Sandy Bridge i7 & i5 Reviews

[waway625]

the performance jump from 1156/1366 Core i7 to SB 2500/2600 is like the performance jump we saw from Q6xxx to Nehalem no?

[Manicdan]

Na its the same for dedicated gpus, when i run boinc in backgorund while I play TF2, my avg fps is a lot lower then when I stop boinc. And thats with a hexacore and a HD5870.

the test was compared across mulitple platforms, and only SB had losses in the game when cinebench was using up more cores

i honestly woulnt call this bad at all considering its trying to manage the TD of 2 components with the same heatsink, this is actually how almost every laptop will work in the near future when fusion based cpus are most of the market (not saying AMD fusion, just all cpu+gpu packages in laptops)

it will require testing to be much more detailed in how the results were obtained instead of just reading the FPS and calling it a day. i hope this means we will see really nice utilities from both AMD and Intel for showing us the turbo speeds and TDP usage for each part of the package during gameplay. when everything is perfect hopefully in the next 10 years, the cpu and gpu will both be running at 100% load in gaming, because the cpu will be downclocked until its maxed out, and the gpu overclocked, with per frame balancing to maintain the 100% load on both parts of the chip.


what this does mean though is that you probably do not want to have your cpu encoding in the background while you play warcraft if your using SBs integrated GPU

[Hornet331]

How many cores can TF2 utilize ? If you limit boinc to 4 cores and TF2 uses up to 2 cores then you shouldn't see much of a drop off.

TF with MT uses 2.x threads, but you allways notice a drop since the BONIC projects are heavy on the cache. It was the same with every cpu i owned till now (C2D, C2Q, and even on X4). You notice it less in gpu bound games like crysis, but you still have a considerable fps drop.

That SB has an additional concern with the TDP limit isn't exactly helping this either.

[kl0012]

Did anyone noticed SB hardware transcode test?
http://www.anandtech.com/show/4083/t...2100-tested/9#

It's just me or picture with a police car in fact looks much better after SB transcode then after x86 & ATI transcode (not to mention terrible Nvidia quality). It seems like some kind of AA was applied to the picture during transcoding.

[Mechromancer]

There is no Intel vPro on their K processors? That is almost a deal breaker for me!

[informal]

QuickSync is pretty good at what it's designed for but there is one big issue with it : you can't use it on desktop if you have discrete gfx card in your system. AT indicates that intel may be preparing some sort of switch (software?) later on,so it may be resolved. On notebook this is not a problem,but majority of users won't use their notebooks for such a task anyway(or they will use it only when they really have no other choice).

[Lanek]

Did anyone noticed SB hardware transcode test?
http://www.anandtech.com/show/4083/t...2100-tested/9#

It's just me or picture with a police car in fact looks much better after SB transcode then after x86 & ATI transcode (not to mention terrible Nvidia quality). It seems like some kind of AA was applied to the picture during transcoding.

He explain it on the article, Quick Sync loose some sharpness ( nothing to do with AA ), this help for this image ( the police car ), but normally it's not desirable to get this lost of sharpness.

[AKM]

the performance jump from 1156/1366 Core i7 to SB 2500/2600 is like the performance jump we saw from Q6xxx to Nehalem no?

http://www.computerbase.de/artikel/p...formancerating

According to these charts clock for clock, the jump from Yorkfield to Nehalem is a little higher.

[kl0012]

He explain it on the article, Quick Sync loose some sharpness ( nothing to do with AA ), this help for this image ( the police car ), but normally it's not desirable to get this lost of sharpness.

Yes, i saw it. But in fact "lost of sharpness" affected nothing except much better looking edges (especially white line on the police car) after SB hardware transcoding. So i guess this is not just a "lost of sharpness" but some kind of hardware post-processing mentioned at IDF.

[Manicdan]

Yes, i saw it. But in fact "lost of sharpness" affected nothing except much better looking edges (especially white line on the police car) after SB hardware transcoding. So i guess this is not just a "lost of sharpness" but some kind of hardware post-processing mentioned at IDF.

i think the batman example looked great
the bond one however was identical between the x86 and 6870, yet the batman one was clearly having different brightness between the two, so i dont know if a setting got messed up or what

this is a nice feature to have and i hope that amd fusion bring competition here so we get better and better quality and speeds in the next few years

also did they ever compare file sizes between all the end results?

[SKYMTL]

Finally: http://www.hardwarecanucks.com/forum...rs-review.html

[DarthShader]

the test was compared across mulitple platforms, and only SB had losses in the game when cinebench was using up more cores

At first, I'd think it's a trival case of being starved for memory bandwidth. But if you say only SB showed this behavior, then I'd say it's the L3 cache... so memory starving after all. The IGP sharing the cache seems to be a big deal for SB - Dr.Who even suggested the IGP will challenge Llano thanks to it. First step towards this goal has been done - the HD5450 is challenged.

There are surely some sophisticated algorithms taking care of dividing the available cache among the IGP and cores, depending on needs. If your cores aren't heavily taxed, then most cache is probably reserved for the IGP. At full load, the cores need the cahce to handle it, not much left for IGP, effective bandwidth drops, performance plumets.

QuickSync is pretty good at what it's designed for but there is one big issue with it : you can't use it on desktop if you have discrete gfx card in your system. AT indicates that intel may be preparing some sort of switch (software?) later on,so it may be resolved.

Well, LucidLogix already announced a software switching sollution for SB. But to work, you will need a monitor plugged in the mobo's video socket. This means, you need to have a mobo with a video socket. (duh)

That means P67 users will have to use traditional x86 encoding on their machines.

[Raqia]

Do any of these reviews feature a GTA IV bench?

[[XC] Hicks121]

Finally: http://www.hardwarecanucks.com/forum...rs-review.html

Thank you!

[chuchnit]

Finally: http://www.hardwarecanucks.com/forum...rs-review.html

Any chance you guys will do a face off of top of the line boards? Say P67A-UD7 vs Maximus IV Extreme vs MSI Big Bang Marshal or whatever is their top end at launch? I am interested in multi-GPU performance since manufacturers are getting creative with the PCIe.

[Mad1723]

Do any of these reviews feature a GTA IV bench?

Hardware.fr does feature that test, quite a nice improvement over i7 9xx

[clayton]

I'm assuming you can't overclock on the H67 chipset. Such a shame.

[rozzyroz]

did anyone overclock it and check the igp performance vs stock?

[cowie]

Do all the 2600k chips have the same max multi?
Sorry i got alittle lost i all those reviews.

[Olivon]

Do all the 2600k chips have the same max multi?
Sorry i got alittle lost i all those reviews.

No, they don't seem to have the same max multi cowie

[Manicdan]

did anyone overclock it and check the igp performance vs stock?

anand did, i think

all i remember was ~85% OC saw just ~19% more perf

[cowie]

No, they don't seem to have the same max multi cowie

Do i pay less for the one that has a weak multi?

[kl0012]

anand did, i think

all i remember was ~85% OC saw just ~19% more perf

You mean this one:


I think there is Anand's typo here. I suppose, the clock was increased from 1100 MHz to 1550Mhz (not from 850 to 1550) which is 40% clock increase.

[Manicdan]

You mean this one:
[image remove]

I think there is Anand's typo here. I suppose, the clock was increased from 1100 MHz to 1550Mhz (not from 850 to 1550) which is 40% clock increase.

yup that one,

i think its accurate

just look at HD2000 to HD3000, its 2x larger and only like 40-60% faster. some rare cases its near double, but most seemed like a half jump

i really think they reached their limits on the GPU without making major changes

[Lanek]

Finally: http://www.hardwarecanucks.com/forum...rs-review.html

Nice job ... 1 thing, on hardware used listdescription you write LGA1156 ( instead of LGA1155 ). too bad you don't have a 950 or 975 for compare. more accurate of the jump between the 920 and 980x ( who show his limit on memory test vs the 950-975)