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

[tential]

https://www.reddit.com/r/Amd/comments/6mfcpp/aibs_get_the_final_source_bios_for_radeon_rx_vega/
AIBs get the final source BIOS for Radeon RX Vega on August 2
source:https://www.3dcenter.org/news/der-release-fahrplan-der-herstellerdesigns-zur-radeon-rx-vega
Looks like amd is hitting a time crunch. From the info in the link doesn't look like amd will have aib cards until September.

 

[guskline]

This "bantering" back and forth about a gpu that hasn't even launched yet remind's me of the infamous Abbott and Costello routine:
https://www.youtube.com/watch?v=kTcRRaXV-fg

 

[Snarf Snarf]

This would make no sense at all that they would vastly gimp their $1000+ card just to spring a August surprise.
AMD can't be THAT incompetent to have pretty much all major improvements "disabled".

It also makes no sense at all that AMD hasn't released a new branch of the drivers for FE owners, which surely must be better than the launch drivers in some ways, unless there isn't any room for improvements that are really meaningful.
No hotfixes at all.
I bet they are working on game specific optimizations, and only for a handful of AAA games, to not make them look as bad as with the FE drivers.

I know you say it'd be gimping, but Lisa did promise a 1H release. If the drivers weren't ready and they hadn't announced a delay to investors in the Q2 call there would likely be legal repercussions for not disclosing any problems with Vega's launch to investors. AMD is really bad at launches but this just takes it to a whole new level, there's no way this architecture is ready for launch in its current state.

Something tells me we're still not seeing the full picture yet, whether this is because RTG really was hamstrung with budget, or if what Raja said in the AMA about losing sight of graphics is actually the reason why they've fallen so far behind.

Edit: Thinking to the road map now, what if Navi and it's scalability is actually in reference to the sizes of the GPU's. If AMD is having issues getting GCN to scale from APU's to High End GPU it's possible that this card is more of an afterthought for the data center and Navi is the true solution their scaling woes.

 

[swilli89]

Only forum nerds will look at those stats.

Consumers will look at performance, power consumption, and the average perception of the card among other factors. This will be the worst selling amd card in recent amd history

You really have grown quite poisonous to these forums. Big reason I think why many are escaping to other, more technical and less emotional forums for discussion.

And let me point out that awful logic you just laid out. You just (seemingly insulting literally everyone here) said that only "forum nerds" look at stats, then go on to say average consumers will be looking at power consumption numbers. Are you serious?

 

[swilli89]

This "bantering" back and forth about a gpu that hasn't even launched yet remind's me of the infamous Abbott and Costello routine:
https://www.youtube.com/watch?v=kTcRRaXV-fg

Yeah its ruined these forums. Mods allow certain posters who just come here to spew negativity to derail the discussion away from a technical one. I think its obvious who does it.

Infraction issued for moderator callout.
-- stahlhart

http://forums.anandtech.com/showthread.php?t=60552

12) Do not contact individual moderators. Please report violations of these guidelines by using the "report" feature located at the bottom of each post. Please direct all questions about specific moderator actions to our Moderator Discussions sub-forum. Please be patient with moderation requests and inquiries. Users found in violation of this directive will be given only one warning and pointed in the right direction before further action is taken.

13) Baiting moderators will not be tolerated nor will Mod Call Outs. Any action that reasonably can be considered baiting a moderator, or multiple consecutive actions that heavily push the boundaries of any of these guidelines will result in an instant short term vacation. Repeated violation of this rule may result in a permaban.
--stahlhart

 

[PeterScott]

Explain this, then. Vega has higher throughput per clock than Fiji. Yet in games sometimes we see decrease in performance per clock, vs Fiji.

Looking around some more, this looks more like Fiji is hamstrung by consumer drivers. 2014 Hawaii chip with fewer units than Fiji, also destroys Fiji in those benchmarks.

http://www.tomshardware.com/reviews/firepro-w9100-performance,3810-9.html

 

[n0x1ous]

Looking around some more, this looks more like Fiji is hamstrung by consumer drivers. 2014 Hawaii chip with fewer units than Fiji, also destroys Fiji in those benchmarks.

http://www.tomshardware.com/reviews/firepro-w9100-performance,3810-9.html

Hawaii had good DP compute, Fiji had it stripped out.

 

[PeterScott]

Hawaii had good DP compute, Fiji had it stripped out.

That only further makes the case that that Fiji has unusually poor "pro" benchmarks, not that Vegas represent some untapped potential not showing up in games.

 

[Glo.]

Looking around some more, this looks more like Fiji is hamstrung by consumer drivers. 2014 Hawaii chip with fewer units than Fiji, also destroys Fiji in those benchmarks.

http://www.tomshardware.com/reviews/firepro-w9100-performance,3810-9.html

You have quoted FirePro review. Professional GPUs which have signed professional drivers do excel in those tests, to the degree, in which as PCPer tested, 550$ professional GPU was able to outperform 1199$ Titan X in some tests.

This is the difference of Professional drivers, which Vega Frontier does not have. Vega Frontier is targeted at professionals, but does not ship with signed, professional drivers. For those - you will have to pay 500-1000$ more with Vega WX 9100.

This is direct Apple vs Apples comparison between both GPUs.

For clues, that this is the case, look no further than to PCPer review of Vega GPU in SpecPerf.

 

[PeterScott]

You have quoted FirePro review. Professional GPUs which have signed professional drivers do excel in those tests, to the degree, in which as PCPer tested, 550$ professional GPU was able to outperform 11.99$ Titan X in some tests.

This is the difference of Professional drivers, which Vega Frontier does not have. Vega Frontier is targeted at professionals, but does not ship with signed, professional drivers. For those - you will have to pay 500-1000$ more with Vega WX 9100.

This is direct Apple vs Apples comparison between both GPUs.

For clues, that this is the case, look no further than to PCPer review of Vega GPU in SpecPerf.

It looks more like AMD just enabled the performance increases in the drivers for Vega FE. After all isn't the claim that this card is NOT for gaming and is for pro usage?

 

[Glo.]

It looks more like AMD just enabled the performance increases in the drivers for Vega FE. After all isn't the claim that this card is NOT for gaming and is for pro usage?

If AMD increased the cache sizes for CU's, and the size of Registry Files, this is what you will see in compute benchmarks, per clock, vs Fiji, regardless of drivers. This is sheer compute horsepower difference, between the GPUs. Its all about the "width" of data per clock that Vega is able to push through.

It appears that AMD actually increased the cache sizes vs Fiji(Actually Polaris already did this), and Registry Files.

Hence the IPC increase touted by AMD in their marketing slides is actually correct.

 

[tential]

You have quoted FirePro review. Professional GPUs which have signed professional drivers do excel in those tests, to the degree, in which as PCPer tested, 550$ professional GPU was able to outperform 1199$ Titan X in some tests.

This is the difference of Professional drivers, which Vega Frontier does not have. Vega Frontier is targeted at professionals, but does not ship with signed, professional drivers. For those - you will have to pay 500-1000$ more with Vega WX 9100.

This is direct Apple vs Apples comparison between both GPUs.

For clues, that this is the case, look no further than to PCPer review of Vega GPU in SpecPerf.

Could you please link to this review so we can view this data as a group?

 

[Glo.]

Could you please link to this review so we can view this data as a group?

Its PCPer review, that has been so much discussed by forum users, its strange that you ask for this link.
https://www.pcper.com/reviews/Graph...B-Air-Cooled-Review/Professional-Testing-SPEC

P2000 is 550$ GPU. It uses GP106 Chip, cut down, to 1024 CUDA Cores. So it is weaker than GTX 1060 3 GB. How come is it faster than Titan X, in some tests?

Isn't this effect of drivers? Quadro P5000 is essentially a GTX 1080. You have proof that Signed, professional drivers have affect on performance of GPUs.

 

[Kenmitch]

After reading the last couple days of comments....I'm still with the wait for the official launch of Vega RX before passing judgement crowd.

 

[tential]

Its PCPer review, that has been so much discussed by forum users, its strange that you ask for this link.
https://www.pcper.com/reviews/Graph...B-Air-Cooled-Review/Professional-Testing-SPEC

P2000 is 550$ GPU. It uses GP106 Chip, cut down, to 1024 CUDA Cores. So it is weaker than GTX 1060 3 GB. How come is it faster than Titan X, in some tests?

Isn't this effect of drivers? Quadro P5000 is essentially a GTX 1080. You have proof that Signed, professional drivers have affect on performance of GPUs.

Comments in bold aren't needed. This is a technical discussion, lets keep it civil. No need to derail this and make it personal.

That's interesting results. I don't know enough about the actual benchmark in question, but yes that proves the underlined point. Looks like we can see it both ways. I've tried finding these recent datapoints for easy comparisons and it's not hard. We're just stuck needing more benches to run with pro and non pro cards.
I don't get why people don't keep $1000s+ professional GPUs on hand for ALL reviews.... /s

Edit: added /s since for the bold since although that should be extremely clear for all of us, some of us may just miss it in the moment or something and not realize it's a joke...

 

[Glo.]

Comments in bold aren't needed. This is a technical discussion, lets keep it civil. No need to derail this and make it personal.

That's interesting results. I don't know enough about the actual benchmark in question, but yes that proves the underlined point. Looks like we can see it both ways. I've tried finding these recent datapoints for easy comparisons and it's not hard. We're just stuck needing more benches to run with pro and non pro cards.
I don't get why people don't keep $1000s+ professional GPUs on hand for ALL reviews....

It is because nobody is using gaming cards for professional usage. The performance drops are too huge.

 

[Elixer]

I know you say it'd be gimping, but Lisa did promise a 1H release. If the drivers weren't ready and they hadn't announced a delay to investors in the Q2 call there would likely be legal repercussions for not disclosing any problems with Vega's launch to investors. AMD is really bad at launches but this just takes it to a whole new level, there's no way this architecture is ready for launch in its current state.

Right, I know that is why they pushed it out the door.

They also claimed that we should expect a similar level of performance (30%-70%+) advantage over the most nearly priced competitor, just like Ryzen.

The 1070 MSRP was $379, 1080 was $599, 1080ti was $699.
So, at minimum, the RX will be $379, and at maximum it will be $699, if they can find something that is at least 30% faster than those to cover what they said at the June 5th conference.
I can only find Hitman where Vega FE vs 1070 was 32% faster, so, price should be $379(4GB?) for Vega RX (assuming same clock).

 

[Snarf Snarf]

At $379 I think Vega would sell quite well. If they sell enough FirePro or Instinct cards for data center the gaming market is mostly irrelevant to them in terms of ASP.

 

[PeterScott]

If AMD increased the cache sizes for CU's, and the size of Registry Files, this is what you will see in compute benchmarks, per clock, vs Fiji, regardless of drivers. This is sheer compute horsepower difference, between the GPUs. Its all about the "width" of data per clock that Vega is able to push through.

It appears that AMD actually increased the cache sizes vs Fiji(Actually Polaris already did this), and Registry Files.

Hence the IPC increase touted by AMD in their marketing slides is actually correct.

Even a W7100 with half the execution units of Fiji, bests it on these "Pro" benchmarks. That makes it blatantly obvious that drivers can unlock massive performance gains in these particular benchmarks.

It is MUCH more likely that all we are seeing here is the difference between Fiji consumer and [semi]Pro drivers on Vega RX.

Its PCPer review, that has been so much discussed by forum users, its strange that you ask for this link.
https://www.pcper.com/reviews/Graph...B-Air-Cooled-Review/Professional-Testing-SPEC

P2000 is 550$ GPU. It uses GP106 Chip, cut down, to 1024 CUDA Cores. So it is weaker than GTX 1060 3 GB. How come is it faster than Titan X, in some tests?

Isn't this effect of drivers? Quadro P5000 is essentially a GTX 1080. You have proof that Signed, professional drivers have affect on performance of GPUs.

This is actually just more evidence that the Vega FE is getting higher workstation scores than Fiji because of drivers, not HW advances.

 

[Tup3x]

I have no idea what has happened with Vega.

On paper it should not perform like it is performing. It has higher throughput than Fiji, and Polaris, both on compute and graphics, yet in graphics it performs like it has lower throughput.

I don't know.

Is AMD's marketing this rubbish to launch unfinished product to lower the hype about it, and surprise with performance at the release of gaming cards?

I would not count on this. Its hard for me to also think this GPU is a failure, because on paper this is very brilliant, and modern architecture, with very high throughput.

On paper it's actually really close to Fiji. Just clock speed is higher. Which leads to another question: how that was achieved? It does have few improvements that should make x amount of improvement. I think those seem to work well in some professional workloads. But those do not necessarily translate to games (although in this case it's kinda pointless to compare Vega FE to gaming cards in professional workloads, because the drivers). Also I think AMD is focusing way too much in compute.

So, did they reduce the IPC to get the clocks up and the new improvements offset the lost IPC so that clock per clock it is more or less the same as Fiji? That could work IF you can get those clock speeds high enough. However it looks like AMD has issues clocking it high enough. That's likely the problem that they are facing. It can't be process' fault solely since GT 1030 runs surprisingly high clocks out of the box while being really energy efficient (I have one passive cooled version in our HTPC and it boost way over 1700 MHz at stock). It uses comparable process (14nm Samsung).

When RX comes out we'll find out for sure.

 

[stockolicious]

Vega has way too many things negative about it vs the gtx 1080 that isn't all washed away by a simple 15% performance advantage(if lucky) and a 10% lower price tag.

Vega has not been released yet. maybe we can agree that Vega doesn't need to be a 1080TI beast to get allot of attention because we forget that AMD is the only company with high end CPU's and GPU's (now that bulldozer retired) . Ryzen and Vega were being built at the same time and both have Infinity fabric features (although i don't know what the special sauce is). I believe "attach rate" on these cards will be high because you put them together and get a great product at a VERY competitive price. Ryzen/Vega = Marketing not required. Maybe the special sauce is the "integration"?

 

[Elixer]

So, you can't undervolt Vega FE at this time...

"VEGA Frontier Edition clock, performance, power scaling // 1600MHz of Flamethrower"
He is guessing you would need a 600W-700W power draw to maintain that speed.
Ouch.
The blower is really, really bad for doing anything of value.

Memory O/Cing is the way to go for Vega (it seems), just got to manage the heat.

Drivers can't fix power draw, without it impacting performance.

 

[OatisCampbell]

I am void of motives, dont have energy for all that. My point is simple if you held out this long AND is willing to shell out dollars for top shelf, if vega turns out to be dud and you are turning to green you might hold out for another 6 months before actually buying into green.

Vega won't be a "dud" compared to a 580 that's basically a 480 that's basically a 7970. AMD has been selling that level of performance a very long time.
This is something new, to AMD at least.

 

[Magee_MC]

So, you can't undervolt Vega FE at this time...

"VEGA Frontier Edition clock, performance, power scaling // 1600MHz of Flamethrower"
He is guessing you would need a 600W-700W power draw to maintain that speed.
Ouch.
The blower is really, really bad for doing anything of value.

Memory O/Cing is the way to go for Vega (it seems), just got to manage the heat.

Drivers can't fix power draw, without it impacting performance.

It sounds to me like that Reddit post that said none of the power efficiency systems that are supposed to be a part of Vega have been activated is correct for the FE.

The only thing that makes sense is that AMD ran into some huge problem with their new systems and had to release it as the FE to meet their statements to investors. Charlie at SA had a tweet where he referenced this saying that AMD took a short term hit for long term credibility and did the right thing by delaying Vega. https://twitter.com/CDemerjian/status/872445675056898048

This whole muddle of a FE Vega release has me thinking that once RX Vega is released it will be faster and less power hungry than we've seen so far with the FE, the big question is how much more can they get out of it when it is finally released? How much power savings will be produced when the full Vega complement of systems is enabled and how much speed will be added.

All in all, this release has me thinking of Winston Churchill's famous quote. Paraphrased and applied to Vega FE, this release, it is a riddle, wrapped in a mystery, inside an enigma; but perhaps there is a key. That key is AMD's long term interests.

 

[Elixer]

If Vega RX is not on some new silicon, then, I just don't see how it can possibly be any faster with that amount of power that is being eaten. There is just no more room to go without falling off the TDP cliff.
They got the same problem with Ryzen & Polaris, it just can't go any higher, the clocks are pretty much pushed as far as they will go, and no amount of driver tweaking can fix silicon issues to get more speed.

Yes, you can limit the max amount of power, at the cost of performance with drivers, but, all power gating should already been enabled at the hardware level.
Since they are trying to push the clocks as high as possible, it makes sense that everything that could be disabled via BIOS of the card has, and they are just using the drivers to control the max power the chip can eat to stay within the TDP limits they picked for the card.

You need to be able to undervolt to verify though, and that is disabled it seems.