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

[Glo.]

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!

It appears you want to spin my words the way you decided they mean.

I have said about Programmable Geometry Pipeline, and Primitive Shaders.

Which of the features is increasing Geometry Throughput of Vega architecture to up to 11 Triangles per clock with 4 Geometry Engines?

We do not know whether it is active in Vega FE drivers, just like DrawStream Binning rasterizer was not active, per applications testing this. We have not seen any analysis of this feature on performance. Yet, you acclaimed Vega, and this feature is a failure. Supposedly I have had zero reputation to claim anything. What about in this context about your reputation?

Can you stop spreading FUD? I want to know and understand the hardware. All you are doing is spreading FUD about it.

And what about numbers you were referring to, Cat Merc has asked you this, and you still not answered this question.

 

[Muhammed]

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

It's not even active in AMD marketing slides, or AMD leaked benches? AMD was vocal about DSBR not being active, but they never spoke about primitive shaders being not active.

We know nothing about Primitive Shaders yet. We do not know whether it is active in Vega FE drivers, just like DrawStream Binning rasterizer was not active, per applications testing this. We have not seen any analysis of this feature on performance.

You are the one who claimed 70% more performance through Primitive Shaders, not me, even though you knew nothing about it. Now that official conflicting information are online, you claim to try to understand the hardware. Typical!

I have said about Programmable Geometry Pipeline, and Primitive Shaders.

They are one and the same.

 

[PeterScott]

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

We have RX leaks, we have AMDs own claims about RX. Unless you want to argue that it isn't active in RX either.

 

[Glo.]

It's not even active in AMD marketing slides, or AMD leaked benches? AMD was vocal about DSBR not being active, but they never spoke about primitive shaders being not active.

That is the reason why its stupid to say that it is active when you don't know that.

You are the one who claimed 70% more performance through Primitive Shaders, not me, even though you knew nothing about it. Now that official conflicting information are online, you claim to try to understand the hardware. Typical!

And I stand by that. Vega, with higher throughput of cores, and higher Geometry Throughput than Gp102, can be 30% faster than GTX 1080 Ti.

They are one and the same.

Proofs? Links? Or you just made stuff up, like about Primitive Shaders being active in Vega FE drivers?

Funniest part, I know the answer for this. But I want YOU to show everybody how little you know about the architecture you argue, to prove your bias.

Let me give you a hint. If Primitive Shaders are driver feature, why does Programmable Geometry Pipeline has name: "Programmable"?

Second question. If you know so much about those features, with what situation will Primitive Shaders help, and with what situation will help Programmable Geometry Pipeline and two times higher throughput? What affect will those features have on specific framerates? Minumums or Maximums?


Actually I cannot even bother. Primitive Shader will help with Minimum Framerates, and potentially averages, but peak framerates will always be defined by Geometry Throughput(Peak). So Maximum Framerates can only be increased by higher Geometry Throughput: Programmable Geometry Pipeline.

And BTW. You still have not answered Cat Merc question.

 

[Kenmitch]

Not really into the inner workings of gpu manufacturing and design, but in the end as a uneducated consumer I can see the room for improvement in things that are functional but not optimized. In the end there can be both functional and not fully functional scenarios.

AMD is somewhat doomed when the majority of consumers only want them for leverage on pricing for their brands of choice. Seems like a lot of users want AMD to be competitive with both Intel and Nvidia for nothing more than better pricing on their products.

 

[tential]

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.

Ok, surely I'm not the only person surprised about this?
After all the posts you've made about Vega, defending it, etc. you're not even going to wait for benchmarks? All the posts you've said telling people to wait for RX Vega.....

This post just contradicts what you've been saying down to the core...

Edit:
Either way, welcome to the club.

 

[Glo.]

Ok, surely I'm not the only person surprised about this?
After all the posts you've made about Vega, defending it, etc. you're not even going to wait for benchmarks? All the posts you've said telling people to wait for RX Vega.....

This post just contradicts what you've been saying down to the core...

Edit:
Either way, welcome to the club.

Vega architecture is not Only Vega 56 and 64. It is also Raven Ridge, and Vega 11, and Vega 12 GPUs.

I will most likely not get Raven Ridge APU with 4C/8T, 12-16 CUs and HBM2 on top of it. This is the range of needs I have, in terms of performance and affordability.

4C Coffee Lake+ GTX 2050 Ti - that is in my range of affordability.

Me going Volta has nothing to do with its performance, but actually lack of hardware I could buy. I am efficiency fan. If 4C/8T APU with 12-16 CU's and HBM2 would offer me GTX 1050 Ti level of performance - I would gladly wait for it.

 

[dogen1]

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

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

Raja said this on reddit something like a month ago.

 

[Konan]

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

https://www.facebook.com/TweakTown/posts/10156534650604942

TweakTown As I said in my article last week, I doubted the 100MH/s article and didn't believe it. All I can say is I was right. Nowhere near 100MH/s on Vega 64 Liquid Cooled Edition -Anthony

 

[Karnak]

Vega with almost 2GHz? Although I doubt that tbh.

http://www.3dmark.com/spy/2205628

 

[ozzy702]

https://www.facebook.com/TweakTown/posts/10156534650604942

I wouldn't expect anywhere near 100MH/s anyway, that's just pure nonsense. Honestly, AMD needs at least 50MH/s for the cards to be worth it when you factor in price and power consumption. If AMD hit's 60MH/s I'll be very surprised and I'll fork over some cash I have sitting around from selling a bunch of 480s.

 

[Head1985]

It looks like vega sucks at mining.Oh yes.Best news in this year.
http://www.overclock.net/t/1636235/...rx-vega-64-nowhere-near-100mh-s#post_26280222
from Tweaktown

You will not buy RX Vega for mining

 

[Muhammed]

New It looks like vega sucks at mining.Oh yes.Best news in this year.

Would you rather buy it (Vega 64) if it scores the same as 1070 @1080p in some DX11 games?

 

[tential]

More confusion from AMD.
https://gaming.radeon.com/en-us/rxvega/
https://www.reddit.com/r/Amd/comments/6t89ie/radeon_pack_confusion_monitor_discount_and/

I've been following the Radeon pack website because I'm really interested in buying the full pack (i.e. 1700X + MoBo + Games + FreeSync Monitor). A few days ago the website said

Plus, get $200 Off if you buy a Samsung 34″ Ultra WQHD Curved Monitor with Quantum Dot and Radeon™ FreeSync technology (C34F791).
Source: https://web.archive.org/web/20170806193109/https://gaming.radeon.com/en-us/rxvega/

Now it says

PLUS, GET UP TO $200 USD DISCOUNT FROM PARTICIPATING RETAILERS ON SELECT RADEON™ FREESYNC CAPABLE MONITORS.* Exact monitor may not be available in your area
Source: https://gaming.radeon.com/en-us/rxvega/ at the time of this post

Maybe they realised that only offering a discount on one monitor (that has its issues) won't sell them any packs. I mean the release is very soon, so negotiating last minute rebates might be possible but I didn't think they would do that.

That's not the only confusing thing though. To my knowledge there isn't any confirmed information out there about which discounts are available where. In the community forum AMD employee ray_m says:

Unfortunately, at this point in time, we have no information on what promo's may or may not be available in Europe & South America.
Source: https://community.amd.com/thread/218694

What do you think? Are they still negotiating deals for the monitor rebate outside of USA, Canada, Australia, and Singapore?

This is really confusing, I'd like some clarification from AMD.

Summary:
Monitor discount for Samsung monitor on website listed.
Now website edited and monitor discount info is gone.

 

[Malogeek]

More confusion from AMD.
https://gaming.radeon.com/en-us/rxvega/
https://www.reddit.com/r/Amd/comments/6t89ie/radeon_pack_confusion_monitor_discount_and/

So now the monitor discount is for.... well we'll have to wait and see.....
No one knows.

The usual different continent issues that either something fell through or haven't had it finalized yet. USA it's easy but Europe is a lot more restrictive in their rules.

Whatever the problem, if it's sorted out by the time you can actually order them then all good. More confusion isn't good at the moment though.

 

[Head1985]

Would you rather buy it (Vega 64) if it scores the same as 1070 @1080p in some DX11 games?

I am upgrading from GTX1070 2140/9500 to vega 56 or 64.Need to see reviews first with vega 56 OC.
I will test cards in this games in 2k/4k and some like alien isolation in 5k

 

[Muhammed]

I am upgrading from GTX1070 2140/9500 to vega 56 or 64.

You can forget about the Vega 56 completely with this overclock of yours, as for Vega 64, it will depend on the title. And you can quote me later on that.

 

[Head1985]

You can forget about the Vega 56 completely with this overclock of yours, as for Vega 64, it will depend on the title. And you can quote me later on that.

I dont think so..GTX1070 2140/9500 is still like 15% behind stock reference GTX1080 in 99% games.It should be around stock reference vega 56performance.AIB card will be 10% faster and maybe another 10% from OC thats 20%.
If vega 56 memory can be oc same way like on vega64 there should be pretty small gap vs vega 64 at same clock.Probably 5-8%.

 

[Konan]

I am upgrading from GTX1070 2140/9500 to vega 56 or 64.

A GTX 1070 to a Vega 56 is more of a lateral move not an upgrade. I wouldn't bother at all.

 

[Hi-Fi Man]

I am upgrading from GTX1070 2140/9500 to vega 56 or 64.Need to see reviews first with vega 56 OC.
I will test cards in this games in 2k/4k and some like alien isolation in 5k

You mean you're sidegrading, right? What's the point?

 

[Muhammed]

It should be around stock reference vega 56performance.AIB card will be 10% faster and maybe another 10% from OC thats 20%.

Ha! Come Monday and you will find out you are severely mistaken. Basically your only option from AMD is Vega 64, and that is also a questionable upgrade given the performance.

 

[Head1985]

Vega 64 and 56 are pretty much same cards you know.If you can OC HBM2 on 56 to or above vega 64level there will be almost zero difference in performance.thats why i waiting for vega 56 OC results

4096/3584=14% more SP

On nv side
2560/1920=33.3% more Sp and 25% more bandwidth

 

[Hi-Fi Man]

Vega 64 and 56 are pretty much same cards you know.If you can OC HBM2 on 56 to or above vega 64level there will be almost zero difference in performance.thats why i waiting for vega 56 OC results

4096/3584=14% more SP

On nv side
2560/1920=33.3% more Sp and 25% more bandwidth

So what's the point?

 

[Maverick177]

So can we all in unision agree Vega is DOA ?

 

[Qwertilot]

99% so for gaming, yes I think. Considerably too late and power consuming for what it does.

Seems to be rather more chance of them finding a useful niche or two in compute etc somewhere.