It's easy. In first chart, 1M/2T 5400K scores a bit faster than 2M/4T 5800K due to the fact that 1 thread from itunes(and Lame) is bounced on 2 frontends in 5800K vs 1 frontend of 5400. It's easy to grasp. 5400 obviously has no problem with 1 thread as 1 frontend is involved with fetching,decoding and scheduling ops. 5800K has 2 modules and 1 thread bounces on two separate frontends of these modules(and thus on 2 FlexFP units instead of being all done on one FlexFP unit like in the case of 5400 model). Turbo clock difference between 5400K and 5800K is exactly 10%,penalty difference due to inefficient scheduling varies it's around 10% or sometimes even more. The fact that 5800K and 5400K were neck and neck(3% difference) fits perfectly with the thread bouncing on 2 modules explanation.
On to the second chart. Original bulldozer review uses FX8150 which has 4.2Ghz Turbo clock. Score is 102 seconds. Newest A10 review has 8150 locked @ 3.8Ghz with all modules running (it says 8C/8T in the chart). So we have a single thread benchmark running on 4C/4T and 8C/8T CPUs which both run at locked 3.8Ghz frequency. In new test 3.8Ghz 8150 scored 108 seconds. How much is 4.2Ghz (old review Turbo clock since iTunes is monothread) divided with 3.8Ghz (new test with locked to 3.8Ghz 8150)? 4.2/3.8=1.1 or 10% clock difference. Time difference is ~6% and that is about what you should expect with variable Turbo if thread jumps from core to core. 4.2Ghz is applied to whatever core the itunes thread jumps to but efficiency is lost and that the 3% loss you see (or non-perfect scaling from 3.8Ghz fixed clock to stock clocked 3.6-4.2Ghz FX8150 in old test).
It all very easy if you use logic
😉.
As for the IPC of Piledriver we have 2 tests that THG ran. First is itunes at both chips locked @ 3.8ghz. This I explained already. Difference is ~15%, you can't go around this . It's a fact.
Second test they ran just confirms that 1st test (itunes ) was not a fluke. We see 15% difference in 3dstudio which is FP heavy workload.
I will do a 3rd test comparison just for you.
Take THG's Ivy Bridge review and Lame results(single thread,was not included in original FX8150 review for some reason). FX8150 scored 2:10 or 130s . Lame is single threaded workload and FX8150 runs at 4.2Ghz ,just like in the case of itunes in original FX8150 review by THG. 5800K that has the same 4.2Ghz single core Turbo as FX8150
scores 1:58s or 118s . Here we
again see 5400K is able to pull ahead due to its single module nature(no penalty for 1 thread bouncing,another proof my claim about itunes is correct
😉 ).
So 118s for 5800K @ 4.2ghz Vs 130s for FX8150 @ 4.2Ghz. 118/130=0.9 or 10% faster than 8150 at same clock. Even Lame shows 10% IPC improvement and Lame is an old integer benchmark.
edit: Just to add,check
stock 5800K and
8150 results in single thread itunes benchmark. Both chips have the sharing penalty and both have 4.2Ghz Turbo for these single thread benchmarks
😉. Voila,same 15% figure appears as in "locked to 3.8Ghz" THG test,what a surprise
🙂
itunes(1 thread)
5800K-1:27 or 87s
8150 -1:42 or 102s
Difference is 87/102=0.85 or 15%
I also mentioned Lame numbers before but here they are for reference purposes:
Lame(1 thread)
5800K-1:58 or 118s
8150 -2:10 or 130s
Difference is 118/130=0.9 or 10%
Trinity has no L3 cache. Whether it helps or not is debatable but it's more costly to go to slower RAM (when upper caches miss) than to go to much faster L3.
