Late week i7 920 C0 vs D0 920

[harbin]

3844a361 c0 did 4.5g on air with HT off, haven't got it stable it yet. Air cooling can't handle the heat at 4.5g for sure, temp goes up to 100c within 1 minute after LinX.

So far, I haven't seen 4.5g D0 stable on air either, HT on or off.

http://i43.tinypic.com/73h5qc.png

 

[bipartisanpwnage]

from what i have seen d0's are as variable as c0's but the d0's seem to be reaching a slightly higher than c0's average speed. a lot more people are coming out saying their d0s are hitting 4.3ghz w/ ht on low volts than c0's, but you are also getting alot of people having trouble getting 3.8ghz w/ d0. nice c0 though

 

[Rick James]

I think it's a push. My C0 overclocks fine. It all depends on how good your setup is

 

[Idontcare]

Topic Title: Late week i7 920 C0 vs D0 920
Topic Summary: are they same?

No two C0's are the same, let alone a C0 and a D0 where the mask-set itself has been changed. Entropy (process variability) ensures it.

You are really asking whether there is considerable overlap in the distribution of clockspeed vs. Vcc capabilities (shmoo plot) of the more recent C0 chips and the D0 chips.

Provided the overlap is non-zero (and it most assuredly is non-zero as we are talking about the same underlying 45nm process technology for both steppings) then you always have a non-zero chance of snagging a C0 that outperforms a D0 (C0 is from the +sigma side fot the mean, D0 from the -sigma side of the mean, etc) or snagging a D0 that performs worse than your current C0.

This is why YMMV is applied to all end-user experiences.

 

[chizow]

Well first off, you're looking at a pretty specific subset of people testing for Windows boot + SuperPi at high clockspeeds. I think most people actually shoot for stability (Prime/OCCT/Linpack) before claiming stable speeds.

That said, I'll still take my chances on D0 being the better clockers still, as we have a very small sample of D0 with very good results so far, compared to a very large sample of C0 with relatively few similarly good results. We also know overall yields were good enough to shift clock distribution up another speed bin/notch.

Although I will say that seeing some of these bad early D0 results has reduced my urge to upgrade to a D0 as soon as they become available. Perhaps the 920 D0s got the dregs in the 940/975 ramp up?

 

[Idontcare]

Originally posted by: chizow
Perhaps the 920 D0s got the dregs in the 940/975 ramp up?

We've definitely become spoiled by Intel's 65nm/45nm Core2 clockspeed/inventory management which down-binned lots of fully capable higher speed CPU's to the lower clocked SKU's.

Historically we did not expect silicon from all SKU's to be capable of reaching the same peak GHz plateau, we expected the higher stock GHz chips to peak at a higher OC GHz than the lower GHz stock chips.

From the sounds of it, we may see some elements of history repeating with the D0's.

My personal suspicion here is that this has less to do with C0 vs D0 binning and more to do with the release of Nehalem EP and the supply/demand pressures this is putting on the binning/inventory management of the i7 920's.

 

[Markfw]

My personal suspicion here is that this has less to do with C0 vs D0 binning and more to do with the release of Nehalem EP and the supply/demand pressures this is putting on the binning/inventory management of the i7 920's.

Could you elaborate on this ?

 

[harbin]

Originally posted by: Markfw900

My personal suspicion here is that this has less to do with C0 vs D0 binning and more to do with the release of Nehalem EP and the supply/demand pressures this is putting on the binning/inventory management of the i7 920's.

Could you elaborate on this ?

Yes, please.

 

[chizow]

Originally posted by: Idontcare

Originally posted by: chizow
Perhaps the 920 D0s got the dregs in the 940/975 ramp up?

We've definitely become spoiled by Intel's 65nm/45nm Core2 clockspeed/inventory management which down-binned lots of fully capable higher speed CPU's to the lower clocked SKU's.

Historically we did not expect silicon from all SKU's to be capable of reaching the same peak GHz plateau, we expected the higher stock GHz chips to peak at a higher OC GHz than the lower GHz stock chips.

From the sounds of it, we may see some elements of history repeating with the D0's.

My personal suspicion here is that this has less to do with C0 vs D0 binning and more to do with the release of Nehalem EP and the supply/demand pressures this is putting on the binning/inventory management of the i7 920's.

Yep that certainly makes sense, as there's no lower Bloomfield-based SKU than the 920. Even though there's lower clocked server parts with less cache, enterprise parts always get binning preference over desktop, except for maybe EE. It does sound like they're straining to meet demand on both the new desktop SKUs (940/975) and all the EP parts as you mentioned.

i7 is very different than recent Core 2 examples though, as Core 2 had really low-end parts like the Pentium Dual Core (I played around with a 1.35VID E2140 myself). I guess the oddity is Q6600 was the lowest Quad SKU, but still overclocked very well compared to the higher binned SKUs. Perhaps all Quads got their two cores from a higher speed bin to begin with.

 

[lopri]

I think what IDC says is that C0 i7 940 and 965 aren't selling well and being pushed to be labed as 920. And all Xeons are D0 so basically whatever left-over C0 chips will be sold as 920.. meaning there will be gems among them.

 

[lopri]

BTW, the OP's 920 is simply incredible. It really whets my appetite!

 

[Idontcare]

Originally posted by: harbin

Originally posted by: Markfw900

My personal suspicion here is that this has less to do with C0 vs D0 binning and more to do with the release of Nehalem EP and the supply/demand pressures this is putting on the binning/inventory management of the i7 920's.

Could you elaborate on this ?

Yes, please.

Read Chizow's follow-up post as he restates it nicely.

To add, basically what I am saying is that the C0 920 SKU was/is populated with both the garbage binouts (high VID, high power dissipation, poor OC'ing) as well as the intentionally underclocked chips (too much supply) that could have just as easily been a stock 940 or a stock 965 because the demand for those 940's and 965's were already met with ample supply.

Now along comes Intel releasing the Xeon Nehalem, suddenly the demand side of the equation doubles, triples, quadruples for those chips with desirable GHz/VID capabilities and this portion of the i7 920 distribution is suddenly reallocated towards the population of those Xeon SKU's (either lower GHz and even lower VID or higher GHz with good VID).

Caveat: I am using VID as a catch-all label to basically categorize chips that have good power consumption characteristics for a given clockspeed, I am not necessarily using the term VID to strictly and only speak to the specific Vcc setting of a given Nehalem.

This is of particular relevance to the i7 920 D0's as all Nehalem Xeons are D0's, no C0's. Just as all 940's and 965's are C0, no D0's.

http://processorfinder.intel.c...ProcFam=528&SearchKey=

So getting a 920 D0 requires an entirely different set of expectations as to the likely capabilities of the silicon in the chip when attempting to extrapolate expectations set from 950's and 975's as verse to how expectations were set for 920 C0's by extrapolation of OC'ing capability of 940's and 965's.

Is the horse dead yet?

 

[aigomorla]

u know what i find funny.

is how people are trying to go for 4.5ghz on a 920 with 1600mhz ram.

Thats what i find FUNNY! <--- if you dont understand what im getting at, you fail on i7 overclocking.

Big clue, your MAX QPI DRAM is way exceeded. Your MT's is also way exceeded.


Few things id like to share with those of you who are on i7s.

From my experience...

1. they clock better on a odd multi vs a even multi.
2. Your gonna exceed your QPI/DRAM speed when you try to go after 4ghz + for 1600mhz ram.
3. 1.65vcore is max dram voltage only if your cpu vcore is 1.15.

 

[harbin]

Originally posted by: aigomorla
u know what i find funny.

is how people are trying to go for 4.5ghz on a 920 with 1600mhz ram.

Thats what i find FUNNY! <--- if you dont understand what im getting at, you fail on i7 overclocking.

Big clue, your MAX QPI DRAM is way exceeded. Your MT's is also way exceeded.


Few things id like to share with those of you who are on i7s.

From my experience...

1. they clock better on a odd multi vs a even multi.
2. Your gonna exceed your QPI/DRAM speed when you try to go after 4ghz + for 1600mhz ram.
3. 1.65vcore is max dram voltage only if your cpu vcore is 1.15.

You are saying I should use higher RAM or put RAM frq. lower? I'd like have ddr3 2000, but it's too expensive.

 

[lopri]

Never worry I didn't comprehend what aigo says, either.

I haven't touched the i7 but looking at your screenshot, your QPI multi is x18, CPU multi x21, memory divider 1:3. (DDR3-1290) NB is running @2580 which is x2 of memory frequency. Everything looks right to me. Granted your NB is running somewhat slow compared to the CPU (due to 1:3 divider), but apparently there is nothing stopping it?

If your divider were to be 1:5 (like from other screenshots I've seen) your memory would be 1075MHz, thus putting the NB frequency to 4300MHz. Yet like you said that'll cost $$$.

 

[aigomorla]

Originally posted by: lopri
Never worry I didn't comprehend what aigo says, either.

Well make things simple.

its like trying to get 400+ fsb on the first gen quads. with DDR2 667 ram.

Which is one of the reasons why i bought my classified, since it handles higher QPI a lot better then the normal boards.

and i said with the intensions of 4.5ghz.

4.5ghz is very hard. Linpack stable, ive only seen mine with HT ON.

 

[harbin]

Originally posted by: lopri
Never worry I didn't comprehend what aigo says, either.

I haven't touched the i7 but looking at your screenshot, your QPI multi is x18, CPU multi x21, memory divider 1:3. (DDR3-1290) NB is running @2580 which is x2 of memory frequency. Everything looks right to me. Granted your NB is running somewhat slow compared to the CPU (due to 1:3 divider), but apparently there is nothing stopping it?

If your divider were to be 1:5 (like from other screenshots I've seen) your memory would be 1075MHz, thus putting the NB frequency to 4300MHz. Yet like you said that'll cost $$$.

The memory I am using ocz ddr3 1600 really need a lot voltage to get to 1600Mhz, which is not I wanted as I already have a lot of heat from CPU itself.

 

[harbin]

Originally posted by: aigomorla

Originally posted by: lopri
Never worry I didn't comprehend what aigo says, either.

Well make things simple.

its like trying to get 400+ fsb on the first gen quads. with DDR2 667 ram.

Which is one of the reasons why i bought my classified, since it handles higher QPI a lot better then the normal boards.

and i said with the intensions of 4.5ghz.

4.5ghz is very hard. Linpack stable, ive only seen mine with HT ON.

What you have achieved is amazing, but the 975 is very expensive, way beyond 95% of PC users. And your 4489Mhz is using a little bit slow qpi link, and I might be wrong, is qpi link speed or width also very important for performance?

 

[Idontcare]

Originally posted by: harbin
What you have achieved is amazing, but the 975 is very expensive, way beyond 95% of PC users. And your 4489Mhz is using a little bit slow qpi link, and I might be wrong, is qpi link speed or width also very important for performance?

This techreport article looked specifically at answering this question.

http://techreport.com/articles.x/16044/3

You want to take note of the differences in the graphs between the "Core i7-940 (sim)" versus "Core i7-940" as the only difference is the uncore clockspeeds.

To some extent, you're simply forced to turn up the uncore clocks if you want to run faster memory with the Core i7. But if you own a Core i7 and want to tune it for absolutely optimal performance, you might want to push the uncore clock as high as possible, even if it means using lower QPI and memory multipliers to keep things stable.

 

[harbin]

Originally posted by: Idontcare

Originally posted by: harbin
What you have achieved is amazing, but the 975 is very expensive, way beyond 95% of PC users. And your 4489Mhz is using a little bit slow qpi link, and I might be wrong, is qpi link speed or width also very important for performance?

This techreport article looked specifically at answering this question.

http://techreport.com/articles.x/16044/3

You want to take note of the differences in the graphs between the "Core i7-940 (sim)" versus "Core i7-940" as the only difference is the uncore clockspeeds.

To some extent, you're simply forced to turn up the uncore clocks if you want to run faster memory with the Core i7. But if you own a Core i7 and want to tune it for absolutely optimal performance, you might want to push the uncore clock as high as possible, even if it means using lower QPI and memory multipliers to keep things stable.

Good info!

 

[aigomorla]

i dont think your system will hold linpack with QPI that high.

also @ 100C your cpu should either

1. Throttle, actually i think your PAST throttling now.
2. Shut down.

I dont think your system will pass a 30min prime even.

Care to prove me wrong on the prime?? im curious if it will pass or not.