IDF Sandybridge overclocking - 4.9Ghz

[Kuzi]

The SB OC clocks are pretty good, but the 20-25% IPC increase over Bloomfield (if true) is REALLY impressive!

 

[extra]

What it is telling me is to temper my expectations on the Bulldozer clock speed. If Intel has a better HKMG process than GF, and still isn't seeing much higher clocks than they got from 45nm, I am not sure how the GF process will do.

I believe that the SOI and perhaps better low-k might help mitigate this, as well as the difference in design (BD appears to be designed for high clock speeds). Even so, my expectation that we would see a 4GHz stock SKU from AMD by the end of 2011 is sufficiently tempered.

I am not an expert on this, but if IDC seems to think this is low, then I know to expect even less from GF due to the inherently weaker gate first HKMG implementation.

BD might clock super high, who knows. Might clock super low. Hard to say. The rumors are that it is designed to clock high...Doesn't really matter though, since we know literally nothing about it's IPC and such (other than IPC will be higher than phenom II).

I wouldn't be too surprised if we do see a stock 4ghz cpu from AMD, though, in 2011 (and from Intel). I bet 4ghz might be a reasonable stock clock for the highest sku 32nm phenom II (given how easily thubans hit 4ghz).

 

[Martimus]

BD might clock super high, who knows. Might clock super low. Hard to say. The rumors are that it is designed to clock high...Doesn't really matter though, since we know literally nothing about it's IPC and such (other than IPC will be higher than phenom II).

I wouldn't be too surprised if we do see a stock 4ghz cpu from AMD, though, in 2011 (and from Intel). I bet 4ghz might be a reasonable stock clock for the highest sku 32nm phenom II (given how easily thubans hit 4ghz).

Yeah, I was probably a little to rash with my pessimism. Just taking a minute to think about it, I remember why I believed it would clock to 4GHz in the first place.

I hope that we will see a higher stock clocked Sandybridge part than the 2600K. I do not like to overclock a computer that I am relying on being stable, and it seems that there may be a lot of upward room for Intel to release higher clocked SKU's based on this test.

Actually now that I have taken a minute to think about it, I shouldn't put much faith in this test at all, and I will wait until the products get released. My mind seems to be pretty scrambled lately. Things are slipping my mind that normally wouldn't ...

 

[PlasmaBomb]

25% IPC boost is crazy amazing...especially when that comes on top of Bloomfields already impressive IPC.

You think this boost is basically coming from the addition of that L0 uop $?

Or are we looking at a special compile of Cinebench that takes advantage of some of the new ISA extensions?



http://en.wikipedia.org/wiki/Strain_engineering

It is actually yet another one of those major "pros" to the lengthy list of performance advantages that are intrinsic to doing a disposable gate (also called replacement gate or "gate last") integration scheme.

Don't get me wrong, my rant above wasn't to say that 4.9GHz on air isn't great, means we'll probably be seeing some 6GHz water-cooled rigs in the forum. Hmmm...6GHZ...

6GHz :biggrin:

God almighty .

NDA is laughable / We have proof see what it says . LOL . watch the the test. It shows 8 threads running . It could only be 1 SB .

Well it was a pretty stupid NDA to begin with... lets see, we know it is a K series processor of which the choice is 2600K and 2500K right?

 

[ElFenix]

don't they always show off a huge overclock? i'll bet they had P4 chips running near that way back when.

 

[ilkhan]

I hope that we will see a higher stock clocked Sandybridge part than the 2600K. I do not like to overclock a computer that I am relying on being stable, and it seems that there may be a lot of upward room for Intel to release higher clocked SKU's based on this test.

The chips may be capable, but I doubt Intel is going to release >2600k priced chips ($300-350, although didn't the thing say <300 euro?) on s1155. At the higher price points Intel thinks you should be on s1366 (and later s2011).

 

[Idontcare]

The chips may be capable, but I doubt Intel is going to release >2600k priced chips ($300-350, although didn't the thing say <300 euro?) on s1155. At the higher price points Intel thinks you should be on s1366 (and later s2011).

If I can get a 2600K for $300-$350 I'm gonna bust out my VapoLS unit again...sitting in the basement for a couple years now. Nothing loves a VapoLS like a CPU with an unlocked multiplier.

 

[Acanthus]

If the average is above 5.0ghz on water with a sub $300 part, that will be enough for me to settle with quad core again and not wait for hex.

I'd be looking at the 50% performance increase that I usually wait for.

 

[Acanthus]

The SB OC clocks are pretty good, but the 20-25&#37; IPC increase over Bloomfield (if true) is REALLY impressive!

I'm very skeptical of sweeping statements like that. It is likely updates to microcode that are driving that increase, and it isn't across the board so 20-25% is misleading.

It is likely something like 3-25% depending on the app. Encoding, encryption, and compressing data are the largest gains from what i've seen.

 

[deputc26]

I'm very skeptical of sweeping statements like that. It is likely updates to microcode that are driving that increase, and it isn't across the board so 20-25% is misleading.

It is likely something like 3-25% depending on the app. Encoding, encryption, and compressing data are the largest gains from what i've seen.

Honestly I'd still be impressed if it was 12.5% average gain, I was expecting Sandy Bridge to just be Nehalem with on-die GPU and minor tweaks.

 

[DrMrLordX]

If the average is above 5.0ghz on water with a sub $300 part, that will be enough for me to settle with quad core again and not wait for hex.

I'd be looking at the 50% performance increase that I usually wait for.

Isn't the 2600K supposed to be around $500-$550? The 2500K is supposed to be in the $200 range or something?

 

[OCGuy]

I think 5.0 on water is setting the bar a little high. I would be amazed, but not disappointed with less.

 

[jihe]

For a processor that people were panicking about not being able to overclock, I'd say 4.9GHz on air is pretty good.

If I could hit 4.5GHz on my own setup I'd be very pleased.

You now have to pay a premium to overclock, that's what the fuss was all about.

 

[IntelUser2000]

Isn't the 2600K supposed to be around $500-$550? The 2500K is supposed to be in the $200 range or something?

The roadmaps say 2600 and 2600K are successors to 870/875K/950, and they are priced at $342 for the K versions and $294 for the regular ones.

I'm guessing then 2500=$190 2500K=$210

 

[OCGuy]

The roadmaps say 2600 and 2600K are successors to 870/875K/950, and they are priced at $342 for the K versions and $294 for the regular ones.

I'm guessing then 2500=$190 2500K=$210

Are we talking about lots or retail?

 

[IntelUser2000]

Huh? Those are the official 1ku pricing. I assume since the Sandy Bridge followups are in the same price bracket they'll be priced similar.

 

[DrMrLordX]

oh only $342? Alright, nm.

edit: still want to know if uncore/qpi multis are unlocked for the K-series chips.

 

[ilkhan]

The s1155 chips dont have a QPI bus. I'm not sure what uncore ("system agent") would include besides memory (on its own multi), so probably not that relevant anymore anyway.
Yeah, the sys agent is mem controller, DMI bus, PCI-E bus, and "display engine"(?). Nothing that seems critical to me.

 

[DrMrLordX]

Memory controller. Increasing the memory controller's clockspeed can improve memory bandwidth/latency, which is one of the reasons why uncore overclocking has helped Nehalem chips in the past. It also helps K10 chips (NB clockspeed) even when L3 is not present.

 

[ydnas7]

Honestly I'd still be impressed if it was 12.5&#37; average gain, I was expecting Sandy Bridge to just be Nehalem with on-die GPU and minor tweaks.

apart from one being a 1% performance gain the rest are between 10% to 20% performance gain. include the odd low one and its at 13% increase in IPC. Now this excludes the effect of Sandybridge's turbo and is measured against a part using turbo, this probably ignores the benefits that SB brings to encoding, vectors, encryption. It excludes the benefit SB has with an IMC over arrandale, and it was with 6mb cache available, not the 8mb on the i7 models

 

[Kuzi]

I'm very skeptical of sweeping statements like that. It is likely updates to microcode that are driving that increase, and it isn't across the board so 20-25% is misleading.

It is likely something like 3-25% depending on the app. Encoding, encryption, and compressing data are the largest gains from what i've seen.

I agree, my expectation was that SB would be ~10% faster per clock on average. Of course apps that will use the new instructions will get a much larger jump in performance.

 

[ydnas7]

I agree, my expectation was that SB would be ~10&#37; faster per clock on average. Of course apps that will use the new instructions will get a much larger jump in performance.

the early anand results were 13%, but this was against a part with turbo enabled, and with full 8mb cache, so between taking comparing a SB without turbo to Nehalem without turbo, the result is probably much closer to 20% increase, then ipc increases again if the full cache is available so an increase in IPC is probably in the region of 20-25% for an apple to apple comparison. use code that can use the new instructions and it may jump a lot higher, or it may change at all.
SB vs arrandale gets an IMC and all its goodness...

SB also has quite different performance in HT compared to Nehalem, it averages out to be similar, but per application it will range from being half less effective to half more effective than Nehalem, clock per clock

 

[Idontcare]

SB also has quite different performance in HT compared to Nehalem, it averages out to be similar, but per application it will range from being half less effective to half more effective than Nehalem, clock per clock

Awesome table summary above :thumbsup: thanks for contributing that

Why the increased sigma in ROI for HT on SB?

If anything, with there being even more resources per core in SB vs. Nehalem, I would have expected the mean value to increase and the sigma to decrease. (higher HT performance on average, less penalties for the extraneous corner-conditions)

 

[Bman123]

4.9ghz on air is very nice, the water guys will hit over 5ghz easily. I dont see how anyone can complain about SB hitting 4.9 on air. Sure it could of been higher but it also could of been lower. This should mean the avg joe schmo here should hit 4ghz+ easily and that is very nice

 

[edplayer]

they did show the cooler used, it was a big heatpipe one, but not as big as a noctua or something lol. No doubt this was a completely cherry picked sample, but still . . . on air.

This means we'll probably see 4.5ghz on air become common, and maybe 5ghz on water become commonplace. Yes pwease.

fudzilla is claiming that it was done on a stock cooler:

http://www.fudzilla.com/processors/processors/processors/intel-overclocks-sandy-to-49ghz