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

[Arachnotronic]

Jeez, overclocking the Asus adds another 140W+ of power consumption. 450W of board power to beat a stock GTX 1080.

Wow. Pathetic.

 

[zinfamous]

First AIB review I believe, for the Asus RX Vega 64 Strix Gaming:

http://hexus.net/tech/reviews/graphics/109078-asus-radeon-rx-vega-64-strix-gaming/

Hmm, I wonder if the 56 will see an even greater improvement with better cooling, since it is way more efficient overall. If it manages a +8-10% boost from a tricooler or even liquid, that would be an excellent card.

 

[zinfamous]

Jeez, overclocking the Asus adds another 140W+ of power consumption. 450W of board power to beat a stock GTX 1080.

lol, that is just awful.

 

[Despoiler]

lol, that is just awful.

IKR! Oh Asus what are you doing?

 

[insertcarehere]

IKR! Oh Asus what are you doing?

I don't see how it's Asus's fault that Vega 64 was already well off the performance-efficiency curve to begin with...

 

[parvadomus]

I can only see one positive point about the Vega architecture: it requires less memory bandwidth than their previous architectures.
I expect a really good basic gaming APU based on DDR4 + Vega.
About Vega10 die: its a flop, its even worse than Fiji, performance per watt had a regression, instead of improving.

 

[Rifter]

I don't see how it's Asus's fault that Vega 64 was already well off the performance-efficiency curve to begin with...

Exactly, Asus is just polishing a turd, not there fault its a turd to begin with.

 

[crisium]

Asus Strix Fury actually factory undervolted their card. The only AIB to do so, and it make it remarkably close to GTX 980 in efficiency.

https://tpucdn.com/reviews/ASUS/R9_Fury_Strix/images/perfwatt_3840.gif

IIRC, HardOCP also showed nearly the same efficiency (which is why that review was torn apart here for the contradictions).

I don't expect quite the same here, but hopefully final version tunes the voltage.

I still find the "Power Save" 2nd bios on the TPU review interesting for Vega 64:
https://tpucdn.com/reviews/AMD/Radeon_RX_Vega_64/images/perfwatt_1920_1080.png

It's still very far away, but I think a winning AIB design could consider a "Green Edition" that's undervolted and maybe even underclocked slightly. Sure a user will be able to do this themselves when they buy the Strix Vega 64, but an out of box very efficient (relatively) Vega would look nice on efficiency charts. I guess the Nano will be like this, but that will be louder from the small single fan design than if we got a Strix like this.

But yeah, Vega 64 just has no business hitting 1900mhz, and imo even 1600mhz.

 

[Despoiler]

I don't see how it's Asus's fault that Vega 64 was already well off the performance-efficiency curve to begin with...

See crisium's post. Asus undervolted their Fury cards and it doesn't look like they bothered here even though it's been shown that Vega responds well to it. Also, Asus cooler designs tend to be half arsed on AMD cards. Reused parts from their Nvidia cards that aren't optimized for coverage of the AMD chip. We will have to wait for a tear down to see if they've improved.

 

[EXCellR8]

oh wow, that's a ton of power for a card that already requires a heap of energy to run non-OC'd.

I'd love to see the 64x2 with a good OC. If you want to see my prediction of that just google "Ukraine power plant accident 1986"

 

[NomanA]

Jeez, overclocking the Asus adds another 140W+ of power consumption. 450W of board power to beat a stock GTX 1080.

To clarify for those who didn't read the article, ASUS ROG equipped system uses 455W of *system* power with its factory clocks and BIOS with higher allowed TDP. The card runs about 5-6% faster than the reference Vega 64.

They also overclocked the core in a separate test to 1980MHz, which must be well beyond the reasonable efficiency curve. They gained additional performance but the system power usage increased to 530W.

In any case, 455W of stock Asus Vega 64 system power usage was about 40W higher than the reference design equipped system.

Vega 56 with this type of cooler and without a miner tax will be an awesome card. A Sapphire or Gigabyte Gaming version will probably have better coolers too. If only, they can be found for $400 or under. but that doesn't look likely in the next 3-5 months.

 

[antihelten]

Jeez, overclocking the Asus adds another 140W+ of power consumption.

No it doesn't. It goes from 455W without OC, to 532W with OC. So an increase of 77W, not 140W+

450W of board power to beat a stock GTX 1080.

System power, not board power.

Still really bad, but not quite that bad (you can probably subtract 100W or so for the rest of the system).

Edit: ninja'ed by NomanA above

 

[Phynaz]

No it doesn't. It goes from 455W without OC, to 532W with OC. So an increase of 77W, not 140W+

I failed at math. It's 121W additional over a reference card.

Card is still pulling 430W+

 

[Guru]

Guys what do you think will happen with Raja? He's been doing really great on the software side of things, AMD really have a full packaging that is actually better what Nvidia is currently offering, their open source and semi open source features and code is also great, their driver updates are frequent and with really good game optimizations.

That said in terms of hardware its been shambles, the RX 400/500 series were decent, certainly the RX 400 were saved by board partners who made the card competitive and with the RX 500 series they basically released the cards that were supposed to be released before, but couldn't manage.

No one can recommend the GTX 1060 6GB over the RX 580 8GB, the RX 570 is absolutely king of the mid to low range(of course I'm talking pre mining craze and prices and availability going bonkers).

Vega is an utter disaster, at computing its really good, Nvidia just released a driver that increases Titan xp performance on average about 5% in various programs, in Maya in certain tasks up to 60% faster, etc... Vega still has an advantage, certainly its priced better, but not a clear cut winner at all.

 

[Magic Hate Ball]

Guys what do you think will happen with Raja? He's been doing really great on the software side of things, AMD really have a full packaging that is actually better what Nvidia is currently offering, their open source and semi open source features and code is also great, their driver updates are frequent and with really good game optimizations.

That said in terms of hardware its been shambles, the RX 400/500 series were decent, certainly the RX 400 were saved by board partners who made the card competitive and with the RX 500 series they basically released the cards that were supposed to be released before, but couldn't manage.

No one can recommend the GTX 1060 6GB over the RX 580 8GB, the RX 570 is absolutely king of the mid to low range(of course I'm talking pre mining craze and prices and availability going bonkers).

Vega is an utter disaster, at computing its really good, Nvidia just released a driver that increases Titan xp performance on average about 5% in various programs, in Maya in certain tasks up to 60% faster, etc... Vega still has an advantage, certainly its priced better, but not a clear cut winner at all.

Vega is tailored for the high margin markets of workstation and compute. It's a jack of all trades design.

Gaming suffers, but hopefully AMD's margins improve with the MI25 and WX9100 products doing well in their respective markets, especially in their bang for buck proposition.

More money means more R&D budget.

 

[raghu78]

I can only see one positive point about the Vega architecture: it requires less memory bandwidth than their previous architectures.
I expect a really good basic gaming APU based on DDR4 + Vega.
About Vega10 die: its a flop, its even worse than Fiji, performance per watt had a regression, instead of improving.

Thats correct. Vega 10 regressed in perf/watt and perf/sq mm compared to Fiji. Basically AMD just wasted all the advantages of a full process node jump with some bizarre engineering. Even in professional workloads Vega 10 will lose to GP102 if you compare P6000 with Quadro drivers against WX9100 with Firepro drivers. All round fail.

 

[Rifter]

Guys what do you think will happen with Raja? He's been doing really great on the software side of things, AMD really have a full packaging that is actually better what Nvidia is currently offering, their open source and semi open source features and code is also great, their driver updates are frequent and with really good game optimizations.

That said in terms of hardware its been shambles, the RX 400/500 series were decent, certainly the RX 400 were saved by board partners who made the card competitive and with the RX 500 series they basically released the cards that were supposed to be released before, but couldn't manage.

No one can recommend the GTX 1060 6GB over the RX 580 8GB, the RX 570 is absolutely king of the mid to low range(of course I'm talking pre mining craze and prices and availability going bonkers).

Vega is an utter disaster, at computing its really good, Nvidia just released a driver that increases Titan xp performance on average about 5% in various programs, in Maya in certain tasks up to 60% faster, etc... Vega still has an advantage, certainly its priced better, but not a clear cut winner at all.

Hopefully he gets canned, because if navi is as bad as vega Nvidia is going to have 95%+ of the gaming market share by 2019-2020.

Vega is going to get murdered by volta, by this time next year AMD is not going to have a hope vs 2060/2070 let alone the flagship 2080/2080Ti, navi will be the only glimmer of hope, it absolutely has to deliver.

 

[tviceman]

Hopefully he gets canned, because if navi is as bad as vega Nvidia is going to have 95%+ of the gaming market share by 2019-2020.

Vega is going to get murdered by volta, by this time next year AMD is not going to have a hope vs 2060/2070 let alone the flagship 2080/2080Ti, navi will be the only glimmer of hope, it absolutely has to deliver.

Lets just face reality: They aren't going to close the gap entirely with Navi vs. Volta. It just isn't going to happen. Polaris and Vega had fairly steep mountains to climb to catch Pascal, but instead of clawing back perf/w, the gap widened. So rather than saying "catch up to" Volta, or hoping that Navi "delivers," AMD just needs to move in the right direction.

Unless Volta is horrible (which it won't be, we've already seen Nvidia improve perf/w by 50% with GV100 over GP100) and Navi is a miracle (again, it won't be), the best we should honestly hope to get is that instead of a further widening gap in perf/w, AMD is able to close some of the gap. AMD getting more competitive in perf/w means that they won't have to push all their chips to the limit at stock speeds AND they can reach more notebooks, SFF machines, and avoid the horrible stigma of having a 345w GPU that competes with a 180 watt GPU.

 

[antihelten]

I failed at math. It's 121W additional over a reference card.

Card is still pulling 430W+

Something notable is going on with Hexus number though. In the original review the reference Vega 64 used 352W under load, and now all of a sudden it uses 411W, a rather significant 59W more.

The test notes indicate that the original review was done with a 6700K at 4.4GHz, whereas the Strix review was done with the 6700K at 4.6GHz or 200MHz more. As such the 59W extra can probably be ascribed to this higher overclock, and without it the Strix card would likely be using a little under 400W for the system, and thus somewhere around 300W for the board (and somewhere around 350-400W when overclocked).

Hopefully he gets canned, because if navi is as bad as vega Nvidia is going to have 95%+ of the gaming market share by 2019-2020.

Vega is going to get murdered by volta, by this time next year AMD is not going to have a hope vs 2060/2070 let alone the flagship 2080/2080Ti, navi will be the only glimmer of hope, it absolutely has to deliver.

Canning Raja isn't going to help if the Vega debacle is due to a lack of resources devoted to RTG from AMD.

Also I don't really see what the problem is with Vega getting murdered by Volta. Pascal will also be murdered by Volta. Yes I know that AMD marketing was saying stuff like "poor Volta", but Vega was quite clearly always intended to compete with Pascal (the original plan was probably GP102 and not GP104 though), and Navi against Volta.

With that being said Navi absolutely needs to do better than Vega, if AMD hopes to maintain any kind of market share.

 

[ozzy702]

Lets just face reality: They aren't going to close the gap entirely with Navi vs. Volta. It just isn't going to happen. Polaris and Vega had fairly steep mountains to climb to catch Pascal, but instead of clawing back perf/w, the gap widened. So rather than saying "catch up to" Volta, or hoping that Navi "delivers," AMD just needs to move in the right direction.

Unless Volta is horrible (which it won't be, we've already seen Nvidia improve perf/w by 50% with GV100 over GP100) and Navi is a miracle (again, it won't be), the best we should honestly hope to get is that instead of a further widening gap in perf/w, AMD is able to close some of the gap. AMD getting more competitive in perf/w means that they won't have to push all their chips to the limit at stock speeds AND they can reach more notebooks, SFF machines, and avoid the horrible stigma of having a 345w GPU that competes with a 180 watt GPU.

All this. The gap will be widened between Volta and Navi. NVIDIA is just a well oiled machine and has a lot more resources to draw from. AMD doesn't have the resources and is stuck using Global Floundaries.

 

[Rifter]

Lets just face reality: They aren't going to close the gap entirely with Navi vs. Volta. It just isn't going to happen. Polaris and Vega had fairly steep mountains to climb to catch Pascal, but instead of clawing back perf/w, the gap widened. So rather than saying "catch up to" Volta, or hoping that Navi "delivers," AMD just needs to move in the right direction.

Unless Volta is horrible (which it won't be, we've already seen Nvidia improve perf/w by 50% with GV100 over GP100) and Navi is a miracle (again, it won't be), the best we should honestly hope to get is that instead of a further widening gap in perf/w, AMD is able to close some of the gap. AMD getting more competitive in perf/w means that they won't have to push all their chips to the limit at stock speeds AND they can reach more notebooks, SFF machines, and avoid the horrible stigma of having a 345w GPU that competes with a 180 watt GPU.

Dont get me wrong i dont think they are going to completely close the gap. But if they dont start making serious gains soon they are doomed.

If navi flops im not sure how much longer AMD will be able to float there GPU division. Once Nvidia hits 90-95% of the market share everyone is going to not give a crap about AMD's GPUs, no one will optimize for them, vulkan will die off, and any chance of getting market share back will be very much a uphill battle more so than it is now, at least now they have some dev support. Why would you optimize for only 5% of the market it makes zero economical sense. AMD needs to retain at least 20% market share to at least have a hope of staying in the game.

 

[CatMerc]

Lets just face reality: They aren't going to close the gap entirely with Navi vs. Volta. It just isn't going to happen. Polaris and Vega had fairly steep mountains to climb to catch Pascal, but instead of clawing back perf/w, the gap widened. So rather than saying "catch up to" Volta, or hoping that Navi "delivers," AMD just needs to move in the right direction.

Unless Volta is horrible (which it won't be, we've already seen Nvidia improve perf/w by 50% with GV100 over GP100) and Navi is a miracle (again, it won't be), the best we should honestly hope to get is that instead of a further widening gap in perf/w, AMD is able to close some of the gap. AMD getting more competitive in perf/w means that they won't have to push all their chips to the limit at stock speeds AND they can reach more notebooks, SFF machines, and avoid the horrible stigma of having a 345w GPU that competes with a 180 watt GPU.

Vega closed some of the gap, it's just the equivalent of the RX 580 right now. Pushed to the limit.

In power saver 2nd BIOS, the card is at the heels of Pascal in efficiency and a good bit above 400 series Polaris (the none pushed to the limit ones).

20% more perf/watt than the best Polaris has to offer.

 

[Qwertilot]

You have to be a bit careful with this undervolting/power saving stuff and things because you're then getting vaguely near having to compare to the laptop variants of Pascal.

Do wish someone could persuade AMD to ship cards with halfway sane stock settings though! The small bit of extra performance really, really isn't worth this much power.

 

[mohit9206]

Hmm, I wonder if the 56 will see an even greater improvement with better cooling, since it is way more efficient overall. If it manages a +8-10% boost from a tricooler or even liquid, that would be an excellent card.

There's no such thing as excellent or terrible card just excellent and terrible prices and at $500 (actual non bait and switch msrp), Vega 56 is opposite of excellent

 

[EXCellR8]

Radeon RX Vega 64 demand continues to exceed expectations. AMD is working closely with its partners to address this demand. Our initial launch quantities included standalone Radeon RX Vega 64 at SEP of $499, Radeon RX Vega 64 Black Packs at SEP of $599, and Radeon RX Vega 64 Aqua Packs at SEP of $699. We are working with our partners to restock all SKUs of Radeon RX Vega 64 including the standalone cards and Gamer Packs over the next few weeks, and you should expect quantities of Vega to start arriving in the coming days

literally an hour of browsing online web boards and gaming communities would have revealed the nature of the demand that is "exceeding expectations..."