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

[KompuKare]

Aaaaaand AMD's typical factory overvoltage strikes again.

1.2v -> 1.075v for the same 1600MHz!?

This is ridiculous.

I'm fairly sure it's not so much their typical factory overvolt, as that AMD still do not seem to any decent binning (at least for the GPUs).

This has been a noticeable problem since at least Tahiti, but despite AMD's card getting tons of bad publicity (7970 GHz, Hawaii etc.), it looks like they haven't learned any lessons.

Obviously, nobody wants to throw away good dies, but the last time AMD even tried to divert leaky parts from the top SKU was with Tahiti and the 7870LE.

As Hawaii showed aggressive binning actually can contribute to higher profits
as the poor initial reviews (at least partially to blame on the high power usage due to wanting to squeeze all dies with all shaders working to full 290X forced them sell the 290X in the end for very little.

Basically:

• Top SKU (290X etc) well binned (and saving 20-50W)
• Second SKU (290 etc.) with shaders bad/fused
  
• Third SKU (7970LE type thing) from dies potentially able to be full shader-count but leaky

even if it look like the third SKU will cost them (assuming they are essentially able to run all the shaders), because of how it makes the 1st SKU look better (less power, less noise etc.), overall they would probably make more money.

 

[Headfoot]

Remember that successful to you != successful to AMD RTG. If the card generates a return on investment costs, increases ASP of RX cards, gets market share back in the high end (which currently is at 0 for this gen), and gains market share in the data center it is a success to AMD. Also I said 2 posts down 650 is too high, it has to be 599 for the best of the best AIB cards. The success of Vega architecture is not going to be decided by the RX alone, it is going to have a full product stack with APU's included. Assuming they lose the top tier halo battle, there's still plenty of success that can happen below that point, some of you guys are way too prone to using hyperbole when talking about graphics cards

Yeah of course im talking about successful to the consumer, i.e. me. This is a forum of consumers. None of us are AMD.

 

[Bouowmx]

Gamers Nexus liquid-cooling mod of the Frontier Edition (air): http://www.gamersnexus.net/guides/2986-amd-vega-hybrid-mod-results-overclocking-liquid-vs-air

400 W max through PCI Express cables (likely 450 W total because 150% power limit of 300 W), for a 3DMark Fire Strike score close to mid-point between NVIDIA GeForce GTX 1080 and 1080 Ti.

 

[Snarf Snarf]

The goalposts just get moved around and have no real meaning.

Yeah of course im talking about successful to the consumer, i.e. me. This is a forum of consumers. None of us are AMD.

I don't have goalposts, it's fair to say that a product is not appealing to you specifically. It's your money and your hobby, you should be happy with your purchases. I'm saying that Vega as a whole can still make AMD money and if it does increase market share it will be viewed internally as a success. This whole "if it doesn't meet my needs this product is absolute garbage and a complete failure" mentality that goes on this forum is toxic.

We get it, you aren't impressed, people will still buy this product for one reason or another. I've said multiple times that the FE launch is an unmitigated disaster, and in its current state that card is a steaming pile for a gaming only purchase. I'm not an AMD apologist, I just happen to realize that the level of emotional outrage that happens on here sometimes is completely off base. We all realize something has gone terribly wrong with Vega launch, if you want to speculate why, or how AMD managed to get this so wrong then go for it. My opinion has been, and still is, that FE only exists to meet 1H target. Either the silicon needed another spin and a tweak, or the drivers are completely non-functional and AMD is damage controlling saying it's just "a different branch". Too many missing features, clock speed behavior is odd, TBR doesn't seem functional, voltage is set way too high at stock, and the list goes on. We're going to know the truth soon once RX Vega launches and our favorite reviewers and forum members get the cards in hand to really dig deep.

TLDR: Let's all be friends and speculate peacefully without the hyperbole and emotional bashing of a yet to be released product.

 

[tential]

. I'm saying that Vega as a whole can still make AMD money and if it does increase market share it will be viewed internally as a success. .

No my problem is your post doesn't actually say anything. I'm focusing on the part here thats relevant because the rest of your post starts to get personal or talks about how things can benefit amd.

This comment I've quoted says nothing. It's the blandest of comments. That's my problem. You basically have said "if this product generates real revenue and increases market share it will be good."
You haven't said anything of substance.


Something of substance would be providing a rational for how Vega will increase marketshare, generate enough revenue to be a success, and be a successful product overall, and how that could happen, which is something you aren't able to do in reference to Vega.

So TLDR;
Don't throw out vague comments. Provide a reason for your statements. And stop trying to make it personal with me. You want this place to be civil, then stick to Vega, which is what I'm asking you about.

 

[Snarf Snarf]

So TLDR;
Don't throw out vague comments. Provide a reason for your statements. And stop trying to make it personal with me. You want this place to be civil, then stick to Vega, which is what I'm asking you about.

My apologies if you think I'm coming directly after you, I'm not. I have nothing against you or your ideas. You said I moved goal posts, and I responded that I don't have any goalposts.

Objectively speaking, Vega is going to be a stack of products, and it appears that the halo product is going to lose head to head for the second generation running barring a miracle or the greatest sandbagging job in the history of GPU launches.

How can Vega still be a commercial success you say? The FE performs pretty admirably in professional work loads, combine that with the packed math feature and at the very least will be a very attractive card once the FirePro and Instinct variants are released with professional quality signed drivers. On the APU front the signs of FE being able to run at almost 1v 1600MHz locked is a promising sign, it shows that better scaling is possible and aggressive binning combined with a lower clock speed of around 1400 Mhz or below with a quad core of Haswell level IPC will be an excellent product. My point being that RX Vega can be a flop but the Vega architecture can still be a huge success for AMD. I'd love to speculate more on the basis of the steamy turd that is FE but all signs point to it being a half baked product not ready to properly evaluate.

 

[tential]

What would you do about the Power consumption issue of vega as well as the poor gaming performance scaling which is what an apu would be used for?
I'm not sure why you bring up the Vega stack now, but if that's the case, the rest of the stack is in far worse position than the fe.

The apu being the worst of the bunch due to Vega doing nothing for it in the 10 form. I honestly thought Vega 11 was needed for apus.

Edit:the undervolt is promising but without knowing the actual power consumption we don't really know. Since it's amd with hbm, we don't even know if anything is working or reporting correctly currently.

 

[tviceman]

What would you do about the Power consumption issue of vega as well as the poor gaming performance scaling which is what an apu would be used for?
I'm not sure why you bring up the Vega stack now, but if that's the case, the rest of the stack is in far worse position than the fe.

The apu being the worst of the bunch due to Vega doing nothing for it in the 10 form. I honestly thought Vega 11 was needed for apus.

Edit:the undervolt is promising but without knowing the actual power consumption we don't really know. Since it's amd with hbm, we don't even know if anything is working or reporting correctly currently.

Everything can undervolt, though. Setting the power limit on Pascal at a particular level can significantly impact power draw with minimal impact on performance. Heck, my i7-6700k and 7700HQ are both undervolted.

Undervolting is not an AMD specific phenomenon, it just happens to be a more focused topic with AMD products because, as of late, there is no headroom left in the tank to OC.

 

[Veradun]

I'm fairly sure it's not so much their typical factory overvolt, as that AMD still do not seem to any decent binning (at least for the GPUs).

This has been a noticeable problem since at least Tahiti, but despite AMD's card getting tons of bad publicity (7970 GHz, Hawaii etc.), it looks like they haven't learned any lessons.

Obviously, nobody wants to throw away good dies, but the last time AMD even tried to divert leaky parts from the top SKU was with Tahiti and the 7870LE.

I don't think you can really say this is a yeld issue. You can still bin accordingly and have the same parametric yelds with a fairly broad selection of die qualities to selle the less power hungry as top bin and the fournaces as lowest bin. BUT the more you bin selectively and the more testing costs you.

My bet this "factory overvoltage" is so big on AMD products because of budget, not because of yelds.

tl;dr: they want to lower their testing costs, so instead of doing a lot of binning they set really high voltage targets and forget about it.

 

[beginner99]

I don't think you can really say this is a yeld issue. You can still bin accordingly and have the same parametric yelds with a fairly broad selection of die qualities to selle the less power hungry as top bin and the fournaces as lowest bin. BUT the more you bin selectively and the more testing costs you.

My bet this "factory overvoltage" is so big on AMD products because of budget, not because of yelds.

tl;dr: they want to lower their testing costs, so instead of doing a lot of binning they set really high voltage targets and forget about it.

Makes sense. Golden chip from AMD doesn't mean high oc but a little OC and a lot lower voltage.

 

[Zstream]

What would you do about the Power consumption issue of vega as well as the poor gaming performance scaling which is what an apu would be used for?
I'm not sure why you bring up the Vega stack now, but if that's the case, the rest of the stack is in far worse position than the fe.

The apu being the worst of the bunch due to Vega doing nothing for it in the 10 form. I honestly thought Vega 11 was needed for apus.

Edit:the undervolt is promising but without knowing the actual power consumption we don't really know. Since it's amd with hbm, we don't even know if anything is working or reporting correctly currently.

You have no idea wth you're talking about. You have no idea of performance, power, cost of Vega non-fe, or even the APU.

Stick to things you do know, like Vega FE, and how a majority of its features are disabled.

 

[Jackie60]

The fact that so many of Vega FE's features are as yet disabled suggests gaming performance could skyrocket with the Vega RX launch. I don't think it's a good idea for Vega RX's gaming performance to be revealed until the card is either widely available or imminently widely available. In the past Nvidia have always sought to scupper a new AMD launch with a product of their own and given the 1080ti is year old architecture we can be fairly sure Nvidia have something up their sleeve ready to drop as soon as AMD's hand is revealed. That's what I hope is the case and would explain the lack of gaming optimisations activated in the VEGA FE. Here's hoping!

 

[Elixer]

The fact that so many of Vega FE's features are as yet disabled suggests gaming performance could skyrocket with the Vega RX launch. I don't think it's a good idea for Vega RX's gaming performance to be revealed until the card is either widely available or imminently widely available.
...

This is not true if you believe AMD's Senior Manager of Game Engineering.
https://forum.beyond3d.com/posts/1989522/

Not gimped = not disabled.

If you want to argue with him that the drivers are in fact gimped, well, good luck.

 

[3DOSH]

s

This is not true if you believe AMD's Senior Manager of Game Engineering.
https://forum.beyond3d.com/posts/1989522/

Not gimped = not disabled.

If you want to argue with him that the drivers are in fact gimped, well, good luck.

Thats not necessarily true. It could that the features had to be disabled because they are not ready/unstable yet. Gimped means it was disabled for no reason.

 

[Aenra]

Guys.. English
Simple words, simple meanings.

'Not gimped' means that it is not purposefully made to underperform, malfunction, etc.
Ergo saying 'not gimped' does -not- preclude the possibility that it --IS-- underperforming, malfunctioning, etc., except without anyone wanting it to. It just.. is; for now.

Why do we complicate the easy stuff? ^^

 

[Mopetar]

What would you do about the Power consumption issue of vega as well as the poor gaming performance scaling which is what an apu would be used for?

Any GPU has poor power efficiency when its fed high-levels of voltage. Raven Ridge APUs aren't going to throw as much voltage at the GPU as a desktop part, even if that means lower clock speeds.

You should also be smart enough to reason that the FE isn't going to be indicative of final gaming performance for Vega.

Even if the performance isn't world shattering, it hardly needs to be as AMDs competition in the mobile space is Intel who has terrible performance on the GPU side. AMDs APUs have always been held back by anemic CPU performance, but Ryzen solves that. They could honestly ship Raven Ridge using their previous generation GCN tech and still be miles ahead of Intel. Using Vega is just a bonus.

The apu being the worst of the bunch due to Vega doing nothing for it in the 10 form. I honestly thought Vega 11 was needed for apus.

I wouldn't be surprised if the APU is a single piece of silicon with the GPU and CPU on the same die. They probably won't get to having a modular GPU approach until Navi and the 8 cores from a Ryzen die would be overkill, never mind that HBM doesn't appear to be available in the quantity or at the price AMD would need to make a true mobile SoC. Instead just use a single Ryzen CCX and base the GPU around Vega tech. Even if it's just 8 CUs, they should be well ahead of Intel's offerings.

 

[Cloudfire777]

Pretty sure the Vega Hybrid HardwareNexus built, with the OC, is the best case scenario for Vega FX

That should be a good indication I think. Go check out their results

 

[Snarf Snarf]

Guys.. English
Simple words, simple meanings.

'Not gimped' means that it is not purposefully made to underperform, malfunction, etc.
Ergo saying 'not gimped' does -not- preclude the possibility that it --IS-- underperforming, malfunctioning, etc., except without anyone wanting it to. It just.. is; for now.

Why do we complicate the easy stuff? ^^

When Raja said that whole "I wish you could tell my driver team it was Polaris+" and "making good drivers is the hardest part of GPU design" I think that pretty much answers that debate. They aren't intentionally non-functional, just accidentally. I'm not sure which is worst TBH but it is what it is. Lets just hope someone is sneaky at one of these Vega tour events and pulls up some more info on the hardware for a quick picture

 

[lobz]

This is not true if you believe AMD's Senior Manager of Game Engineering.
https://forum.beyond3d.com/posts/1989522/

Not gimped = not disabled.

If you want to argue with him that the drivers are in fact gimped, well, good luck.

This. Please decide for all people, what they really mean when they write something different. Because that's clearly your superpower.

 

[Elixer]

When Raja said that whole "I wish you could tell my driver team it was Polaris+" and "making good drivers is the hardest part of GPU design" I think that pretty much answers that debate. They aren't intentionally non-functional, just accidentally. I'm not sure which is worst TBH but it is what it is. Lets just hope someone is sneaky at one of these Vega tour events and pulls up some more info on the hardware for a quick picture

"Accidentally"? If that was the case, then they would have released a hotfix.
There have been no updates at all, even though there have been updates for all their other cards.

Besides, you don't think they have test suites to test if things are actually working as expected?
(Yes, this is for the Pro stuff, but the non-Pro drive also have test suites as well)

For designers, content creators, and game developers, Radeon Pro Software Crimson ReLive Edition delivers outstanding productivity and stability with up to 30 percent performance improvements in key applications2 and undergoing twice the OEM platform testing, three times more ISV certification testing and 1.5 times more stress testing than the previous AMD professional graphics driver.3

2) Up to 30% faster performance using 16.50 and Catia on AMD FirePro™ W7100 than with 14.502.1019. Testing conducted by AMD Performance Labs as of November 7th, 2016 on a test system comprising of Intel E5-1650 v3 3.50GHz, 16GB RAM, Win7 64-bit SP1, AMD FirePro™ W7100, software drivers 16.50 and 14.502.1019 for SPECviewperf benchmark. Benchmark Application: SPECviewperf 12.1 under official run. Subtest: catia-04. FirePro W7100 (with 16.50) score: 74.48. FirePro W7100 (with 14.502.1019) score: 57.33. Performance Differential: 74.48/57.33 = ~29.9% higher score on FirePro W7100 with 16.50. PC manufacturers may vary configurations, yielding different results. Performance may vary based on use of latest drivers. SPEC® and the benchmark name SPECviewperf® are registered trademarks of the Standard Performance Evaluation Corporation. For more information about SPECviewperf, see https://www.spec.org/gwpg/gpc.static/vp12.1info.html. FS-6

3) AMD internal data, testing of Radeon Pro 16.Q4 Enterprise Driver compared to FirePro™ 14.502.1019 driver.

While they may not be optimized the best they could be, as seen in the past by other driver releases (and the quote above) that get better with time, to purposely disable parts of their own $1000 card is, as Rys said, ridiculous.

Will we see 30% gains or higher? Possible for specific applications/games.
A bit over 2 weeks to find out just how much better the newer drivers will be.

 

[Snarf Snarf]

A released hotfix implies that they have a fix for it already made... Like you say though we will know in about 2 weeks for sure what's been going on.

There's also a good chance that this plays out like Ryzen and once apps start to release Vega patches we'll see the gap close on the 1080ti. Even then 30-40% is a really big gap to close, but that is within the best case scenario range that we saw from Ashes performance patch. That is some impressive gains in pro work though from updates, I'm just not going to hold my breath for that kind of uplift in gaming.

 

[lobz]

"Accidentally"? If that was the case, then they would have released a hotfix.
There have been no updates at all, even though there have been updates for all their other cards.

Besides, you don't think they have test suites to test if things are actually working as expected?
(Yes, this is for the Pro stuff, but the non-Pro drive also have test suites as well)


While they may not be optimized the best they could be, as seen in the past by other driver releases (and the quote above) that get better with time, to purposely disable parts of their own $1000 card is, as Rys said, ridiculous.

Will we see 30% gains or higher? Possible for specific applications/games.
A bit over 2 weeks to find out just how much better the newer drivers will be.

Are you really not seeing the possibility that these features are not _disabled_ per se, just not fully ready to be enabled and they rather turn them all on when they are fully functional and stable? I know I would.. and in the meantime, the FE card fares rather well against its intended competition in the areas at the card is marketed at.
Geez just wait till end of july, you will have plenty of opportunity to bash rx vega if the gaming performance is still lackluster

Sent from my VTR-L09 using Tapatalk

 

[Elixer]

Are you really not seeing the possibility that these features are not _disabled_ per se, just not fully ready to be enabled and they rather turn them all on when they are fully functional and stable? I know I would.. and in the meantime, the FE card fares rather well against its intended competition in the areas at the card is marketed at.
Geez just wait till end of july, you will have plenty of opportunity to bash rx vega if the gaming performance is still lackluster

I am not trying to bash here, I am trying to temper the exceedingly high expectations that people are having by coming up with all sorts of reasons why it isn't performing better than it is. Yes, we should know in ~2 weeks, and possibly know more on Tuesday, July 25, after 5PM EST which is when AMD's quiet period is over and they are going to be asked questions, and they can finally tell us things.

As for features being disabled / not ready / gimped / whatever you want to call it, I have already pointed out what AMD have said. Is it possible he was kidding? Sure, but, realistically, no.
I am just saying the base code is there, and it is being utilized. That still leaves the door open to optimizations as I said above, we just don't know how far it can be optimized.

 

[Cloudfire777]

I am not trying to bash here, I am trying to temper the exceedingly high expectations that people are having by coming up with all sorts of reasons why it isn't performing better than it is. Yes, we should know in ~2 weeks, and possibly know more on Tuesday, July 25, after 5PM EST which is when AMD's quiet period is over and they are going to be asked questions, and they can finally tell us things.

As for features being disabled / not ready / gimped / whatever you want to call it, I have already pointed out what AMD have said. Is it possible he was kidding? Sure, but, realistically, no.
I am just saying the base code is there, and it is being utilized. That still leaves the door open to optimizations as I said above, we just don't know how far it can be optimized.

If Vega RX was that much faster than FE they would have released the card already.
Or have shown direct comparisons with 1080Ti on several games. Let sites do a preview on its performance.

TDP over Vega FE is basically unchanged.

Vega FE Air: 300W
Vega XT: 285W

Vega FE AIO: 375W
Vega XTX: 375W

So dont expect a 1700MHz Vega that HardwareNexus tested.

 

[Glo.]

If Vega RX was that much faster than FE they would have released the card already.
Or have shown direct comparisons with 1080Ti on several games. Let sites do a preview on its performance.

Or AMD's marketing department is rubbish to the degree that they deliberately released crippled GPU, with inactive features, wrong BIOS, that not enables full control over voltages, to lower expectations, and destroy the hype train, and then to show "real" performance of the architecture, with RX Vega release.