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

[urvile]

Yes, they did that.

If vega 64 had performance that was better than the gtx 1080. Then I would I have been all over a pair of AIO vega 64. I have a monster power supply so I don't really care about power draw. At that level of perfomance the price would have been right but it's basically an AMD version of the gtx 1080 and I can get an actual gtx 1080 for LESS. So why bother with vega?

As I mentioned I have a freesync monitor but that's not enough to make me buy vega. It's a bit of a hassle to sell it but I am in a position where I can buy the g-sync version and then sell it.....so I don't feel locked in. Before I go gsync though I am going to keep using my current monitor and see how it goes for gaming.

Anyways. I am a bit disappointed with how it turned out I was looking forward to going full red team with my latest build.

Caveat: Of course this all relates to pricing in australia but it's relevant to me. YMMV.

 

[Elixer]

If vega 64 had performance that was better than the gtx 1080. Then I would I have been all over a pair of AIO vega 64. I have a monster power supply so I don't really care about power draw. At that level of perfomance the price would have been right but it's basically an AMD version of the gtx 1080 and I can get an actual gtx 1080 for LESS. So why bother with vega?

You hit the biggest point on why gamers would get a Vega, and that is, if they already had a freesync monitor, and don't want to invest more $$$ in a g-sync monitor.

The Vega56 is rather interesting though, if indeed it can come within 1-2% of a Vega64 via the BIOS change, then, that paints things slightly differently, once more people can confirm that, then, that card would be a better value for pretty much all freesync users. Sure, still won't beat a 1080 in everything, but, it will be a better bang for the buck.

AMD doesn't have a top tier card that can compete with nvidia's top end (1080ti), (and no, there is no amount of driver tweaking they could do to beat a 1080ti in everything), there is no sugar coating it, they blew it big time, and their PR machine was just horrible.

For non-gamers, jury is still out, and it is highly dependent on workloads.

The only other reason to want Vega is, you want to support the underdog / think AMD is a better company than nvidia (ie, more open/ uses open standards) / don't want to pay the g-sync tax.

 

[Veradun]

As a purchaser of a graphics card why should I care about what Raja thinks the bigger picture is? What is the tangible benefit to me?

As I already said multiple times, even before Raja did it yesterday, that chip is meant to fight three different chips in three different fields. For gaming it is a GP104 competitor. As a purchaser of a graphic card for gaming you'll do as anyone: evaluate gaming performance, features, price and power and make your choice.

When instead you are judging the architecture the bigger picture is what you want to look at.

Vega 10 does not support 1/2 rate FP64 like GP100 does. [...]

Vega 10 has peak FP16 performance on par with V100. There are tradeoffs when designing chips, more so if you chip has to work in vastly different fields. I never said Vega kills GP100, I said it has to compete with it, for example in deep learning.

Exactly, i care about one thing and one thing only, real world gaming performance. I could care less what it does in compute workloads, if this is a compute only card and thats where AMD thinks the big picture is then they should have never released it in the RX line.

Exactly. But they don't design around you, deal with it :>

 

[krumme]

If you live in scandinavia (with high prices and VAT) there is still a lot of cards at komplett to be had at slightly below lowest 1070 price. There is only one sapphire left for the 6 september but there is still 84 xfs comming the 15 september and 100+ MSI comming the 22 september. They are not ordered, but are up for grabbing, because when ordered the number counts down. They are not confirmed, but usually komplett is as sureproof as it can be. Its evident they are launch partner for scandinavia. Besides 26 rx56 gigabytes the 2 october it seems thats the initial batch. Next is AIB comming probably october.

 

[showb1z]

€430 for RX56, that's not even a high price. I'd buy that in a heartbeat. Best I can find that ships here is €509.

 

[Despoiler]

I wasn't disagreeing about their importance, I was asking whether it would be possible to evaluate if this was bottlenecking Vega's performance without front end bottlenecking issues confounding the results. Does MSAA performance not depend at all on geometry performance? If it doesn't then Vega's 8x MSAA issues may be evidence of being ROP bottlenecked (assuming those issues aren't a driver bug).

According to sebbi the ROPs handle MSAA. If you goto page 1 of that thread he explains that simple shaders are ROP bound. I think it would be possible to measure throughput somehow. The problem with a test like that is that no one would program a game like that.

I do not believe Vega is ROP bottlenecked when applying 8x MSAA. Fiji never showed an issue. Presumably the ROPs are at least as good as Fiji. No? When I say at least as good, it's important to remember that ROPs are not created equal. I suspect it's just a driver issue. There have been more than a few times where AMD has had to patch the driver specifically for 8x MSAA performance. 17.6.2 patched Dirt 4's 8x MSAA performance. Granted it's never been an across the board type fix, but it serves to highlight that the driver does do something different with 8x MSAA.

 

[Rasterizer]

According to sebbi the ROPs handle MSAA. If you goto page 1 of that thread he explains that simple shaders are ROP bound. I think it would be possible to measure throughput somehow. The problem with a test like that is that no one would program a game like that.

Presumably the ROPs are at least as good as Fiji. No? When I say at least as good, it's important to remember that ROPs are not created equal. I suspect it's just a driver issue. There have been more than a few times where AMD has had to patch the driver specifically for 8x MSAA performance. 17.6.2 patched Dirt 4's 8x MSAA performance. Granted it's never been an across the board type fix, but it serves to highlight that the driver does do something different with 8x MSAA.

True, but we do have B3D Suite results for pixel fill rate:

However, having done some research into the topic, doesn't the fact that Fiji performed relatively better at higher resolutions preclude it being ROP bound? If Vega was ROP bound, shouldn't its performance worsen relative to Pascal at higher resolutions, rather than staying the same or improving?

 

[Rifter]

You hit the biggest point on why gamers would get a Vega, and that is, if they already had a freesync monitor, and don't want to invest more $$$ in a g-sync monitor.

The Vega56 is rather interesting though, if indeed it can come within 1-2% of a Vega64 via the BIOS change, then, that paints things slightly differently, once more people can confirm that, then, that card would be a better value for pretty much all freesync users. Sure, still won't beat a 1080 in everything, but, it will be a better bang for the buck.

AMD doesn't have a top tier card that can compete with nvidia's top end (1080ti), (and no, there is no amount of driver tweaking they could do to beat a 1080ti in everything), there is no sugar coating it, they blew it big time, and their PR machine was just horrible.

For non-gamers, jury is still out, and it is highly dependent on workloads.

The only other reason to want Vega is, you want to support the underdog / think AMD is a better company than nvidia (ie, more open/ uses open standards) / don't want to pay the g-sync tax.

Honestly its more the MSAA issue for me over much else, otherwise vega 56 looks pretty good vs a 1070. It looks like vega drops 20-30% Vs Nvidia with 8xMSAA. I have alot of older games in my catalog, and i use 8xMSAA in most of them. Sure now the 20-30% MSAA penalty is not to bad because i only have a 60Hz screen and any midrange or better card can hit that even with a 30% perf penalty, but i intend to upgrade to a 144Hz G/free sync panel with my GPU upgrade and then the 20-30% MSAA performance loss will be an issue when pushing for 144FPS.

Releasing a top tier(well for AMD its top tier) GPU that takes a huge hit in the most popular and effective AA implementation is a huge blunder, how they missed this in internal testing ill likely never know.

The perf/watt is bad to, but for me personally its the MSAA issue more than anything else that makes vega a bad buy.

 

[Qwertilot]

Very unlikely they missed it, so presumably its quite hard to fix for whatever reason(s). Either just flat out hard or the knock on consequences would be undesirable.

 

[Glo.]

True, but we do have B3D Suite results for pixel fill rate:

However, having done some research into the topic, doesn't the fact that Fiji performed relatively better at higher resolutions preclude it being ROP bound? If Vega was ROP bound, shouldn't its performance worsen relative to Pascal at higher resolutions, rather than staying the same or improving?

What you see in this graph is one thing. Draw Stream Binning Rasterizer being disabled.

 

[Despoiler]

Honestly its more the MSAA issue for me over much else, otherwise vega 56 looks pretty good vs a 1070. It looks like vega drops 20-30% Vs Nvidia with 8xMSAA. I have alot of older games in my catalog, and i use 8xMSAA in most of them. Sure now the 20-30% MSAA penalty is not to bad because i only have a 60Hz screen and any midrange or better card can hit that even with a 30% perf penalty, but i intend to upgrade to a 144Hz G/free sync panel with my GPU upgrade and then the 20-30% MSAA performance loss will be an issue when pushing for 144FPS.

Releasing a top tier(well for AMD its top tier) GPU that takes a huge hit in the most popular and effective AA implementation is a huge blunder, how they missed this in internal testing ill likely never know.

The perf/watt is bad to, but for me personally its the MSAA issue more than anything else that makes vega a bad buy.

You must play some really old games if you think MSAA is the most popular AA form these days. The minority of moden games have MSAA because it conflicts with rendering methods. Over on the B3D Vega thread sebbi was asked about the 8xMSAA issue and his response was long the lines that he doesn't care because he is beyond MSAA. Look at the Vega reviews for what AA they use. Techreport Vega reviews only 3 of 8 games support MSAA. They only used it on 2 of the 3. You will see something over @ the [H] review also. They won't choose MSAA unless they have to. Methods like FXAA, TXAA, SMAA, CMAA, TSSAA have replaced MSAA. Personally I will choose other methods over MSAA. If I have to use something like FXAA because it's the only choice(fairly common) I'll use Reshade to do the FXAA and turn on sharpening and various other filters to make it look amazing.

 

[Rasterizer]

What you see in this graph is one thing. Draw Stream Binning Rasterizer being disabled.

Isn't pixel fill rate a function of ROPs times clockspeed? I'm not sure how DSBR would affect it? Isn't the primary benefit of DSBR related to saving memory bandwidth?

 

[Glo.]

Isn't pixel fill rate a function of ROPs times clockspeed? I'm not sure how DSBR would affect it? Isn't the primary benefit of DSBR related to saving memory bandwidth?

Theoretical Fillrate - yes.

 

[PeterScott]

You must play some really old games if you think MSAA is the most popular AA form these days.

He didn't say it was the most popular, he said he used it. Likely because he prefers it. MSAA is polygon aware, and does AA on jagged edges but preserves sharpness. Many of us prefer it when available.

The post process methods like FXAA are not polygon aware. They simply blur everything on the screen and many of us hate that.

 

[maddie]

Saw that GDDR5 memory prices rising ~ 30%. Actually all memory prices are rising.

HBCC to the rescue? How likely 4GB Vega 56 in the near future?

 

[Despoiler]

He didn't say it was the most popular, he said he used it. Likely because he prefers it. MSAA is polygon aware, and does AA on jagged edges but preserves sharpness. Many of us prefer it when available.

The post process methods like FXAA are not polygon aware. They simply blur everything on the screen and many of us hate that.

Your reading comprehension needs some tuning.

Releasing a top tier(well for AMD its top tier) GPU that takes a huge hit in the most popular and effective AA implementation is a huge blunder

Thanks for telling me how FXAA works. As if I didn't know... You clearly don't know what Reshade can do. There is a reason why I mentioned it when talking about FXAA.

 

[Rifter]

You must play some really old games if you think MSAA is the most popular AA form these days. The minority of moden games have MSAA because it conflicts with rendering methods. Over on the B3D Vega thread sebbi was asked about the 8xMSAA issue and his response was long the lines that he doesn't care because he is beyond MSAA. Look at the Vega reviews for what AA they use. Techreport Vega reviews only 3 of 8 games support MSAA. They only used it on 2 of the 3. You will see something over @ the [H] review also. They won't choose MSAA unless they have to. Methods like FXAA, TXAA, SMAA, CMAA, TSSAA have replaced MSAA. Personally I will choose other methods over MSAA. If I have to use something like FXAA because it's the only choice(fairly common) I'll use Reshade to do the FXAA and turn on sharpening and various other filters to make it look amazing.

Indeed, most games i own are in the 3-10 year old category, i tend to wait till they are 2-3 years old and buy them for $5 on sale. Only games i buy new are MMO's, or popular multiplayer games i want to play while there is still a active multiplayer community. Single player i always wait for $5-10 pricing. Id say 80% of my 400 game steam library is greater than 3 years old.

 

[Rifter]

He didn't say it was the most popular, he said he used it. Likely because he prefers it. MSAA is polygon aware, and does AA on jagged edges but preserves sharpness. Many of us prefer it when available.

The post process methods like FXAA are not polygon aware. They simply blur everything on the screen and many of us hate that.

Fully agree, FXAA is horrid. MSAA is best IQ IMO.

 

[Rifter]

Your reading comprehension needs some tuning.



Thanks for telling me how FXAA works. As if I didn't know... You clearly don't know what Reshade can do. There is a reason why I mentioned it when talking about FXAA.

Perhaps i should have clarified that statement with most popular in the games i play.

 

[Olikan]

MSAA is polygon aware, and does AA on jagged edges but preserves sharpness.

Mmm... So, the "MSAA bug", might be a geometry bottleneck?

 

[Rifter]

Mmm... So, the "MSAA bug", might be a geometry bottleneck?

I think it was mentioned above that it was the ROP's that do MSAA but all the hardware stuff is mostly over my head. This is why i focus on reviews and benchmarks that laymen can understand. Real world performance i understand.

 

[krumme]

Hitman928 with rx 56 with Custom with -1015 mV and +50% Power score 6839 in timespy graphics score.
I get aprox 6980 on a rx 64 using same settings
With same memory clock on the rx56 its more or less precisely the same score.

I dont know if Hitman score is specifically the graphics but mine is and i guess his is too.

 

[Headfoot]

Fully agree, FXAA is horrid. MSAA is best IQ IMO.

There are some extremely good TXAA/TAA implementations out there that IMO are superior to MSAA because they fix jaggies in transparent (non polygon) objects like textures too and have no weirdness in motion. Cryengine's TXAA and The Division's spring to mind.

 

[Malogeek]

There are some extremely good TXAA/TAA implementations out there that IMO are superior to MSAA because they fix jaggies in transparent (non polygon) objects like textures too and have no weirdness in motion. Cryengine's TXAA and The Division's spring to mind.

This. Temporal AA methods that aren't simply post-filters are the future and provide the best experience overall, especially in motion. Temporal SMAA is gorgeous.

 

[PeterScott]

There are some extremely good TXAA/TAA implementations out there that IMO are superior to MSAA because they fix jaggies in transparent (non polygon) objects like textures too and have no weirdness in motion. Cryengine's TXAA and The Division's spring to mind.

TXAA is just as bad as FXAA, if not worse:
https://www.gamespot.com/forums/pc-mac-linux-society-1000004/txaalarge-sheet-of-plastic-29319415/

These post process AA modes "fix" AA in everything, because they treat the screen like bitmap image and just blur everything.

There are essentially various flavor of the month blur filters. Cheap to implement, so often used instead of the superior but more expensive MSAA.