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

[Janooo]

It's not this, it's that you're saying there are features that amd just needs to flip a switch for. Or that you make sound easy to have happen. If these were simply things they had to enable in drivers they would have done so.

Amd says these features are on and then we have forum members saying to wait for the large performance gain once the feature is actually on....

It appears you are not up to speed with this thread:
https://forums.anandtech.com/thread...-updated-aug-28.2515492/page-51#post-39054642

 

[Rifter]

Your request does not make any sense.
He said, the features are not enabled. Who can test something that is not enabled?
Then he continued... "if made to work..."
All of us are guessing what NGG might bring in the future. If AMD does not make it happen, I am not excluding that possibility, then it's going to be one big fat zero.

It makes perfect sense, he said "we know for a fact that several performance enhancing features are not operating at present" and i asked him to prove it with independent benchmarks.

he stated as fact that there are several disabled features that are performance enhancing. I just want proof of his statement, as all the reviews i have seen that try and benchmark with any of these features enabled do not have any good things to say about performance. The gamestar.de review of HBCC for example has this to say about it(it is google translated):

"In Doom, however, the fps decline comparatively strongly, on average, it is about nine percent. The title is, however, clearly the exception, on the average of all measurements the performance with activated HBCC decreases instead only by about four to five percent. In the case of the minimal fps, we also arrive at largely identical results; here, too, HBCC does not offer any advantages in the course of our previous tests.

Either way, HBCC does not usually make a big difference to the game performance of Vega. It may be curious whether this changes with increasing VRAM requirements in games or by new drivers in the future - further HBCC measurements at a later date are in any case firmly fixed."

I have no doubt these features are disabled, i do however have doubt that they are all going to be performance enhancing in a meaningful amount.

 

[guachi]

New Tom's review of the V56 including some extensive power testing... if people still care anymore...

http://www.tomshardware.com/reviews/radeon-rx-vega-56,5202.html

Thank your for the link. Lots of good testing. I don't generally read tomshardware any more but it was thorough enough. I'd be interested in seeing an undervolt/overclock test of the 56, but beggars can't be choosers.

The conclusion seems on par with other 56 reviews. It's a good card that's faster than the 1070. Power hungry but possible to get 90% the performance at 70% of the power.

Now, if only you could get them at $400 or get an AIB of any kind.

 

[EXCellR8]

How many units have they moved? are they out of stock due to receiving a token amount of cards or are they out of stock and actually moving a real number of cards?

my guess is that the actual store only received a handful of cards if any and that the online stock was from a regional warehouse but I honestly have no idea. it wouldn't surprise me if the cards that were initially sold off were all spoken for and that the earliest adopters tried to return them as soon as they could. as a retailer they might be able to restock the cards since I HIGHLY doubt RMA's are even being honored with the already low supply. refunding buyers is probably the most efficient method.

a few of the cards were tested upon being returned and worked fine, so draw your own conclusions I suppose.

 

[maddie]

It makes perfect sense, he said "we know for a fact that several performance enhancing features are not operating at present" and i asked him to prove it with independent benchmarks.

he stated as fact that there are several disabled features that are performance enhancing. I just want proof of his statement, as all the reviews i have seen that try and benchmark with any of these features enabled do not have any good things to say about performance. The gamestar.de review of HBCC for example has this to say about it(it is google translated):

"In Doom, however, the fps decline comparatively strongly, on average, it is about nine percent. The title is, however, clearly the exception, on the average of all measurements the performance with activated HBCC decreases instead only by about four to five percent. In the case of the minimal fps, we also arrive at largely identical results; here, too, HBCC does not offer any advantages in the course of our previous tests.

Either way, HBCC does not usually make a big difference to the game performance of Vega. It may be curious whether this changes with increasing VRAM requirements in games or by new drivers in the future - further HBCC measurements at a later date are in any case firmly fixed."

I have no doubt these features are disabled, i do however have doubt that they are all going to be performance enhancing in a meaningful amount.

HBCC should not have any significant impact for data that can fit within the 8GB and it's not representative of DSBR capabilities. No one has tested the capabilities of the DSBR. I'm going on the whitepaper. They aren't implemented presently, so none of it's benefits are available or can be tested by the public. This is a driver level feature. If you believe they're outright lying, then so be it.

One function is this from the whitepaper. AMD has always suffered from poor utilization vs Nvidia. One reason why it took many more AMD shaders to equal GeForce GPUs. This can close the gap. There was an AMD patent discussed here several years ago with variable shader clusters for different sized wavefronts. This appears to have the same end result in packing a full 64 wavefront with different instances instead of breaking up the wavefronts into smaller values.

"Another innovation of “Vega’s” NGG is improved load balancing across multiple geometry engines. An intelligent workload distributor (IWD) continually adjusts pipeline settings based on the characteristics of the draw calls it receives in order to maximize utilization. One factor that can cause geometry engines to idle is context switching. Context switches occur whenever the engine changes from one render state to another, such as when changing from a draw call for one object to that of a different object with different material properties. The amount of data associated with render states can be quite large, and GPU processing can stall if it runs out of available context storage. The IWD seeks to avoid this performance overhead by avoiding context switches whenever possible. Some draw calls also include many small instances (i.e., they render many similar versions of a simple object). If an instance does not include enough primitives to fill a wavefront of 64 threads, then it cannot take full advantage of the GPU’s parallel processing capability, and some proportion of the GPU's capacity goes unused. The IWD can mitigate this effect by packing multiple small instances into a single wavefront, providing a substantial boost to utilization."

With that said, look at this posted earlier. Another site found a HBCC gain.
https://translate.google.tt/transla...ls/HBCC-Gaming-Benchmark-1236099/&prev=search

""Let's look at the benchmarks. We took some titles from the current PCGH-Parcours to the chest and partly used other settings, as for example with The Talos Principle, which we also tried with powerful, but imaginative 4x Supersample-Antialiasing. This was followed by a test from our old Parcours: Metro Last Light Redux (see benchmark how-to ) with our standard settings for the benchmark "The Crossing" - not the integrated performance test, but a real play scene. This was all the more surprising, as the Radeon RX Vega 64 was able to grow by an impressive 11 or 14%, depending on the driver used."

 

[krumme]

It makes perfect sense, he said "we know for a fact that several performance enhancing features are not operating at present" and i asked him to prove it with independent benchmarks.

he stated as fact that there are several disabled features that are performance enhancing. I just want proof of his statement, as all the reviews i have seen that try and benchmark with any of these features enabled do not have any good things to say about performance. The gamestar.de review of HBCC for example has this to say about it(it is google translated):

"In Doom, however, the fps decline comparatively strongly, on average, it is about nine percent. The title is, however, clearly the exception, on the average of all measurements the performance with activated HBCC decreases instead only by about four to five percent. In the case of the minimal fps, we also arrive at largely identical results; here, too, HBCC does not offer any advantages in the course of our previous tests.

Either way, HBCC does not usually make a big difference to the game performance of Vega. It may be curious whether this changes with increasing VRAM requirements in games or by new drivers in the future - further HBCC measurements at a later date are in any case firmly fixed."

I have no doubt these features are disabled, i do however have doubt that they are all going to be performance enhancing in a meaningful amount.

Raja mentioned hbcc as a feature especially interesting on a 5 year horizon.

Zlatan explained it well in the other thread what is needed when hbcc is treated inclusive or exclusive and some data about the granularity.

You will see proof perhaps in later q4 is zlatans guess.
Its mentioned in the footnotes of the white paper that the 4x peak discard rate is results from the 17.3.. prototype driver.

 

[Krteq]

Its mentioned in the footnotes of the white paper that the 4x peak discard rate is results from the 17.3.. prototype driver.

Yep, branch 17.320 ... we have 17.300 currently

 

[LightningZ71]

Just a question, but, do you think that HBCC as implemented on the VEGA (as we believe it to be) has a use case in coin mining? With the blockchains continuing to grow, and with the 2GB cards almost completely unable to achieve any performance on the larger coins and the 4GB cards showing memory pressure, perhaps using HBCC to implement a totally flat logical GPU memory pool can allow GPU cards to still be relevant for mining work when the chains outgrow even the 8GB VRAM pools of the current higher end GPUs? It will be a long time before the mainstream cards are equipped with 16GB+ VRAM, and, given current DRAM prices, those cards may take a long time to come down to volume purchase price levels. An intelligent memory management engine can keep the unused section of the chain moved to the slower storage while continually updating the local storage with the sections that are currently in use in the background, keeping GPU performance at its maximum.

 

[Muhammed]

AMD never mentioning how much performance they expect from certain features = there is no performance to be gained, I though this is straight forward enough. They are in a dire state, looking for any win possible, they could make a PR announcement and promise min X% from feature Y, but they didn't, because there isn't any that is worthwhile.

For example take DSBR, it was disabled with Vega FE driver, AMD said this to Anand:

the DSBR is not enabled in Vega FE’s current drivers. Whereas we have been told to expect it with the RX Vega launch. AMD is being careful not to make too many promises here – the performance and power impact of the DSBR vary wildly with the software used

Ryan Smith also had this to say about it:

From the tone of the conversations I had, while DSBR will improve things, everyone was quick to point out that the gains would be higher on a more resource-constrained card. Those aren't the kind of comments I'd expect if they thought performance would make a huge jump with DSBR.

Fast forward to RX Vega launch, DSBR has been enabled, and no worthwhile gains have been noticed in games, maybe in some Pro apps. But games? Nothing. Same thing for HBCC, and same thing for incoming Primitive Shaders as well. I had a look at the complete reviewers guide from AMD, the guide never once mentions any of these features having any significant impact on games performance. It looks like these features are designed mainly for APUs and mobile GPUs, or any GPU that fits the criteria of resource constrained cards. As for full fledged cards? They are out of luck.

Same ridiculous claims have been said repeatedly about Fiji's Async Compute and how it is going to move it beyond Maxwell, despite AMD not uttering a word about this. And so, in real games the feature never amounted to more than 5~10% at best. And it never changed the landscape of GPUs or market competitiveness. People should just stop clinging to false hopes, when AMD is not giving any.

 

[Konan]

I think after the dust (and prices) settle down, Vega will be seen as a reasonable AMD choice, the 56 model in particular. It looks bad right now because of disappointment VS the competition, availability, delayed release etc etc. But, it is what it is. A capable enough gaming card for 1440p. Hell, plug the card in and it will work. Whatever. Wanted to see better from AMD but whatever man.

Hopefully the prices do settle down. Vega 64 is in stock in my city. There at the time of writing there are 6 available Vs 25+ GTX 1080 Ti's.

MSI Vega 64 $949.99 CAD
GIGABYTE AORUS GTX 1080 Ti $949.99 CAD

easy choice...

The cheapest Vega 56 equates to $520 USD for the XFX version. The Sapphire version equates to $600 USD. How much of a mark up!

 

[Grubbernaught]

I still take issue with Vega being released without key features enabled.

If primitive shaders really do provide substantial gains (and I hope they do, I want to buy for my second rig) then surely the product should have been delayed as first impressions count and you only get one chance to make them. Take the 290x blower fiasco for example.

I understand the release was factored into the earnings forecast promulgated to investors but surely that does not legally bind them to release a product that does not function as advertised or intended.

Further, people have paid good money for these cards and deserve a straight up answer as to what isn't enabled, why and a timeline for rectification. Perhaps even a beta toggle to turn on what they have right now with the caveat that it will impact stability.

Working silicon has existed for quite a long time now and the fact these features still remain disabled is not a good omen.

I for one would feel much more comfortable laying down bucks for Vega if all the proverbial cards were on the table and knew what I was getting into.

Say what you will about NV, when they release, the card does what they say it it is going to do straight out of the box (3.5 + .5 Gb 970s not so much).

This would be classed as deceptive and unacceptable in any other industry, why is it ok here?

**End rant. Sorry if this comes off snakey but I've just finished 8x12h night shifts and corporate behavior like this chaps my ass at the best of times....

 

[maddie]

I still take issue with Vega being released without key features enabled.

If primitive shaders really do provide substantial gains (and I hope they do, I want to buy for my second rig) then surely the product should have been delayed as first impressions count and you only get one chance to make them. Take the 290x blower fiasco for example.

I understand the release was factored into the earnings forecast promulgated to investors but surely that does not legally bind them to release a product that does not function as advertised or intended.

Further, people have paid good money for these cards and deserve a straight up answer as to what isn't enabled, why and a timeline for rectification. Perhaps even a beta toggle to turn on what they have right now with the caveat that it will impact stability.

Working silicon has existed for quite a long time now and the fact these features still remain disabled is not a good omen.

I for one would feel much more comfortable laying down bucks for Vega if all the proverbial cards were on the table and knew what I was getting into.

Say what you will about NV, when they release, the card does what they say it it is going to do straight out of the box (3.5 + .5 Gb 970s not so much).

This would be classed as deceptive and unacceptable in any other industry, why is it ok here?

**End rant. Sorry if this comes off snakey but I've just finished 8x12h night shifts and corporate behavior like this chaps my ass at the best of times....

Don't mistake some enthusiasts talking tech & possibilities with AMD marketing to users. What exactly did AMD promise you and not delivered this time? You have a card providing X performance at $Y. Worth it or not? Always buy a product on your opinion of it. If you're unhappy with it, don't buy it. For certain, there will be some, myself included, who think there is a good chance of unlocked performance happening in the near future, and others who believe that it all a wash and no great gains expected. So what. I don't get why everyone has to agree, just that we don't lie to score points.

Look at the mining phenomenon. Cards that were acceptable for a certain price are now seen as hugely expensive at mining prices. All products are like that. If you find the present performance of Vega is too low for the price, ignore it. There are some getting upset and saying that I for one, am telling members to wait for increased performance. Here is it folks.

Buy whatever you want . Wait if you want. Don't wait if you think it's a mistake. For sure somebody will be wrong and the other side correct, or maybe a mixture. Just please yourself.

The FE editions are performing adequately for that market and do seem to have satisfied customers. Its the gamers doing all the grumbling. My personal view is that they are struggling to get all driver features ready. Primitive shaders for one and the power management a second.

 

[tential]

Don't mistake some enthusiasts talking tech & possibilities with AMD marketing to users. What exactly did AMD promise you and not delivered this time? You have a card providing X performance at $Y. Worth it or not? Always buy a product on your opinion of it. If you're unhappy with it, don't buy it. For certain, there will be some, myself included, who think there is a good chance of unlocked performance happening in the near future, and others who believe that it all a wash and no great gains expected. So what. I don't get why everyone has to agree, just that we don't lie to score points.

Look at the mining phenomenon. Cards that were acceptable for a certain price are now seen as hugely expensive at mining prices. All products are like that. If you find the present performance of Vega is too low for the price, ignore it. There are some getting upset and saying that I for one, am telling members to wait for increased performance. Here is it folks.

Buy whatever you want . Wait if you want. Don't wait if you think it's a mistake. For sure somebody will be wrong and the other side correct, or maybe a mixture. Just please yourself.

The FE editions are performing adequately for that market and do seem to have satisfied customers. Its the gamers doing all the grumbling. My personal view is that they are struggling to get all driver features ready. Primitive shaders for one and the power management a second.

You'll make any justification necessary for AMD's actions. You don't have to defend everything they do...
Generally people like and dislike aspects of something. They don't just defend every single thing about something...

You never bothered to address the subject of amd not enabling features you just talked around it and gave a mini pitch for why people will buy Vega.

Great. Doesn't change the fact that amd launches with features disabled with 0 time frame on when it will be enabled after taking it up.

Using "if you don't like it don't buy it!" is not furthering discussion on this forum. It's a way to shut down the conversation and subvert it into a conversation on whether or not you'll buy the product rather than addressing the actual concern at hand.

 

[beginner99]

Major features turned off at launch is not typical. We should expect not typical increases in performance.

Do you really believe that the driver can unlock 20%+ performance and AMD is saying jack shit about this? Common. If they knew it's possible, they would be talking about it. Since they are not it's either not possible at all or very unclear if and when such a driver will release (sure not this year).

 

[maddie]

You'll make any justification necessary for AMD's actions. You don't have to defend everything they do...
Generally people like and dislike aspects of something. They don't just defend every single thing about something...

You never bothered to address the subject of amd not enabling features you just talked around it and gave a mini pitch for why people will buy Vega.

Great. Doesn't change the fact that amd launches with features disabled with 0 time frame on when it will be enabled after taking it up.

Using "if you don't like it don't buy it!" is not furthering discussion on this forum. It's a way to shut down the conversation and subvert it into a conversation on whether or not you'll buy the product rather than addressing the actual concern at hand.

This has been a long thread You appear to jump in a thread and not realize that there have been many posts earlier saying many things. I have stated several times that AMD has screwed up the launch. What exactly do you want? An exact word or phrase that would satisfy you? Are you upset that someone has a different opinion? I'm just tired of all the sniping. A recent prolific poster here got very aggressive and made the general mood very nasty. I've had enough of that.

 

[geoxile]

Same ridiculous claims have been said repeatedly about Fiji's Async Compute and how it is going to move it beyond Maxwell, despite AMD not uttering a word about this. And so, in real games the feature never amounted to more than 5~10% at best. And it never changed the landscape of GPUs or market competitiveness. People should just stop clinging to false hopes, when AMD is not giving any.

The async compute claims aren't so ridiculous, it's just a little out of context. The first time Asynchronous Computation was used was by PS4 devs, who, afaik implemented it themselves since the PS4 uses some sort of very low-level openGL derived API designed by Sony. The same concept was standardized in the mantle API specification and DX12 and Vulkan inherited it.

http://fumufumu.q-games.com/archives/TheTechnologyOfTomorrowsChildrenFinal.pdf

Their result is an 18% improvement, which is nothing to sneeze at. But they note that it's saved on shaders. Considering GCN cards on the PC can't even feed the shaders properly I don't think we'll ever see Async Compute be as big a boon as it was for these PS4 devs unless AMD fixes the issues. That said the improved scheduler in NGG sounds interesting. Full Vega has ~13TF worth of ALUs, if AMD could feed it all it'd be a huge boost in performance and with compute being a bottleneck async compute might return bigger improvements too. But right now it's clear Nvidia's by far much better at putting those shaders to work.

 

[Elixer]

I think pretty much everything possible has been said in this thread, minds are pretty much made up one way or the other.

About the only thing left is to wait for AIB cards, and see how they slot into this Vega puzzle.
I am actually curious if the AIB's will make 'mining' cards with Vega, like they did with Polaris.
Unless AMD's partners having been hording chips, instead of selling the reference cards, we are looking at December before we will see ample stock.

 

[Reinvented]

I think pretty much everything possible has been said in this thread, minds are pretty much made up one way or the other.

About the only thing left is to wait for AIB cards, and see how they slot into this Vega puzzle.
I am actually curious if the AIB's will make 'mining' cards with Vega, like they did with Polaris.
Unless AMD's partners having been hording chips, instead of selling the reference cards, we are looking at December before we will see ample stock.

Honestly, I'm waiting for the AIB cards. Not sure why people are paying a premium for stock/reference cards. The AIB cards that are custom tuned by the partners are leaps and bounds better. Hopefully they will come at a reasonable price, if not a bit more than MSRP. However, in it's current state I do not see that happening.

Also, with the shortages on memory, I do not see them catering to miners this time around. At least at this price point, and with this architecture.

 

[Krteq]

Fast forward to RX Vega launch, DSBR has been enabled...

Nope, DSBR is NOT enabled yet.

 

[krumme]

This is Kanter explaining tile based rendering in Maxwell.
http://www.realworldtech.com/tile-based-rasterization-nvidia-gpus/


This is from Vega whitepaper about the DSBR for the pixel engine:

And if we look at the missing HBM2 pseudochannel this is a bit about the effect for the memory systems:
http://www.semicontaiwan.org/zh/sit..._taiwan_2015_ppt_template_sk_hynix_hbm_r5.pdf

And to give the full picture of major disabled features - the geometry utilization we talked about:

 

[Magic Hate Ball]

This is Kanter explaining tile based rendering in Maxwell.
http://www.realworldtech.com/tile-based-rasterization-nvidia-gpus/


This is from Vega whitepaper about the DSBR for the pixel engine:

And if we look at the missing HBM2 pseudochannel this is a bit about the effect for the memory systems:
http://www.semicontaiwan.org/zh/sit..._taiwan_2015_ppt_template_sk_hynix_hbm_r5.pdf

And to give the full picture of major disabled features - the geometry utilization we talked about:

Even if Vega were able to real-world utilize 25% of the potential boost from this NGG feature, that'd be significant in price/performance.

 

[Despoiler]

Vega handily beating a 1080TI OC in PUBG @ 1080p, but then falls way off when the resolution goes up. Very odd results. General driver issues or Vega features being disabled showing where it bottlenecks?

http://www.pcgameshardware.de/Playe...s-Battlegrounds-neue-Benchmarks-Vega-1236260/

 

[Malogeek]

Vega handily beating a 1080TI OC in PUBG @ 1080p, but then falls way off when the resolution goes up. Very odd results. General driver issues or Vega features being disabled showing where it bottlenecks?

http://www.pcgameshardware.de/Playe...s-Battlegrounds-neue-Benchmarks-Vega-1236260/

IMO it's more likely an Nvidia bottleneck at higher resolutions than anything to do with Vega.

 

[Kallogan]

Vega handily beating a 1080TI OC in PUBG @ 1080p, but then falls way off when the resolution goes up. Very odd results. General driver issues or Vega features being disabled showing where it bottlenecks?

http://www.pcgameshardware.de/Playe...s-Battlegrounds-neue-Benchmarks-Vega-1236260/

Vega : the alien gpu

 

[kondziowy]

Seems like Vega is the best gpu on the market to play PUBG on high refresh rate 1080p monitors 🙂

It's a shame nobody tests on very low settings though, as almost every player and every twitch streamer plays PUBG on very low.