JEMC said:
No, I don't have any info. It's just the non-written yet usually true "law" that happens everytime a new gen of cards gets launched. If we look at the last AMD cards we see that the second tier cards, the x8xx ones, usually match the performance of the previous first tier cards, the x9xx ones. HD 4870 ~ HD 5770 HD 5870 ~ HD 6870 HD6970 ~ HD 7870 |
While I mostly agree, here's some comparison of those cards (with Voodoo ratings):
4870vs 5770 - 78VP vs 76VP (VLIW5 800:400:16@750/900; 256bits) vs (VLIW5 800:40:16@850/1200; 128bits)
5870vs6870 - 132VP vs 120VP (VLIW5 1600:80:32@850/1200; 256bits) vs (VLIW5 1120:56:32@900/1050; 256bits)
6970vs7870 - 156VP vs 173VP (VLIW4 1536:96:32@880/1375; 256bits) vs (GCN 1280:80:32@1000/1200; 256bits)
As you can see:
in case of 4870 and 5770 we have VLIW5 with same config, higher clock on 5770, but lower bus width for same performance on average.
In case of 5870 and 6870, we have VLIW5 with more of eveything on 5870, same bus width, for better performance of 5870, lagging only in pixel fill rate
In case of 6970 and 7870, we have VLIW4 vs GCN, and though it seems that 6970 has lot more shaders and TMUs, equilized to 1GHz of 7870, 6970 would be more like 1352:84:28 card - not too different from 7870, but still, considering GCN is newer and quite improved architecture, 7870 wins by some 10%
Now, let's look at why I asked you if you have some inside info on GCN2:
7870 - GCN 1280:80:32@1000/1200; 256bits
8770 - GCN2 768:48:16@unkown clocks; 128bits
I hope you see where I got confused, even if GCN2 is jump as VLIW4->GCN, 8770 with those specs would need (in my guesstimate) to run at 1.5GHz to achieve 7870 level of performance.
Of course, there is a chance that 8770 is not 768:48:16 part, but 1024:64:32, what 8830 is presumed to be, and 8830 is 1280:80:32 part - in that case I can see how 8770 can reach performance of 7870, if GCN2 is that better than GCN.
