AMD Radeon R9 Fury X Reviews

[Rys]

@Davros said:

It's as if my last post on mantle the previous page went unnoticed,

Spoiler

Another misleading slide mantle doesnt have a 105% advantage in bf4 its 5%

 

[Rys]

@UniversalTruth said:

Another misleading slide mantle doesnt have a 105% advantage in bf4 its 5%

It's just another way to type the same......

Anyways, any comments about the 43% performance advantage in the other title?

 

[Rys]

@tabs said:

I would imagine that a Civ game is not particularly GPU bound in the first place, so Mantle is reducing some of the CPU load leading to a noticable increase in framerate.

 

[Rys]

@swaaye said:

Would be great if we could get AVX patches like the 3DNow! days. ( assume devs aren't yet using AVX)

K6-2 needed AVX. Surprise Intel!

 

[Rys]

@fellix said:

 

[Rys]

@rcf said:

But the settings are different. Fury X has FXAA and VSync enabled while with 980ti they are disabled.

 

[Rys]

@Lightman said:

But the settings are different. Fury X has FXAA and VSync enabled while with 980ti they are disabled.

I would also say there is night and day difference between these videos! Besides, has anyone else reproduced these slowdowns in GTA V? I've seen pauses like that on AMD GPU when Windows was on it's last leg and heavily swapping through HDD.
Before I see this replicated on another system I will only consider this as possible drawback of 4GB FB.

 

[Rys]

@fellix said:

3DMark scaling:

AMD Radeon R9 Fury X x1 = 4337
AMD Radeon R9 Fury X x2 = 7557 (74.2% scaling)
AMD Radeon R9 Fury X x3 = 10801 (42.3% scaling)
AMD Radeon R9 Fury X x4 = 13286 (23.0% scaling)

Source

 

[Rys]

@gamervivek said:

AMDmatt needs to do his job better.

https://www.reddit.com/r/hardware/comments/3bdwdi/amd_radeon_r9_fury_x_4way_crossfire_setup/

 

[Rys]

@ToTTenTranz said:

Looking at GPU scores:

- 91% scaling with 2 cards
- 91% scaling with 3 cards
- 89% scaling with 4 cards

This is using 16x/8x/8x/8x PCI-Express 3.0 on a X99.

 

[Rys]

@Rys said:

Can someone summarise the performance data for something that isn't 3DMark?

 

[Rys]

@Alexko said:

There's apparently great scaling in Sniper Elite 3 (near perfect) but that's with Mantle.

 

[Rys]

@CarstenS said:

Can someone summarise the performance data for something that isn't 3DMark?

In short and of course dependant on your preferred level of cherrypicking, the majority of sites agree that it's around 8-10ish percent slower than 980 Ti in 1080p, about 4-5 percent slower in 1440p and almost equal in performance in 2160p - with some sites noting that in some titles in Ultra HD begin to exhibit signs graphics memory overflow. The latter of course depending on the level of anti-aliasing applied with FXAA being better usable on Fury X than MSAA compared to 980 Ti.

 

[Rys]

@firstminion said:

It's relevant that Fury X matches or wins over 980ti on 4K, where Fury's reduced memory should hurt it more.

 

[Rys]

@CarstenS said:

Well, based on average Fps... but since GTX 970 and #memorygate we all know better, right?

 

[Rys]

@ToTTenTranz said:

It's relevant that Fury X matches or wins over 980ti on 4K, where Fury's reduced memory should hurt it more.

But the Fury's reduced memory does hurt it in games that take over 4GB VRAM. For example, GTA V has decent average framerates, but it shows some serious hiccups when turning the camera 180º.

 

[Rys]

@gamervivek said:

Fewer ROPs at lower clocks as well. Seems to have overturned the earlier TeraScale scenario where AMD would fall behind at higher resolutions with fewer ROPs at higher clocks. And apparently it can be pushed quite further one you get HBM to play with.

Without HBM overclocking,

http://www.3dmark.com/fs/5224465

With HBM overclocking,

http://content.hwigroup.net/images/products_xl/313064/17/amd-radeon-r9-fury-x.jpg

 

[Rys]

@gongo said:

How did this lesser known site managed to overclock the HBM while the rest couldn't? GPU-z error?

The core clock was also overclocked higher, as do the CPU clock..so they are not a like for like comparison?

 

[Rys]

@CarstenS said:

Apparently, with a non-review-driver you can do it. AMD investigates themselves.

 

[Rys]

@silent_guy said:

It's relevant that Fury X matches or wins over 980ti on 4K, where Fury's reduced memory should hurt it more.

Everybody who expected FuryX to perform worse at 4K with today's games didn't pay attention to benchmarks for other GPUs. Even with the GTX970, it wasn't trivial to come with convincing examples where the 512MB less made a major difference.

As long as the working set fits the memory, it doesn't matter how much more GB there are on board.

The argument against the FuryX wrt memory is the same as the on that was made against the GTX680 with its 1.5GB vs the 2GB of the 7970. Or against the 3GB of the 780Ti vs the 4GB against the R9 290X.
It may be a problem in the future, but right now, you have to look hard to find any issues with it. The only difference is that the 4GB is a structural limitation of HBM1 where it was a pure cost issue for the GTX680/780Ti in the past.