Overclocked Ivybridge-E vs. overclocked Haswell-E

magellan · Oct 7, 2015

If we were to o'clock a Ivybridge-E to 4.5 GHz and a Haswell-E to 4.5 GHz it's more than likely the Haswell-E's uncore won't be running at 4.5 GHz (i.e. if you don't have one of the ASUS m'boards) but the Ivybridge-E will have its uncore/L3 cache running at the full 4.5 GHz. So if the Haswell-E uncore is still running at below 4 GHz speeds wouldn't that mean the Ivybridge-E could match or exceed the perf. of a Haswell-E at the same overclocked speed of 4.5 to 4.6 GHz.?

EarthDog · Oct 7, 2015

Depends, but in most cases, nope. There is (much) more to it than cache that makes HW-E perform a few percent faster at the same clocks. Cache has almost no bearing on performance in the vast majority of cases. Are there exceptions? Surely.

Woomack · Oct 7, 2015

Cache on IB-E is not affecting anything so much as on HW-E. Even then HW-E is not much faster if you set higher cache frequency just because it mainly affects memory bandwidth and that's already really high. So as ED said, there are more factors what make HW-E faster clock to clock than IB-E. Difference is between 0 and 15% depends from application. In other words 4.7GHz 4930K will be about as fast as 4.5GHz 5820K in many ( mainly new ) applications.

If you try to check full stability of cache clock then you will see that nothing much above 4GHz will be stable. Most chips won't go much above 3.7GHz and this is actually max for boards without OC socket. OC socket is helping mainly in benching while in 24/7 work it's not so important.

magellan · Oct 7, 2015

Woomack said:
Cache on IB-E is not affecting anything so much as on HW-E. Even then HW-E is not much faster if you set higher cache frequency just because it mainly affects memory bandwidth and that's already really high. So as ED said, there are more factors what make HW-E faster clock to clock than IB-E. Difference is between 0 and 15% depends from application. In other words 4.7GHz 4930K will be about as fast as 4.5GHz 5820K in many ( mainly new ) applications.

If you try to check full stability of cache clock then you will see that nothing much above 4GHz will be stable. Most chips won't go much above 3.7GHz and this is actually max for boards without OC socket. OC socket is helping mainly in benching while in 24/7 work it's not so important.

If we have both the HW-E and IB-E cores @ 4.5 GHz, but the L3 cache on the HW-E is @ 3.7 GHz, that's giving up 800 MHz in L3 cache speed to the IB-E. In most cases it would probably be giving up closer to 1 Ghz in L3 cache speed. Considering everything memory related has to go through the uncore, I find it hard to believe that this wouldn't be reflected in at least some benchmarks,

Will Skylake feature synchronized uncore/core speeds like IB-E and SB-E?

EarthDog · Oct 7, 2015

Cache isn't typically a bottleneck. But as we said, there are some benchmarks that show an improvement with it, just not many.

I haven't seen anything on skylake-e, but it shouldn't be much different than skylake. Review on the front page or the anand article goes a bit more in depth if you want to look.

magellan · Oct 8, 2015

EarthDog said:
Cache isn't typically a bottleneck. But as we said, there are some benchmarks that show an improvement with it, just not many.

I haven't seen anything on skylake-e, but it shouldn't be much different than skylake. Review on the front page or the anand article goes a bit more in depth if you want to look.

The uncore is more than just an L3 cache though. It's the interface where cache coherency issues between cores are resolved. As Woomack pointed out it's also the only way for the cores to get to memory.

It might be easy to test the theory by dropping the uncore by 800MHz to 1000MHz (the possible uncore frequency discrepancy between an IB-E and a HW-E overclocked to 4.5 GHz) and see if that negatively affects any benchmarks.

EarthDog · Oct 8, 2015

I'd be interested to see your results on that.

Woomack · Oct 8, 2015

Here is my thread made just after hw-e release: http://www.overclockers.com/forums/showthread.php/750439-DDR3-4930K-vs-DDR4-5820K-comparison
There you have comparison at the same clocks.

EarthDog · Oct 8, 2015

TL;DR version:

A lot depends also from memory controller but it's hard to come any closer in a different platform comparison.

Overclocked Ivybridge-E vs. overclocked Haswell-E

magellan

Member

EarthDog

Gulper Nozzle Co-Owner

Woomack

Benching Team Leader

magellan

Member

EarthDog

Gulper Nozzle Co-Owner

magellan

Member

EarthDog

Gulper Nozzle Co-Owner

Woomack

Benching Team Leader

EarthDog

Gulper Nozzle Co-Owner

Similar threads