[HardOCP] XFX 7950OC review with clock-for-clock comparo to 7970

[Termie]

Link: http://hardocp.com/article/2012/03/01/xfx_radeon_hd_7950_black_edition_video_card_review/8

HardOCP has done a review of the XFX 7950 Black edition, and what's most interesting about it is it allows some direct comparisons of the 7950 and 7970 at almost equal clocks.

(1) 7950 at 900/1375
(2) 7970 at 925/1375

Comparison in four games:
(1) Arkham City: 3%
(2) Skyrim: 6%
(3) BF3: 4%
(4) Deus Ex: 14%
Average: 7%

Given that just the slight core clock difference could account for almost 3% of that, you're talking about very limited gains from the 14% more cores, which is not that different from the 6970 over 6950, which had a theoretical 9% shader advantage and an actual advantage of ~3%.

This could actually bode well for the 7870, with its 1408 cores at 950MHz. With roughly 75% of the cores and ROPs, 75% of the bandwidth, and a ~20% higher clock, it could actually come within striking distance of the 7950!

Also, for all you OCers out there, HardOCP puts up some serious OC numbers for the 7950 (1195/1700). Just one graph to give you an idea:

 

[Troll Trolling]

This is quite clear when comparing with Cape Verde. The HD 7750 got like, 50%+ perf/shader or thereabouts than the HD 7950 (I chose those two because they've the same clock speed).

 

[lavaheadache]

Every game is going to be different depending on how they use these cards. I'm sure you can do the same with other cards such as a GTX 570 and 580.

Some games are going to react better to higher mhz while others like more cores or both.

 

[BallaTheFeared]

7950 doesn't lose any ROPs though. ROPs are probably more important here than shaders. The only place where the texture fillrate did much was in a game that is pretty much pure texture based (Deus Ex).

The cards seem to be ROP bottlenecked, which is probably why they have a harder time with titles like BF3. Losing ROPs is going to probably hurt performance more than losing some shader clusters.

 

[Concillian]

This could actually bode well for the 7870, with its 1408 cores at 950MHz. With roughly 75% of the cores and ROPs, 75% of the bandwidth, and ~20% high clock, it could actually come within striking distance of the 7950!

Not sure you can really say this with any kind of confidence.

7950 --> 7970 at the same clock speed has the same ROP performance.
7870 will have significantly lower ROP performance due to having 24 instead of 32.

If you look at the 7970 vs. 7950 there are three things that can make the loss of a shader cluster only a minor impact:
- the fact that the ROP performance is the same
- the fact that the memory bandwidth is the same
- the fact that the cache per shader and / or TMU increases when you disable a cluster

the first and second bullets won't be true with the 7870. The last bullet? Who knows.

Hopefully we don't have to wait too long to find out.

 

[Termie]

7950 doesn't lose any ROPs though. ROPs are probably more important here than shaders. The only place where the texture fillrate did much was in a game that is pretty much pure texture based (Deus Ex).

The cards seem to be ROP bottlenecked, which is probably why they have a harder time with titles like BF3. Losing ROPs is going to probably hurt performance more than losing some shader clusters.

Good point, but the 7870 has 75% of the ROPs (24 vs 32) and stream processors (1408 vs 1792), and a ~20% higher clock (950 vs 800)...so it's at a 25% disadvantage at most, which the high clock may compensate for (but which will limit OC headroom). That could put it ahead of the 6970, which offers roughly 75% of the performance of the 7950.

Not that I'm necessarily that hot on the 7870 until we see price/performance next week.

 

[superjim]

I doubt the 7870 will have the OCing headroom of the 7950 with stock clocks as high as they are. If it does, it will be a monster at $300. It's performance will give the ROPs a final verdict in performance impact.