How The Radeon OpenGL Performance Has Evolved From The HD 2900XT To RX Vega

[Michael]

@Michael:

Where do I get the 'glmark2' number from?
You have GLmark2 v276.
I have 'glmark2 2013.08.07' for ages, but no #vXXX.

http://openbenchmarking.org/innhold/d7e49977810be79d3097050c62b1a84b51871664

[nuetzel]

http://openbenchmarking.org/innhold/...b1a84b51871664

That was _fast_. ;-)
Thanks!
Dieter

[fallenbit]

@Michael:

Do you plan a comparison in the freesync range on 1440 (vertically) in regard which CPU is already sufficient to drive Vega into it's limit?

Based on Windows reviews already a R3 1200 @ 3.9 GHz should be enough to full load Vega (was a 1080 in the reviews).
The only difference on Linux might be driver CPU efficiency and the resently added mesa thread support, which could favor SMT a. k. the 1400 4C/8T @ 3.9 GHz. But I do not realy know whether these mesa threads are on the CPU. I would be suprised, if the 1600 @ 3.9 could improve the fps over the 1400 (OC'ed!). This is already all GPU limited @ 1440.

So for the 60 Hz gamers an overview of the OC'ed base CPUs at 1440 and what max settings in the game would keep the graphics card in the freesync range would be a different and very interesting test.

I'm aware, that freesync needs 'display' which in turn needs another 6-12 month, if it will ever come.

And yes, other gamers need 120, 144 or 240 Hz, these will not opt for the low range CPUs.

But casual gamers with freesync monitors like 3880x1440 or 3880x1600, doing mostly other stuff (specially the 1600 allows for a better scale of portrait documents (letter, A4) but is insane expensive when compared) will most likely be fine for gaming even with a quad core.
As you do not overclock, the X-versions would be close enough in that regard, the rest is just pro-rating.

But beeing non premium, just asking... :-)

Personally I nearly convinced myself of the R5 1400 @ 3.9 GHz, but the difference to the R3 1200 is around 50€ (105 vs 155 vs 195€ (1600) here in de), which is a bit high for only 4 virtual cores (SMT vs non SMT), but might be a tad more future prove.

Sorry for this long post...

[tildearrow]

Typo:

RX Vega 56/65

[AnonymousCoward]

The outlier seems to be the 6950, which in some cases is much slower than the 6870. Shouldn't it be consistently faster?

[Michael]

The outlier seems to be the 6950, which in some cases is much slower than the 6870. Shouldn't it be consistently faster?

No, HD 6900 series uses a VLIW4 design and was never optimized on the open driver as much as VLIW5

[Strunkenbold]

Well the RX Vega 64 is actually showing some numbers, did you meant the 7950 instead, because its graph is missing?
If so could you confirm https://bugs.freedesktop.org/show_bug.cgi?id=102201 ?

Maybe I was a little bit unclear.

@Michael
There seems to be a typo on Page 6:

Unigine Valley could run on the AMD hardware going back to the Radeon HD 4870 while ran into a hang with it on the RX Vega 64

How could it run into a hang when its actually showing a result? There are also no performance per watt and energy consumption figures.
Instead 7950 is missing from that comparison. A patch for this is already on the ml, thx Marek for the fix!

[onicsis]

But the vast majority of the GPUs sold this days is for laptops. Market of discrete graphic cards for big boxes become a niche.
AMD should focus their creativity on building open source drivers on mobile markets(including MiniPCs with a lot of horsepower and GBs of RAM).
Anyway kudos to them for exceptional performance of latest OSS drivers

[vipor29]

A powerful cpu always helps push more fps. Compare your cpu to the one used for these benchmarks.

i have a core i7-7700k which is pretty high end lol

[GreatEmerald]

Ah, and I still have my old HD 4890 sitting somewhere in a box. And my current GPU (GTX 660) seems to be around the middle between the former and an RX 480... So indeed time for an upgrade. However, it is a bit concerning that the RX 480 seems to be drawing as much (if not more) power as the HD 4890...