing uarch.
As long as the benchmark is using the same quality settings as what you are measuring is the performance difference at the same quality.

Any fanboi comparisons such as "who does the best async" "best AA" best AVX/HSA etc etc should be done either in a separate article or a subsection of the review that is treated in isolation and is not part of the final conclusion, similar to power consumption.

 

[lolfail9001]

If you did you would know that there is no separate option to turn Async compute on or off. It is enabled by default when you choose to run it on Vulkan. So you cannot measure its impact if you can't test it in the first place.

It was initially only enabled for no AA/TSAA and enabling FXAA actually improved FPS compared to no AA on at least few separate occasions. So, no, Async did penalty the performance, unless you want to claim that running without any AA is harder than running with FXAA in general. That's way off track though. So, i will just note that

Now, there are 4 updates

At this rate there will be a BIOS update a day before review embargo.

anyone calculated what single thread score should be for an overclocked 1800X to 4.3ghz or so? Can it reach 180?.

We can safely assume 162 points is 4Ghz score. 180 points would require ~4.45Ghz. Since that MSI board does not even have such setting, i have to assume, it can't.

 

[Magic Hate Ball]

Now, there are 4 updates

https://www.asus.com/us/Motherboards/ROG-CROSSHAIR-VI-HERO/HelpDesk_Download/

Asrock X370 Taichi has a new BIOS from yesterday on their support page as well: http://www.asrock.com/mb/AMD/X370 Taichi/#BIOS

 

[tamz_msc]

It was initially only enabled for no AA/TSAA and enabling FXAA actually improved FPS compared to no AA on at least few separate occasions. So, no, Async did penalty the performance, unless you want to claim that running without any AA is harder than running with FXAA in general. That's way off track though. So, i will just note that
.

What? So you change two settings that are interlinked but claim that one affects more than the other when there is no way to separate them for independent measurement?

Your claim of FXAA being faster than no AA is also false, at least in OpenGL.

We've indeed digressed way too far on this one. Let's decide to stop here.