i7 980X vs i7 2600

jasjeet · 06-03-2015, 10:41 PM #1

Anyone know how these two CPUs handle Dolphin?
What would the difference be if the 980X was at 4Ghz, and the i7 2600 at 3.2Ghz?

**degasus** · 06-03-2015, 11:31 PM #2

A lower-end haswell will run better, even at a lower clock.

jasjeet · 06-03-2015, 11:37 PM #3

Yeah i know this, but i ask about these CPUs as they are costing exactly £0.00.

I think there are some CPU instruction sets that may hurt the 980x in dolphin today compared to the 2600?

**AnyOldName3** · 06-03-2015, 11:50 PM #4

The 980X isn't even a Sandy Bridge chip, so will perform much worse in Dolphin (and, in fact, anything which uses a maximum of four or fewer cores, which is most things) than the 2600 at the same clock rate... but it's frequency is higher, so it won't actually be that far behind.

jasjeet · 06-04-2015, 12:00 AM #5

In the benches ive seen, the clock to clock performance difference is about 10% to Sandy bridge, another 10% to ivy bridge, another 10% to Haswell. So yeah the single core performance is going to be an issue for Dolphin. I will have to test it against my 3770 system.

**Fiora** · (This post was last modified: 06-04-2015, 12:30 AM by Fiora.)

The gain of Haswell over Ivy Bridge is significantly larger in Dolphin (like 30%). Nobody quite knows why.

jasjeet · (This post was last modified: 06-04-2015, 12:43 AM by jasjeet.)

Ah thats a big one, ill have to consider how much use he will get from Dolphin and then make my mind up.

Thanks everyone.

DZCreeper · 06-07-2015, 09:03 PM #8

(06-04-2015, 12:30 AM)Fiora Wrote: The gain of Haswell over Ivy Bridge is significantly larger in Dolphin (like 30%). Nobody quite knows why.

I assumed the performance gain was somehow a byproduct of Haswell adding AVX2, gather-scatter, BMI1, BMI2, ABM, and FMA3 instruction support. The last 4 being responsible is unlikely though, considering AMD chips also have had those instructions for some time.

**Fiora** · (This post was last modified: 06-07-2015, 09:16 PM by Fiora.)

AVX2 isn't even used in Dolphin. Gathers aren't either, and are actually slower than doing every load separately. BMI1 isn't used significantly (not sure if I ever finished my patch to add it, but the gain wasn't significant). ABM is AMD-only. FMA3 does matter, but the gain was like 1 or 2 %.

Haswells were fast long before any post-SSSE3 instruction set was even supported in Dolphin. If I had to guess it might be something more subtle, maybe instruction cache-related? Or the 4th execution pipe? But I really don't know...