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

[Vesku]

Hope AMD is dedicating enough resources for Navi to see a big enough improvement in performance/transistor. Something went very wrong with Vega, a die shrunk Fury would have been out a year ago and been similar to Vega56 gaming performance.

Same with a Vega-sized Polaris with GDDR5X. That might have even turned out better for gaming performance than full Vega.

Whatever RTG has been doing with Vega development over the last year, it better have provided useful institutional knowledge and experience. Simply so we don't see a defacto gaming GPU monopoly in the near future.

 

[tential]

What gaming extras did rx Vega have over founders edition....

 

[Arachnotronic]

Hope AMD is dedicating enough resources for Navi to see a big enough improvement in performance/transistor. Something went very wrong with Vega, a die shrunk Fury would have been out a year ago and been similar to Vega56 gaming performance.

Same with a Vega-sized Polaris with GDDR5X. That might have even turned out better for gaming performance than full Vega.

Whatever RTG has been doing with Vega development over the last year, it better have provided useful institutional knowledge and experience. Simply so we don't see a defacto gaming GPU monopoly in the near future.

What likely happened is that for AMD to get to the kinds of clocks that Vega could achieve on a shoestring budget, the physical design was relatively sloppy -- delivering the required performance but at the cost of great power consumption.

Hope those people who derided Pascal as "Paxwell" now come to understand just how hard it is to get a big frequency increase at good power consumption.

 

[DeathReborn]

GTX 1030 level, playing Doom in Vulcan is the best that you can realistically expect.

It's the codec support (4K Netflix would be nice) more than anything, I have other machines for gaming. An APU is what I have always tried to use for HTPC since they first launched but always ended up with a discrete card too, hopefully no more for that.

 

[Arachnotronic]

In my opinion AMD needs to revert back to the small-die strategy that made the RV770 possible back in the day and make perf/watt their biggest priority. GCN blown up to the size of Fiji and Vega scales very poorly as we have seen from their respective performance. The problem is that the people who built the RV770 - Carrell Killebrew et al., have been laid off. I'm not hopeful that the current RTG leadership can execute with respect to these parameters in a timely fashion.

They weren't laid off -- they left for greener pastures.

http://www.fudzilla.com/25956-amd-graphics-cto-leaves-company

 

[JoeRambo]

If someone told me a few months ago that an out of the box 14nm FinFET Vega 64 delivers worse performance per watt than a 2.5 years old 28nm GTX 980 Ti, I'd call that person crazy

In fact in January there were "leaks" about Vega having 4x power efficiency. I remember those rumors sending certain forum AMD dervishes into frenzy
The scope of AMD failure with Vega is just legendary, book and movie worthy, competing only with Bulldozer.

Hope those people who derided Pascal as "Paxwell" now come to understand just how hard it is to get a big frequency increase at good power consumption.

Good one, haha, AMD would kill for Fury @ 2Ghz at 1080TI wattage

 

[IntelUser2000]

Breaking a GPU into multiple chips, really only benefits chip yield, it would actually increase power consumption and decrease performance so this really isn't a performance solution, it's a chip yield solution, and the problem here is NOT chip yield.

They may have issues with HBM2 yield at those clocks.

Last year's GP100 Pascal had HBM2 operating at 1.4GT/s. This year's GV100 raises it to 1.75GT/s. Vega pushes that even further to 1.89GT/s.

HBM2 was promised to reach 2GT/s. Failure to reach clocks point to production and yield issues. The out of stock notifications in stores are likely due to low supply as well.

My take on your post directly is though, the team may be more experienced in building smaller dies. It could work out better for them. They have been building only smaller dies for years. They avoid large dies on CPUs too.

What gaming extras did rx Vega have over founders edition....

GamersNexus say because RX Vegas are re-spun silicons it has DSBR enabled.

 

[Vesku]

What likely happened is that for AMD to get to the kinds of clocks that Vega could achieve on a shoestring budget, the physical design was relatively sloppy -- delivering the required performance but at the cost of great power consumption.

Hope those people who derided Pascal as "Paxwell" now come to understand just how hard it is to get a big frequency increase at good power consumption.

Die shrunk Fury X would see some "free" clock increase, shifts the MHz curve about 20% iirc the evaluation of 28nm to 16/14nm.

Look at Polaris, within 10% of Vega in terms of optimal MHz. Polaris did not need a hefty chunk of "clocking" die budget to get there. Something else is going on with Vega transistor budget and it's not doing much for gaming performance.

 

[tamz_msc]

They weren't laid off -- they left for greener pastures.

http://www.fudzilla.com/25956-amd-graphics-cto-leaves-company

Charlie@SA says AMD "cut" 10% of their workforce, rumors have it that Carrell Killebrew was fired and Eric Demers left on his own. Then you have that saga involving AMD accusing ex-engineers of stealing confidential information and giving it to NVIDIA, and Carrell being very upset over these accusations against his former colleagues on the Ars Techinca forums.

Getting back to the point: AMD desperately needs another RV770 in order to have any chance of being competitive again.

 

[Face2Face]

I wouldn't mind seeing how well a 1500MHz GTX 980Ti stacks up to VEGA. Performance and power consumption. Wish I still had mine...

 

[UglyDuckling]

Threadcrapping not allowed.



esquared
Anandtech Forum Director

 

[Arachnotronic]

Getting back to the point: AMD desperately needs another RV770 in order to have any chance of being competitive again.

Polaris is doing OK against the 1060 in terms of performance, though it is behind in efficiency. Hasn't stopped it from selling well, though.

 

[Azix]

Breaking a GPU into multiple chips, really only benefits chip yield, it would actually increase power consumption and decrease performance so this really isn't a performance solution, it's a chip yield solution, and the problem here is NOT chip yield.

there is a benefit. vega 56 TGP is 165W. the PCB puts it at ~220W. sharing a PCB could net you almost 2 x vega 56 for close to vega 64 LC TDP. It's a way around limitations because a chip as powerful as that can't be had with vega 64 or reasonably.

 

[zinfamous]

Yeah, I agree that there is a value in pushing the cards hard for the halo effect. But (1) I don't think that AMD targeted 1080-level performance when they designed Vega or, another way to put it, I don't think AMD planned for Vega to be merely 30% faster than two year old Fiji and (2) I don't think AMD wanted to bring out a product for which they had no choice but to push balls-to-the-wall on the power limit right out off the bat.

These factors lead me to believe that AMD targeted a more efficient, higher performing part but fell short by 20-30% and 50 to 100W.

I go back to the earnings call earlier in July when Su (I think it was Su), reiterated that the design focus for RTG had been shifted towards enterprise more and more and on this front, AMD is very, very competitive with performance and even more so with price. From the purported focus of what GCN/Vega is supposed to be in relation to AMD's market goals, it can be fairly argued that Vega is a success.

The problem with AMD here really is about marketing. AMD should have just toned down the marketing for Vega as it relates to gamers, rather than make nonsense ads like "Poor Volta." That was clearly very stupid...unless they were specifically referring to Volta in the enterprise space (which could only be a price/performance issue, imo)...but that wasn't really an enterprise-directed ad. It looked like it was aimed at gamers, to me anyway. That being said, it's not that RX Vega is bad in a vacuum, it's just that AMD is not competing in a vacuum. This would have been a decent release a year ago (still getting crushed with efficiency, though), but it is just very late to be meagerly competing with a mid/high-end card like 1080.

I agree that RX 56 should have been the gaming card, and save the full dies for enterprise where margins would have been much better....but I suppose there is a segment need where AMD wants to claim that it can offer options in the higher end. It just looks like a very poor effort to me. Obviously these are great cards/options for the many numbers of people that don't have something in this performance range (I will be getting an AIB 56 some time in September-November...I guess), but it appears very, very late and a bit too short. Vega seems like more of an enterprise design on the high-end/margin side of the offering, and a low-end-mid APU offering on the HPC/console space. I fully expect Raven Ridge chips to monsters within their class, so it very well could be that the actual design and market goals for Vega are very different than what consumers expected (and to no small fault of AMD's own marketing campaigns, obviously)

 

[Elixer]

This launch (Vega 64) & (yet another) paper launch (Vega 56) is a big fat wart on AMD's rep.

The longest marking campaign in AMD's history, turns out to be a mediocre product at best.
It was full of smoke & mirrors, dishonesty ("Poor Volta"), and just horrible execution caused by delays.

The delays are clearly because of HBM2, and they STILL haven't been able to source chips that meet JEDEC's standard of 2Gbps & 1.2V (min) - 1.26V(max).

The design of Vega was clearly not prepared for the 1080, let alone for the 1080ti.
AMD was forced to throw as much power as they could just to "trade blows" with the 1080, and they won't have an answer for the 1080ti for another year or two?
That in itself is an epic disaster for the very high end.

The 'Vega FE is a PRO card, not for gaming' pretty much proves that in fact, the performance scales almost exactly the same, once everything is equal.
$999 for gimped drivers. Seriously AMD? sigh.

The price point is just not there, the blower cards are $40 too much. The AIB cards should be at $399 & $499, and instead will have an even worse price, and that price will move them out of any "Best bang for the buck" category.
In fact, I would argue that AMD should have never released either of these cards, and instead, let the AIB's handle the launch. The "Too hot, Too loud" meme of the past strikes yet again.

It all boils down to, those that have a Freesync monitor, their only upgrade option is Vega, unless they want to buy a new monitor as well.

People on the sidelines might consider Vega 56, unless Nvidia does price cuts or throws in a few free games, and then AMD will be basically suck between a rock & a hard place.
Once they see that the AIB cards will be $40+ than the MSRP, then, I see very few people buying them.

Things that are meaningless as of today:
"Give AMD time to fix the drivers, fine wine!" --They should have priced that into the equation first then. If the product is unfinished, then price it that way.
"Don't worry, they are working on Navi!" -- See you in 2 years?

 

[tamz_msc]

Polaris is doing OK against the 1060 in terms of performance, though it is behind in efficiency. Hasn't stopped it from selling well, though.

And it wold have done similarly against a GTX 1070 if AMD made a full fat 320mm^2 die with 44 CUs a-la Scorpio. They don't seem to realize that Hawaii has by far been their most balanced configuration, and not making it into a Polaris-based a consumer GPU was a mistake.

 

[Arachnotronic]

And it wold have done similarly against a GTX 1070 if AMD made a full fat 320mm^2 die with 44 CUs a-la Scorpio. They don't seem to realize that Hawaii has by far been their most balanced configuration, and not making it into a Polaris-based a consumer GPU was a mistake.

AMD flew too close to the sun, and now their star Vega has gone nova.

 

[Fallen Kell]

The only good thing Vega seems to accomplish is that it will sell decently until the cryptominers have a better solution.

For gaming, this platform is a complete failure, almost to the same scale as Polaris (I say almost because this almost competes against Nvidia's current mid-highend cards like the 1070 and 1080 (not the TI versions)).

Nvidia still has a card from over a year ago (really 2 years in terms of architecture) that destroys Vega, and can do so on air cooling...

 

[Magic Hate Ball]

And it wold have done similarly against a GTX 1070 if AMD made a full fat 320mm^2 die with 44 CUs a-la Scorpio. They don't seem to realize that Hawaii has by far been their most balanced configuration, and not making it into a Polaris-based a consumer GPU was a mistake.

Yeah, that and a 384-bit memory bus would have been beautiful for a just shy of high end card.

 

[tamz_msc]

The only good thing Vega seems to accomplish is that it will sell decently until the cryptominers have a better solution.

For gaming, this platform is a complete failure, almost to the same scale as Polaris (I say almost because this almost competes against Nvidia's current mid-highend cards like the 1070 and 1080 (not the TI versions)).

Nvidia still has a card from over a year ago (really 2 years in terms of architecture) that destroys Vega, and can do so on air cooling...

I don't understand why Polaris was a failure - if an additional 50W power consumption isn't a problem then an RX 580/480 is equal or slightly better than a GTX 1060 6GB. AMD delivered on their promise of giving previous gen 350$+ performance at mainstream 200$ prices.

 

[DisEnchantment]

I go back to the earnings call earlier in July when Su (I think it was Su), reiterated that the design focus for RTG had been shifted towards enterprise more and more and on this front, AMD is very, very competitive with performance and even more so with price. From the purported focus of what GCN/Vega is supposed to be in relation to AMD's market goals, it can be fairly argued that Vega is a success.

The problem with AMD here really is about marketing. AMD should have just toned down the marketing for Vega as it relates to gamers, rather than make nonsense ads like "Poor Volta." That was clearly very stupid...unless they were specifically referring to Volta in the enterprise space (which could only be a price/performance issue, imo)...but that wasn't really an enterprise-directed ad. It looked like it was aimed at gamers, to me anyway. That being said, it's not that RX Vega is bad in a vacuum, it's just that AMD is not competing in a vacuum. This would have been a decent release a year ago (still getting crushed with efficiency, though), but it is just very late to be meagerly competing with a mid/high-end card like 1080.

I agree that RX 56 should have been the gaming card, and save the full dies for enterprise where margins would have been much better....but I suppose there is a segment need where AMD wants to claim that it can offer options in the higher end. It just looks like a very poor effort to me. Obviously these are great cards/options for the many numbers of people that don't have something in this performance range (I will be getting an AIB 56 some time in September-November...I guess), but it appears very, very late and a bit too short. Vega seems like more of an enterprise design on the high-end/margin side of the offering, and a low-end-mid APU offering on the HPC/console space. I fully expect Raven Ridge chips to monsters within their class, so it very well could be that the actual design and market goals for Vega are very different than what consumers expected (and to no small fault of AMD's own marketing campaigns, obviously)

If you remember Raja's statements at FAD, he pretty much stated Vega is back to basics for AMD in Discrete Graphics. The first Chapter of a Book.
His words....

AMD did very well... with what they had.
- Our company has been through a crunch like AMD, 8 years ago. We struggled with debt while trying to maintain R&D budget. We manufactured HW parts pre-dispositioned to to support features we have not made the SW yet. Our product got ripped at launched, but we stayed afloat and survived and extract maximum profit from the HW.
- Having worked with NVIDIA (NVIDIA as a supplier), there are times when I just thought my god how I wish we have an option.

 

[MBrown]

I have a 980ti and probably wont be upgrading for a while (Nvidia or AMD). But I have to say, sometimes these comment sections seem like they are just a competition to see who can write the most elegant and embellished post on how bad a product is. It cracks me up sometimes.

 

[psolord]

My 7950 called Vega and it said "Son,I'm disappoint".

Then my 5850 called. You don't want to know what it said!

 

[raildogg]

I haven't played modern PC games in years and have been waiting months for a good graphics card to be available. A lot of shortages in the past few months as you guys know.

I think the Vega 56 seems like a very nice product though maybe slightly high in power consumption. But maybe a 600W PSU should be enough for it. I'm hoping some will remain in stock when it is introduced to the market. I mean, I was looking at the 580 but those are impossible to find so the 56 is definitely the way to go. I'm mainly looking at the AMD side in graphics because the availability of Freesync monitors seems to be better. This product seems to be it.

 

[Magic Hate Ball]

My 7950 called Vega and it said "Son,I'm disappoint".

Then my 5850 called. You don't want to know what it said!

Man I gave my 5850 to a now ex-girlfriend for her rig 6 years ago.... I want my card back damnit!