AMD Radeon RX Vega 64 and 56 Reviews [*UPDATED* Aug 28]

[krumme]

A rx64 in powersave mode - not 2. bios - is 48w more watt than a 1080fe
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/29.html

For the games i play in 4k its midway between 1080fe and 1080ti in perf.
IN POWERSAVING MODE THAT IS.
With freesynch thats damn fine for the 1080 minus 10% i paid. Hell even plus 5% its okey.
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/15.html
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/9.html

i dont know where the haters get their numbers but i see a rx64 getting awfully close like 8% to a 1080ti while actually using a bit less watt in bf1 and doom.
And its in the forward looking engines that matters.

I can see that the rx64 with prey and doom for 1080 plus 6% price is sold out where i live. next stock is 2 oct. Guess it mean aib cards then when there is next stock. A long wait.

 

[simeon93]

RX Vega 64 I call it the Graphix beast....

 

[tential]

A rx64 in powersave mode - not 2. bios - is 48w more watt than a 1080fe
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/29.html

For the games i play in 4k its midway between 1080fe and 1080ti in perf.
IN POWERSAVING MODE THAT IS.
With freesynch thats damn fine for the 1080 minus 10% i paid. Hell even plus 5% its okey.
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/15.html
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/9.html

i dont know where the haters get their numbers but i see a rx64 getting awfully close like 8% to a 1080ti while actually using a bit less watt in bf1 and doom.
And its in the forward looking engines that matters.

I can see that the rx64 with prey and doom for 1080 plus 6% price is sold out where i live. next stock is 2 oct. Guess it mean aib cards then when there is next stock. A long wait.

You make a graphics card evaluation based off a 2 game benchmark suite and are questioning why other people don't use your reasoning?
Ok.

I think you're confused. What if retailers need to sell the card at minimun 650$ to even make a minor profit(MSRP is meaningless). Does this not make your whole point pointless

In order to not derail the thread into an economics lesson I PMed you a "full" explanation.

$650 is below the current selling point and below the equilibrium price examples in both posts explaining this(It was clearly a variable either way and not a hard number.) So markets would clear and the card would be sold out since the value of the current crop of people who purchased the card attribute more than $650 of value to it. Ask me further questions in the PM thread as to how this works if needed since it's getting to not be about Vega and more about economics.

 

[CatMerc]

I see what you're saying, but you're also falling into this trap AMD wants you to dig yourself into. Do you think it's a smart move for people to spend $500-600 on the RX Vega 64 graphics card, put it in power saving mode where it ends up 10% slower and and still 25% less efficient than the likely-cheaper GTX 1080? And further, I'm sure if Nvidia wanted to put GTX 1080 into some kind of power saver / mobile mode on the desktop and drop power consumption down to 130 watts for a 10% loss in performance they could easily do that, thus making RX Vega's power saving mode still look as inefficient as it's balanced and turbo modes. But, you see, Nvidia doesn't have to waste time creating driver profiles that function in power saving modes because their cards don't draw an embarrassing amount of power anymore.

Vega is more than 50% larger in die size, has 60% more transistors, and consumes 120+ watts more when going toe-to-toe with the GTX 1080. The 390x was much closer in die size and transistors to it's immediate competitor (the GTX 980) while consuming similar or less power than RX Vega 64.

Vega is not a step in the right direction. Fiji was a huge step in the right direction (they were actually competitive with Fiji, OMG remember that?!). Polaris was a small stumble, Vega is lame duck. AMD can't keep throwing transistors and TDP at the problem; they're at the wall. If they don't heavily innovate they will be even worse off next round because I guarantee you Volta will not be some 5-10% update like Intel's CPU's were for years which allowed AMD to catch Intel in the CPU space.

I don't disagree that Vega is still clearly behind NVIDIA. But I don't believe it's right to say it regressed from Polaris.

As for NVIDIA efficiency mode, their scaling is pretty linear in terms of power and performance. I don't believe you can increase efficiency much more.

 

[tviceman]

I don't disagree that Vega is still clearly behind NVIDIA. But I don't believe it's right to say it regressed from Polaris.

As for NVIDIA efficiency mode, their scaling is pretty linear in terms of power and performance. I don't believe you can increase efficiency much more.

Perf/w from RX 480 - before AMD decided to butcher Polaris's efficiency, is clearly down. (109 - 97)/109 = Polaris is 12% more efficient or (109 - 93) / 93 = Polaris is 17% more efficient depending on which bios is used for full speed Vega.

It's not an improvement, it's not even equal, it's a drop.

 

[tential]

Look at how telling that graph is with the whole GTX 10 series line up just together.
164% From the GTX 1080ti!

@tviceman actually has said this NUMEROUS times in the past and it really clicks now with me.

In my own words:
Overall efficiency defines the top level of performance your architecture can reach.

So with the GTX 1080Ti being 165% ahead of AMD currently, that's a MASSIVE lead.
On top of that the perf/watt of AMD just declined this generation. It may get better with drivers, but meanwhile look at Nvidia

104% with the 980Ti
164% with the 1080Ti.

I don't envy anyone having to compete with a company making improvements like that.

 

[raghu78]

I don't disagree that Vega is still clearly behind NVIDIA. But I don't believe it's right to say it regressed from Polaris.

As for NVIDIA efficiency mode, their scaling is pretty linear in terms of power and performance. I don't believe you can increase efficiency much more.

I would say that Vega atleast in its current form regressed from Polaris in terms of perf/watt and perf/sq mm

https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_56/32.html
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/32.html

https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_56/31.html
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/31.html

Vega is using 8GB HBM2 at 945 Mhz over a 2048 bit bus which brings significant power savings over Rx 480 / Rx 580's 256 bit GDDR5 memory at 8 Ghz. The estimate was the Rx 480 GPU power draw was 110w and GDDR5 power consumption on Rx 480 was close to 45-50w. I am guessing 8GB HBM2 is shaving off atleast 30w from that power draw. Vega is roughly 2x the die size of Rx 480. Vega 10 struggles to increase perf by 70% over Rx 480 inspite of power gains from HBM2. I don't think its wrong to say that Vega 10 has regressed in perf/watt and perf/sq mm from Rx 480.

AMD's scalability problems with GCN are clearly illustrated from the comparison of GTX 1060 vs GTX 1080 and Rx 480 vs Vega.

GTX 1060 - 116w (200 sq mm)
Rx 480 - 163w (232 sq mm)
GTX 1080 - 166w (314 sq mm)
Rx Vega - 292w (486 sq mm)

Nvidia GTX 1080 gets much better perf scaling over GTX 1060 compared to Vega 10 over Rx 480 when you factor in die size and power . GTX 1080 has better perf/watt than GTX 1060 while Vega 10 has lesser perf/watt than Rx 480.

 

[tential]

I would say that Vega atleast in its current form regressed from Polaris in terms of perf/watt and perf/sq mm

https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_56/32.html
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/32.html

https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_56/31.html
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/31.html

Vega is using 8GB HBM2 at 945 Mhz over a 2048 bit bus which brings significant power savings over Rx 480 / Rx 580's 256 bit GDDR5 memory at 8 Ghz. The estimate was the Rx 480 GPU power draw was 110w and GDDR5 power consumption on Rx 480 was close to 45-50w. I am guessing 8GB HBM2 is shaving off atleast 30w from that power draw. Vega is roughly 2x the die size of Rx 480. Vega 10 struggles to increase perf by 70% over Rx 480 inspite of power gains from HBM2. I don't think its wrong to say that Vega 10 has regressed in perf/watt and perf/sq mm from Rx 480.

AMD's scalability problems with GCN are clearly illustrated from the comparison of GTX 1060 vs GTX 1080 and Rx 480 vs Vega.

GTX 1060 - 116w (200 sq mm)
Rx 480 - 163w (232 sq mm)
GTX 1080 - 166w (314 sq mm)
Rx Vega - 292w (486 sq mm)

Nvidia GTX 1080 gets much better perf scaling over GTX 1060 compared to Vega 10 over Rx 480 when you factor in die size and power . GTX 1080 has better perf/watt than GTX 1060 while Vega 10 has lesser perf/watt than Rx 480.

Are we able to separate:
AMD architecture vs Nvidia architecture
Global Foundries vs TSMC
?

What I mean is, how much of the difference is architecture vs manufacturing? How are we able to tell?

 

[tviceman]

I would say that Vega atleast in its current form regressed from Polaris in terms of perf/watt and perf/sq mm

https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_56/32.html
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/32.html

https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_56/31.html
https://www.techpowerup.com/reviews/AMD/Radeon_RX_Vega_64/31.html

Vega is using 8GB HBM2 at 945 Mhz over a 2048 bit bus which brings significant power savings over Rx 480 / Rx 580's 256 bit GDDR5 memory at 8 Ghz. The estimate was the Rx 480 GPU power draw was 110w and GDDR5 power consumption on Rx 480 was close to 45-50w. I am guessing 8GB HBM2 is shaving off atleast 30w from that power draw. Vega is roughly 2x the die size of Rx 480. Vega 10 struggles to increase perf by 70% over Rx 480 inspite of power gains from HBM2. I don't think its wrong to say that Vega 10 has regressed in perf/watt and perf/sq mm from Rx 480.

AMD's scalability problems with GCN are clearly illustrated from the comparison of GTX 1060 vs GTX 1080 and Rx 480 vs Vega.

GTX 1060 - 116w (200 sq mm)
Rx 480 - 163w (232 sq mm)
GTX 1080 - 166w (314 sq mm)
Rx Vega - 292w (486 sq mm)

Nvidia GTX 1080 gets much better perf scaling over GTX 1060 compared to Vega 10 over Rx 480 when you factor in die size and power . GTX 1080 has better perf/watt than GTX 1060 while Vega 10 has lesser perf/watt than Rx 480.

This is all true, but Vega has a higher percentage of transistors dedicated to compute than Polaris. Vega is essentially AMD's Fermi from a GPGPU point of view. It's trying to be a jack of all trades and compete with GP100 in compute while also competing with GP104 in gaming.

 

[tviceman]

Look at how telling that graph is with the whole GTX 10 series line up just together.
164% From the GTX 1080ti!

@tviceman actually has said this NUMEROUS times in the past and it really clicks now with me.

In my own words:
Overall efficiency defines the top level of performance your architecture can reach.

So with the GTX 1080Ti being 165% ahead of AMD currently, that's a MASSIVE lead.
On top of that the perf/watt of AMD just declined this generation. It may get better with drivers, but meanwhile look at Nvidia

104% with the 980Ti
164% with the 1080Ti.

I don't envy anyone having to compete with a company making improvements like that.

If AMD had been able to improve efficiency over RX 480, or at least stay as efficient, then Vega would average ~10-15% faster than a GTX 1080 at 1440p, making Vega's release way, way more interesting and competitive on all fronts.

Now Polaris and Vega need to float AMD until Navi comes (probably very late 2018, if not 2019), while Volta will be absolutely crushing AMD by a 2 generational gap in perf/w,

 

[CatMerc]

Perf/w from RX 480 - before AMD decided to butcher Polaris's efficiency, is clearly down. (109 - 97)/109 = Polaris is 12% more efficient or (109 - 93) / 93 = Polaris is 17% more efficient depending on which bios is used for full speed Vega.

It's not an improvement, it's not even equal, it's a drop.

I said in power saver mode, not at stock. Stock config is more akin to 580 series pushed to the extreme limits for next to no gains.

 

[tviceman]

I said in power saver mode, not at stock. Stock config is more akin to 580 series pushed to the extreme limits for next to no gains.

We are doing circles now. If power saving mode is what Vega "should be at" then all benchmarks should focus on power saving mode. That isn't the case though. Power saving mode is glossed over, if at all, just to show that it exists. We are talking about a $500+ card trying to keep up with the competition. If AMD didn't need to create a power saving mode, they wouldn't have to. Nvidia cuts 10% performance off GTX 1080 in laptop form and saves 25% in power. "Power saving mode" on a flagship product that can't compete is an apology for crap engineering

 

[PeterScott]

While I agree Nvidia is more tenacious than Intel, no one should count out AMD of either space.

Navi is Raja's baby, Vega is just another iteration of GCN with some hardware bits not even being utilized.

People keep using that deflection, like Vega was set in stone 5 years ago before Raja joined (4 and half years ago).

Jim Keller was only at AMD for about 4 years, and we got Zen.

You can't pretend Raja isn't responsible for Vega.

 

[CatMerc]

We are doing circles now. If power saving mode is what Vega "should be at" then all benchmarks should focus on power saving mode. That isn't the case though. Power saving mode is glossed over, if at all, just to show that it exists. We are talking about a $500+ card trying to keep up with the competition. If AMD didn't need to create a power saving mode, they wouldn't have to. Nvidia cuts 10% performance off GTX 1080 in laptop form and saves 25% in power. "Power saving mode" on a flagship product that can't compete is an apology for crap engineering

You're the one creating the circles in your head. My one and only claim is that perf/watt is better than Polaris. Have you seen how horrible 580 perf/watt is? That's what 64 at stock is.

If we're gonna compare Vega to RX 400 series, then it's only fair to test under similar conditions. I.E not pushed to the brink.

Look at how absurd this whole thing is. 4% extra performance for 36% extra power? Does this sound like a reasonable stock config for you?

 

[stockolicious]

Regardless of whether or not its the best gaming card, the primary revenue stream for the RTG is gaming. Same goes for nVIDIA as the majority of their revenue and profits come from the gaming market. Its their core and primary business in terms of video cards. So saying things like "its not meant to be" is a poor attempt at masking their failure. If for all intent and purpose that was their goal, to go for a more compute orientated card and go after those markets, then they've essentially risked their primary business doing so.

From my industrial experience (I know its anecdotal but Im thinking alot of users will share the same sentiment), only a few very specific type of work loads in a prosumer environment (engineering for instance) actually benefit from having a workstation or compute GPU to do the work. Most of the performance comes down to the CPU/Memory and the hard disk..

VEGA has failed to meet performance targets. Its missed the power targets. Its not even that great in terms of compute mainly due to lacking the sofare toolchains and libraries. And its late.

And can we not start one of those "the human eye can only see X FPS so A cards are not required" arguments.. and no your general adobe experience outside very specific work types will not be greatly enhanced by using VEGA as opposed to a GTX1080. Actually Id rather have the 1080 because with VEGA your going to consume more power even for 2D tasks.

"the human eye can only see X FPS so A cards are not required"

please - if your going to put something in quotes why change what i said?

"Regardless of whether or not its the best gaming card, the primary revenue stream for the RTG is gaming"

AMD has never made a penny off its graphics card business, even in past years when they had better cards. The revenue streams for RTG are going to become more diverse and not just gaming.

FYI- the attach rate for AMD dGPU's when you have an AMD CPU is about 55%. AMD is going to sell a ton of those cards on the coat tails of Ryzen right up to Navi - go to dell.com and check out the 41 designs and see what the graphics card options are. Then keep in mind last year they had "zero".

 

[PeterScott]

Why is it that Intel wasn't tenacious? I don't think Intel should have offered more cores. It would have hurt their profit margins for no reason.

That's doesn't mean Intel was resting on chip development. They were definitely trying to increase IPC. Do you think Intel pours that money into R&D got fun? This narrative that Intel stalled cpu development is the largest joke ever

What Nvidia is doing is execution on another level. Like I said when the gtx 1070 came out. Nvidia is executing on a level in that I'd rather own their stock than their gpu.

Nvidia increasing performance and efficiency generation after generation in this manner is just beautiful to see..

The difference is Jensen Huang, Founder and CEO of NVidia. Some founders may not be suited to lead companies but those that are tend to be more tenacious than the average CEO in it for the paycheck.

If NVidia had merged with AMD, instead of AMD merging with ATI, I am sure the Bulldozer fiasco never would have happened, and NVidia CPUs would have been challenging Intel all along.

Smart, driven, invested leadership matters.

 

[raghu78]

This is all true, but Vega has a higher percentage of transistors dedicated to compute than Polaris. Vega is essentially AMD's Fermi from a GPGPU point of view. It's trying to be a jack of all trades and compete with GP100 in compute while also competing with GP104 in gaming.

This is AMD's problem. Nvidia has gotten way more profitable after Maxwell generation that they can afford separate GPUs for compute (GP100) and gaming (GP102 / GP104 / GP106) while AMD is struggling to generate meaningful profits. Hopefully EPYC gives AMD a double digit market share in servers by late 2018 and decent profits to reinvest in RTG towards future generations.

If AMD had been able to improve efficiency over RX 480, or at least stay as efficient, then Vega would average ~10-15% faster than a GTX 1080 at 1440p, making Vega's release way, way more interesting and competitive on all fronts.

Now Polaris and Vega need to float AMD until Navi comes (probably very late 2018, if not 2019), while Volta will be absolutely crushing AMD by a 2 generational gap in perf/w,

https://videocardz.com/69571/amd-reveals-vega-and-zen-roadmaps

AMD's 2018 GPUs are likely to use a higher performance 14nm+ as seen from roadmaps. The launch drivers for Vega do not even enable features like primitive shaders or HBCC. If AMD can improve drivers over the next 6-9 months architecturally Vega might start to look better (right now its a dud). The next thing AMD need is better execution from GF who have been a major handicap for AMD against Nvidia and TSMC. There also seem to be issues with HBM2 not hitting 2 Ghz clocks as expected and needing to be overvolted to even hit 1.89 Ghz. Hopefully by mid-late H1 2018 AMD can get better yields and clocks on HBM2. If AMD can avoid the perf/watt gap from increasing when Volta launches that would be the best they can do in 2018. Obviously that implies AMD have to execute vastly better in 2018 than 2016-2017 both in terms of hardware and software.

 

[Zstream]

If NVidia had merged with AMD, instead of AMD merging with ATI, I am sure the Bulldozer fiasco never would have happened, and NVidia CPUs would have been challenging Intel all along.

Don't be so sure about that. Nvidia doesn't have the best track record with CPU's, and they've had some blunders as well. If Bulldozer came to be, Nvidia might have allocated two cores or something to driver overhead, and CPU performance would be better. This is all hypothetical anyways, but sure is fun to speculate

 

[PeterScott]

Don't be so sure about that. Nvidia doesn't have the best track record with CPU's, and they've had some blunders as well. If Bulldozer came to be, Nvidia might have allocated two cores or something to driver overhead, and CPU performance would be better. This is all hypothetical anyways, but sure is fun to speculate

I am sure. Jensen is a hard driver who knows CPUs, and Knows AMD. He worked at AMD before founding NVidia. Crap like Bulldozer never would have been built under his leadership.

 

[caswow]

I am sure. Jensen is a hard driver who knows CPUs, and Knows AMD. He worked at AMD before founding NVidia. Crap like Bulldozer never would have been built under his leadership.

i guess you forgot the tegra devision that got subsidized by its gpu devision for half a decade. dont tell me they wanted them into cars first they wanted to be in mobile devices even bought a company for their radio bands and lost millions there too. they barely make money on them after losing millions each year.

 

[Guru]

Is DX12 and Vulkan going to help AMD in 2018, considering their cards do have an advantage in these API's and in 2018 we are likely to see a massive shift to DX12 and Vulkan as the preferred API to develop games on, starting even from scratch.

Or will Navi just not be faster enough from Vega, than even that advantage in these low level api's is going to be lost, assuming consumer Volta is actually better than Pascal at Dx12/Vulkan.

 

[richaron]

Perf/w from RX 480 - before AMD decided to butcher Polaris's efficiency, is clearly down. (109 - 97)/109 = Polaris is 12% more efficient or (109 - 93) / 93 = Polaris is 17% more efficient depending on which bios is used for full speed Vega.

It's not an improvement, it's not even equal, it's a drop.

Lol, perf/watt is clearly better than a 580 in the information you provide but somehow you're missing that. It's obvious to everyone that Vega 64 has been pushed way past it's sweet spot of efficiency just like the 580.

But Vega is more efficient than Polaris in 580. Something 100% the opposite of what you are trying to prove. I understand that you have your personal narrative to keep pushing, but best not to provide such easy counter arguments in your own source material o_0

 

[krumme]

We are doing circles now. If power saving mode is what Vega "should be at" then all benchmarks should focus on power saving mode. That isn't the case though. Power saving mode is glossed over, if at all, just to show that it exists. We are talking about a $500+ card trying to keep up with the competition. If AMD didn't need to create a power saving mode, they wouldn't have to. Nvidia cuts 10% performance off GTX 1080 in laptop form and saves 25% in power. "Power saving mode" on a flagship product that can't compete is an apology for crap engineering

Tpu took power saving mode on their entire bm suite.
You gain 30% perf/w efficiency. And get below 980ti fe power usage.
At a loss of meager 4% perf.
Fact.

And its litteraly a push with a button in driver setup with the std bios. Every one buying this card have the ability to do it. If power is an issue for you its simply in all practicallity solved by the cost of 4% perf.

And at 4k that this card is meant to be played it still does it good deal faster in some titles than the 1080 even in this mode. And excuse me we both knows it performs in the most important engines.
The market reaction and reviews shows the doom and trashtalk is way overboard.

Its eg a far faster card where it matters and a gazilion times more future proof than the 6970 vs fermi. And futureproof like we probably get 4-5% perf uplift with wolfenstein via fp16 alone. So its tomorrows card.

We also have to add if you share pro and consumer market you also share cost. So this mm2 trashtalk is also way over board. Amd aparently thought the saved dev cost outweight the mm2 and efficiency drawbacks. I have meant for 5 years its a stupid decision having all that compute build in so its good people start to think the same but in all fairness we dont have access to remotely the info amd have. All logic says they are right so its all asumptions from here.

 

[krumme]

People keep using that deflection, like Vega was set in stone 5 years ago before Raja joined (4 and half years ago).

Jim Keller was only at AMD for about 4 years, and we got Zen.

You can't pretend Raja isn't responsible for Vega.

Yes he is. Unfortunately for us success on the highend consumer market is probably not the most important.
Its obvious they need a new gamer oriented arch to be relevant for eg the mobile market.
Remember guys like raja and keller set the direction and chooses witch seniors have most influence on the design. Like mike clark on the ryzen. Perhaps its people like him needed for better arch.

 

[CatMerc]

Raja's involvement in the project would tell us nothing about how much of what happened is his fault.

Would you blame Jensen for not being able to produce a leading GPU with a shoestring and a paperclip?

Unless we know RTG's budget and staffing concerns, and if Raja actively tried to fix things but was blocked by outside limitations, then we can't know if any of this is his fault.


Now can we please stop trying to make management calls with not even half the information needed to make an informed call? Thanks.