Tuesday, July 5, 2011

Ivy Bridge ES benchmarked(China again)

Update: I have examined coolaler's latest results from his Sandy Bridge i3 sample downclocked to 1.8Ghz at the bottom of the original blog post.His numbers are very low for SB and skew the results in Ivy Bridge's favor from 10 to 17%(could be bios glitch in SB system).

Coolaler's forum has a new Ivy Bridge(IB) leak.This time we have an ES which is 2C/4T,working at 1.8GHz and having 4MB of L3(vs 6MB of L3 in current SB CPUs).

My analysis ,copied from my post at SA forum.

HW used :
IB ES 2C/4T @ 1.8Ghz,4MB L3
SB @ 1.8Ghz, 4C/4T,6MB L3

Tests:

CPU mark99-purely single threaded integer benchmark,cores don't matter.
IB @ 1.8Ghz gets 278pts
SB @ 1.8Ghz gets 276pts
Practically the same results,tiny difference that fits in margin of error.

SPi 1M-purely single threaded benchmark,with mixed int and fp code.
IB @ 1.8Ghz gets 22.19s
SB @ 1.8Ghz gets 20.8s
SB at the same clock is 6.6% faster than IB ES.

Cine 11.5- single and multithreaded benchmark,uses a lot of fp SSE instructions. In this particular case,tester used MT benchmark.In order to figure out how SB which has 2C/4T performs in it,we can find a comparable i3 SB retail chip and its score in C11.5 and then scale it down to 1.8Ghz clock.

IB @ 1.8Ghz 2C/4T gets 1.81pts
SB @ 1.8Ghz 4C/4T gets 2.61pts
SB 2120 model which is 2C/4T @ 3.3Ghz and no Turbo gets 3.19pts.
Compensate for 1.8Ghz clock speed : 3.19x1.8/3.3=1.74pts
So to recap:
IB @ 1.8Ghz 2C/4T gets 1.81pts
SB @ 1.8Ghz 2C/4T gets 1.74pts
SB at the same clock is 4% slower than IB ES.

Summary:
IB ES is roughly on par with SB at the same clock and with the same number of cores and threads.There are some minor differences so actual scores vary from it being 6% slower than SB(SPI) ,being practically equal to it(CPU mark99) to being 4% faster(C11.5) at the same clock/core/thread configuration.All this is based on early ES,so clock speeds will move up and probably Turbo will be a bit more aggressive with IB.This could lead to better overall scores,but knowing intel's ES from the past,they usually perform within 5% from retail parts(at the same clock).This practically leaves only clockspeed/Turbo as a variable.

edit: Note also that there will be mainstream IB parts with 2C/4T and 4C/4T or 8T configuration which will have more L3 onboard,maybe matching or surpassing SB's 6MB of L3.This could bring a few additional % of performance on average in desktop apps.


UPDATE:
Coolaler updated his thread with comparative scores between retail SB 2C/4T @ 1.8Ghz and IB 2C/4T @ 1.8Ghz.But there is some problem with his SB i3 system.

Coolaler's  SB i3 numbers are way slower than what reviews show for retail parts that work @ 3.1GHz (if we would scale them down to 1.8Ghz). Something is wrong with his SB i3.
Example:
In C11.5 his SB i3 2120 gets 3.19pts @ 3.3Ghz here,so it should roughly score 1.74pts.
This lines up with i3 2100 score which bit tech reviewed here.Bit tech got 2.97pts @ 3.1ghz,so i3 2100 @ 1.8Ghz should be getting 1.72pts,practically the same as hardware canucks SB i3. One more example is here which shows a scaled down i3's score of 1.73pts(at 1.8Ghz). All 3 of these results are ~11% better per clock/per core than what coolaler gets with his sample... Or in numbers : 1.72pts(average from 3 sources)/1.55pts(coolaelr's SB i3)~=1.11 or 11%.

Let's see what super pi shows us. Here we have i3 2120's result in super pi 1m. It scores 11.9s @ 3.3Ghz. At 1.8Ghz the SB 2C/4T would get ~21.81s. This again is faster than what coolaler's i3 is getting .The difference is 23.83/21.81~=1.09 or 9%. So coolaler's i3 is 9% slower in super pi than other retail SB i3's,per clock.

We move on to CPUmark99.I found this result of retail i3 2100 on same(coolaler's !) forum here. i3 2100 @ 3.1Ghz scores 475pts. So i3 @ 1.8Ghz should be getting : 475x1.8/3.1~=275pts,or 275/235=1.17 or 17% faster than coolaler's own i3 chip. Hmm,similar pattern occurs. His sample downclocked is from 9% to 17% slower than other i3 SB retail chips out there.

Conclusion : coolaler's i3 @ 1.8Ghz is performing anywhere from 9 to 17% slower than "normal" i3 SB CPUs if they ran at same 1.8Ghz clock. This skews the results a bit and shows his IB @ 1.8Ghz being slightly faster than his SB i3 at the same clock. The reason could be some bios glitch or power management issue.

No comments:

Post a Comment