General Discussions

[black_zion]

I won't link the full charts as they are large, but I will quote their conclusion, and that it goes to show you that the Fury-X's 8.2 Tflops of compute power are meaningless when you consider the RX 590 packs 6.3 and 5500XT a mere 5.1.

https://www.tweaktown.com/articles/9476/amd-radeon-r9-fury-how-does-it-stack-up-in-2020-against-navi/index.html

[black_zion]

Actually, I will post the 1920x1080 charts, there are 2560x1440 and 4K charts as well, but needless to say you won't be playing at those resolutions with the Fury X.

[colesdav]

I played at 4K with a pair of R9 Fury X in DX11 Crossfire and DX12 MultiGPU.
I played on single R9 Fury X at 2K on many games.
Let down by lack of driver support.
I bought them for Compute performance and because they work on ROCm and Blender though, which is not discussed at all. 

[hardcoregames_]

colesdav wrote:

I played at 4K with a pair of R9 Fury X in DX11 Crossfire and DX12 MultiGPU.
I played on single R9 Fury X at 2K on many games.
Let down by lack of driver support.
I bought them for Compute performance and because they work on ROCm and Blender though, which is not discussed at all. 

BIOS and driver headaches seem to be the way with team red. Eventually cards like my RX 480 are sold in volume so the drivers tend to be better tuned.

last time I use dual cards was GTX 260 SLI. My GTX 690 is technically SLI but it is one card.

aside from my X470, i usually get boards for one video card and see what I can buy to leverage it

[colesdav]

Sure - Polaris sold well so lots of focus on the drivers. 

One video card should be enough for people not working on Compute ot Blender.
DX11 Crossfire is dead and not supported any more.
Latest AAA games that work will likely stop working in future as games are modified, since no AMD Crossfire support.
DX12 MultiGPU has little support and even less benefit.

[hardcoregames_]

colesdav wrote:

Sure - Polaris sold well so lots of focus on the drivers. 

One video card should be enough for people not working on Compute ot Blender.
DX11 Crossfire is dead and not supported any more.
Latest AAA games that work will likely stop working in future as games are modified, since no AMD Crossfire support.
DX12 MultiGPU has little support and even less benefit.

none of the major engines today support dual cards anymore

so for gaming multiple cards do not make sense

the Intel Phi card is a power pig but useless for gaming, they are selling for < $99 on eBay for cards that were > $2000 a few years ago

[hardcoregames_]

I have a Sapphire R9 Fury with 3 fans on it, 12" long and it a beast.

I have used the card at 1920x1080 and all games even now are playable. At 1920x1080 many games do not even load the R9 Fury that much, while some games are more demanding.

I also have the RX 480 8GB which is what I am using at present.

My R9 Fury does not work on my X570 for some reason but does work on the older X470. Both platforms are setup for benchmarking. I may build a new rig when the X670 surfaces with the Ryzen 4000 surface.

[colesdav]

Remember that R9 Fury X was fabbed on TSMC 28nm versus Polaris and Vega on Global Foundries 14nm, RX 590 on 12nm,  and now Navi on TSMC 7nm.
The process advantages from 28nm to 14nm is a large jump as is the jump from 14nm to 7nm.

[hardcoregames_]

colesdav wrote:

Remember that R9 Fury X was fabbed on TSMC 28nm versus Polaris and Vega on GlobalFoundries 14nm, RX590 on 12nm,  and now Navi on TSMC 7nm.
The process advantages from 28nm to 14nm is a large jump, as is the jump from 14nm to 7nm.

back in the day when video cards were 55nm and needed 200W each, bought a Corsair TX850V2 to handle a pair of them. Back then many games were able to be improved with dual cards, but not all.

My RX 480 is 14nm but it is not a flagship card, it was cheap when I grabbed it with an eye on VRAM which seemed to be a bottleneck for gaming at 4K.

Now that I have results to work with, I can now see more clearly forward for my requriements.

Vega 56 cards are available with 8GB and Vega 64 cards also are available with 8GB VRAM. A vendor I saved on eBay has thousands of old video cards, he sent me a photo, card is dead not to worry he has lots of spares. He had a huge pile of RX 470 4GB which are not as powerful as my RX 480 so not much use but lots of potato rigs can use those old cards.

[pokester]

I did get a long life out of my 7950 and 7970. No complaints and the drivers were never an issue either, at least pre-Wattman. It wasn't until Battlefield 1 that I ran into a game they could no longer play. I only got a year out of my RX 580 before I outgrew it. Never had a Fiji or Nano but their lesser brothers were darn fine cards too even without the higher bandwidth memory. 

[hardcoregames_]

pokester wrote:

I did get a long life out of my 7950 and 7970. No complaints and the drivers were never an issue either, at least pre-Wattman. It wasn't until Battlefield 1 that I ran into a game they could no longer play. I only got a year out of my RX 580 before I outgrew it. Never had a Fiji or Nano but their lesser brothers were darn fine cards too even without the higher bandwidth memory. 

so far i have 12 months into this RX 480 pending the RTX 2080 taking over

never was able to get my R9 Fury to work on my X570, it does work on the X470 which is bizzare

[pokester]

Out of curiosity on those Fiji vs later GPUs test, I would be curious if you would get the same results with current drivers vs old drivers. I always thought that the Wattman drivers killed performance on my 7950 and they got worse not better as time went on. 

[black_zion]

colesdav‌ there's an interesting challenge for you since you still have a Fury X, run 3DMark11 (that's free now), on it with drivers a few months after release and current drivers. I say 3DMark11 because to my knowledge AMD and nVidia both still spend time optimizing drivers for it.

[colesdav]

Hi,

I own 3DMark with all possible options..

Please take a look at my 3D Mark Firestrike Scores here:
AMD Red Team 3DMark Scoreboard 

I already looked at the Fury X drivers and HBM Overclocking versus "current"  drivers for those tests.

The FireStrike scores is #2 because it is limited by the Ryzen 2700X undervolted and tuned to run at 4.3 - 4.4 GHz.
The RAM is only running at 3200MHz.

If I get a Zen3 I will rerun the Quad Fury X scores again, and I should be able to improve all scores. 

Best perfoming FuryX Drivers were Pre- Vega 56/64 release and pre-removal of HBM overclocking.
AMD Crimson ReLive 16.12.2 driver was the last one that actually worked with HBM overclocking on Fury X. 

It was completely disabled in later drivers. .

I ran those results ~ November 2019 I think... so that would have been comparing to 19.11.1 drivers.
I will try to look again at Fury X driver performance again some time in future.
Right now all of my PC with R9 Fury X are stuck on 19.12.1 drivers for blender use.

Thanks.

[pokester]

I think that is what we are asking is what is the performance difference between those early 16.12.2 drivers and the current ones. 

[colesdav]

Yes I understand.
I just do not have time to run the comparison right now.
The 16.12.2 drivers were O(5%) faster than 19.11.1 on single R9 Fury X.

[pokester]

I wasn't expecting you too. I was just curious myself. I think we just know you are really in to testing stuff and appreciate your contributions in that area. I am sure we both understand you don't need or have to test that. It really doesn't matter as it is quickly becoming irrelevant outside of compute functions. 

Just that information between the 16x and 19x and the drop I saw with the 2020 drivers, my guesstimate would be nearing a 10% decline in speed over time. 

[colesdav]

I will test the FuryX on 16.12.2 versus 20.5.1 or next driver for you.

I am interested myself to see it.

It is just I have machines with multiple R9 Fury GPUs running flat out for Blender work I am doing and I keep everything on Adrenalin 19.12.1 driver.
Adrenalin 2020 interface is a complete and utter car crash for MultiGPU and in my case I have the additional complication that some of my PCs are running R9 Nanos and R9 Fury cards on eGPU over Thundebolt 2 and 3. None of that is working in Adrenalin 2020 either.

[pokester]

It's awesome if you do but if you really don't have time it is not necessary. I was just curious as others have been if Fiji cards really just died because of architecture or abandonment in the drivers. 

[black_zion]

Would be nice if TweakTown would do an article on it as well, since it doesn't quite seem logical that the Fury X is slower than an RX 590 is slower than a RX 5500XT despite raw computational power decreasing. It would tie in nicely to the discussion of the next generation consoles since both of their computational power figures have been touted greatly by both them as well as game developers.

[hardcoregames_]

black_zion wrote:

Would be nice if TweakTown would do an article on it as well, since it doesn't quite seem logical that the Fury X is slower than an RX 590 is slower than a RX 5500XT despite raw computational power decreasing. It would tie in nicely to the discussion of the next generation consoles since both of their computational power figures have been touted greatly by both them as well as game developers.

I don't have enough video cards to compare all the models very close as of yet.,

[colesdav]

I have been running FireStrike tests on RX Vega Liquid with my default DX11 and DX12 HBM2 overclock with Adrenalin 19.12.1 + 20.5.1 drivers to compare those results with PowerColor Red Dragon RX5700XT. I will post those results later. I think they are a more interesting set of data.