AMD Vega (FE and RX) Benchmarks [Updated Aug 10 - RX Vega 64 Unboxing]

[lobz]

Apparent Vega hash rates -

Some bench from vega 64 with nicehash 1.8.1.0

keccak 0.724 GH/s
DaggerHashimoto 31.275 MH/s
DaggerDecred 31.317 M/0.940G H/s
DaggerPascal 30.993 M/0.930G H/s
DaggerSia 30.718 M/0.922G H/s
Decred 2.659 GH/s
CryptoNight 800.000 H/s
Lbry 0.192 Gh/s
Pascal 1.668Gh/s
X11Gost 13.400 Mh/s

Claymore 9.8 ETH + SIA
33MH/s - 1011MH/s

we could have used some perspective
but thanks, nonetheless

 

[Rifter]

Any word from AMD yet that Vega will have all the DRM it needs to enable 4K streaming and UHD Blu-ray?

I would say no, when i contacted them about polaris being able to stream 4k netflix(playready 3 DRM) they pretty much blew it off and from how i read the response pretty much said they have no interest in adding it. I doubt thats changed for vega but who knows for sure i guess we will in a few days.

 

[Malogeek]

I would say no, when i contacted them about polaris being able to stream 4k netflix(playready 3 DRM) they pretty much blew it off and from how i read the response pretty much said they have no interest in adding it. I doubt thats changed for vega but who knows for sure i guess we will in a few days.

If it's not in Raven Ridge then they've officially abandoned the HTPC market. They used to be the goto with their APUs.

 

[nathanddrews]

If it's not in Raven Ridge then they've officially abandoned the HTPC market. They used to be the goto with their APUs.

To be honest, I don't think they can be competitive in the laptop space if they don't support at least the codecs and DRM that Kaby Lake currently does. With AMD trying to push FreeSync 2 (HDR, etc.), why abandon the primary sources of HDR content? I dunno, I hope they deliver us from the 4K wilderness...

 

[MrTeal]

Apparent Vega hash rates -

Some bench from vega 64 with nicehash 1.8.1.0

keccak 0.724 GH/s
DaggerHashimoto 31.275 MH/s
DaggerDecred 31.317 M/0.940G H/s
DaggerPascal 30.993 M/0.930G H/s
DaggerSia 30.718 M/0.922G H/s
Decred 2.659 GH/s
CryptoNight 800.000 H/s
Lbry 0.192 Gh/s
Pascal 1.668Gh/s
X11Gost 13.400 Mh/s

Claymore 9.8 ETH + SIA
33MH/s - 1011MH/s

Well, that's pretty terrible. Unless that pretty much doubles, no one is going to be paying $600 for Vega strictly for mining.

 

[ozzy702]

we could have used some perspective
but thanks, nonetheless

If those numbers are true and AMD doesn't have some secret sauce solution, Vega is mediocre at mining. It's expensive and sucks juice. For reference, my 1070s pull ~ 31mh/s while dual mining and pull ~ 110watts from the wall off a Platinum PSU. I hope for AMD's sake that they have something up their sleeves because if they don't it's game over.

 

[Elixer]

Vega 56 I have no idea if systems being compare to are even the same, I doubt it.
Basically a bit better than reference 1070.

FS 19357, 19320
FSE 9506, 9213
FSU 4851, 4741
TS 6238, 6156

compare to

https://www.reddit.com/r/Amd/comments/6t6avy/vega_56_3dmark_benchmark/

 

[PhonakV30]

Vega 56 I have no idea if systems being compare to are even the same, I doubt it.
Basically a bit better than reference 1070.

compare to

https://www.reddit.com/r/Amd/comments/6t6avy/vega_56_3dmark_benchmark/

ROFL.Told ya Navi won't be competitive to volta.

 

[CatMerc]

ROFL.Told ya Navi won't be competitive to volta.

A bit early to make that call on Navi don't you think?

It's not the first time for either company to have a commanding lead that evaporated in a gen or two.

 

[Samwell]

Slowly i'm loosing my hopes in AMD, they are just ridiculous.

@ryszu Any idea when Vega's primitive shaders are likely to be exposed via a Vulkan extension? A ball park estimate would be good to have

Rys Sommefeldt‏ @ryszu
No plans to expose them programatically today, but we do review the state of play periodically so that might change. No plans rn though

https://twitter.com/ryszu/status/895587041269952512
So they implement primitive shaders as a great feature, put it in every presentation, but have no plans to make it accessible? In DX12 that's not possible, but in Vulkan you could use a extension and do that. Seems more like a feature to show off to console makers as marketing for the next deals.

 

[maddogmcgee]

If it's not in Raven Ridge then they've officially abandoned the HTPC market. They used to be the goto with their APUs.

Yep, the only AMD chips I own these days are media centre APU's.

 

[Glo.]

Slowly i'm loosing my hopes in AMD, they are just ridiculous.


https://twitter.com/ryszu/status/895587041269952512
So they implement primitive shaders as a great feature, put it in every presentation, but have no plans to make it accessible? In DX12 that's not possible, but in Vulkan you could use a extension and do that. Seems more like a feature to show off to console makers as marketing for the next deals.

This pretty much sums up the situation, for AMD. And the reason why I am going Coffee Lake+Volta setup early next year.

Edit. There is continuation on this topic:

https://twitter.com/ryszu/status/896304786307469313
The driver takes care of it for you. Just render as normal.

 

[tviceman]

Vega 56 I have no idea if systems being compare to are even the same, I doubt it.
Basically a bit better than reference 1070.

compare to

https://www.reddit.com/r/Amd/comments/6t6avy/vega_56_3dmark_benchmark/

Pretty much expected this after all the cloak and dagger of RX Vega blind testing. AMD didn't say Vega 64 is the "new king" at $499 because it will only trade blows at DX12. Vega 64 is going to be ~5-10% slower across the board than a GTX 1080, and Vega 56 will be ~5-10% faster than a GTX 1070.

 

[Head1985]

Its not bad.GTX1070 2140/9400 manage 9700GPU points in firestrike extreme(1440p).And thats pretty much max out GTX1070.99% GTX1070 will score less.
So stock throttling vega 56 will have +- same performance as max out GTX1070.

 

[CatMerc]

https://mobile.twitter.com/ryszu/status/896327467987873793

Looks like maybe Primitive Shaders are automatically applied from the driver and don't require developer input?

 

[Samwell]

https://mobile.twitter.com/ryszu/status/896327467987873793

Looks like maybe Primitive Shaders are automatically applied from the driver and don't require developer input?

That's interesting. Always thought it's different, but there's so much miscommunication out there. But this leads to the question, is it active in drivers and the gains are small so far or is it deactivated?

 

[Glo.]

Actually it is pretty confusing. Even my fellow friends from game dev community are... surprised with this information.

About drivers, we should see immediate effect of this on performance, but, typically for AMD hardware it does not appear this way. Drivers quality for GCN5?

 

[CatMerc]

https://mobile.twitter.com/ryszu/status/896344188039680000

Rys confirmed it clearly.

 

[PeterScott]

https://mobile.twitter.com/ryszu/status/896327467987873793

Looks like maybe Primitive Shaders are automatically applied from the driver and don't require developer input?

That really isn't good news either, because then all these benchmarks leaks should then already be benefiting from Primitive Shader support, and they still aren't very good.

 

[Muhammed]

That really isn't good news either, because then all these benchmarks leaks should then already be benefiting from Primitive Shader support, and they still aren't very good.

AMD's own numbers don't even indicate these shaders doing any tangible difference in performance.

 

[CatMerc]

AMD's own numbers don't even indicate these shaders doing any tangible difference in performance.

What numbers are you referencing?

 

[Glo.]

AMD cannot "value" up the performance benefits, because this will rely on situation, and complexity of the scene.

All you can do is say: "at least 50% of geometry culled will go elsewhere", but how big affect on performance it will have will rely on the complexity, and the total amount of triangles culled.

 

[godihatework]

That really isn't good news either, because then all these benchmarks leaks should then already be benefiting from Primitive Shader support, and they still aren't very good.

So much for the magic improvements posters have spent the last five pages flogging...

 

[Muhammed]

AMD cannot "value" up the performance benefits, because this will rely on situation, and complexity of the scene.

But you did, you claimed when developers will implement them Vega will be 30% faster! And now we know they are automatically driver activated, which means Vega has no hope, it's performance is not competitive enough as it is with primitive shaders! Shows what you know about things!

 

[Zstream]

But you did, you claimed when developers will implement them Vega will be 30% faster! And now we know they are automatically driver activated, which means Vega has no hope, it's performance is not competitive enough as it is with primitive shaders! Shows what you know about things!

I'm not sure we can state that it's active in the FE edition, until that point, it's pointless arguing.