Nehalem Overclocking @ XS

[Charles Wirth]

Day 1 done, some huge results and learned a lot.

Dinos, bclock is ratio'd into the multiple, change the multiple and try more bclock.

I was on the EX58 Extreme @ 208 bclock (wprime 1024) today, bios F4E = bugged multi.

I had the Smackover at 210 bclock, easy once you you know whats working.

[Movieman]

Day 1 done, some huge results and learned a lot.

Dinos, bclock is ratio'd into the multiple, change the multiple and try more bclock.

I was on the EX58 Extreme @ 208 bclock (wprime 1024) today, bios F4E = bugged multi.

I had the Smackover at 210 bclock, easy once you you know whats working.

Yell when your "guest" leaves and I'll be over..

[Oliver]

sounds awesome Charles

Francois and you must have a blast i reckon

cant wait to see some numbers

keep pushing it

[dread77]

Day 1 done, some huge results and learned a lot.

I was on the EX58 Extreme @ 208 bclock (wprime 1024) today, bios F4E = bugged multi.

Not fair at all.... Hey Francois, are you interested for a visit in sunny Greece???

We want all settings used, all bios versions used, detailed feedback in everything...

[SF3D]

What are you talking mate

QPI is Quick path interconnect speed. It is marked as GT/s Gigatransfers per second. You see this value in CPU-Z front page "QPI link" At stock speed it is saying ~3200MHz, when it should say 6400GT/s(2x3200GT/s). When you raise BCLK this value will raise too and with lowest QPI multiplier (36x) it will go up to ~8000GT/s before it locks. Now if we would have lower multipliers than 36x we could get our BCLK higher before that ~8000GT/s limit.

If you remember AMD 64 days and 1X to 5X HT multipliers, you will see what this issue is. Back then you had to lower HT multiplier to 4X or 3X to get your HT speed up to 200 to 250.

After all this processor is designed for servers and these really high QPI speeds were not what they had in mind. Maybe this will be fixed with new chipsets or 1160 processors. If QPI multipliers on X58/core i7 are limited to 36x on hardware level, there is nothing we can do. Just LN2 to CPU and Chipset and we might see some scaling

Are we on the same page now

I quote myslef

I got answer to my question about hardware level limits. QPI multipliers are locked in CPU's registry, so there is no way to unlock them. That would mean higher BCLK clocks, but then the difference between basic and extreme edition cpu's would be thinner. Intel don't like that of course.

This case is clear for me now

Bad news for 920/940 overclockers

[Fr3ak]

Thanks for the info! Confirmed what I was thinking =)

That limit is not generally bad for overclockers, but for people using extreme cooling methods it is.

[qurious63ss]

I quote myslef

I got answer to my question about hardware level limits. QPI multipliers are locked in CPU's registry, so there is no way to unlock them. That would mean higher BCLK clocks, but then the difference between basic and extreme edition cpu's would be thinner. Intel don't like that of course.

This case is clear for me now

Bad news for 920/940 overclockers

Hmm... then why are all these people in here getting >4ghz and some extremers over 5ghz on 920/940? I guess we must all expect 6ghz to be good nowadays.

[dinos22]

wish Intel would just allow us to remove those temps sensors

or if someone could figure out a mod so that we can run these CPUs at proper CPUs and none of this retarded below 5GHz on LN2

[BenchZowner]

There's surely a way, and with some good theory and tries we could make it.
But who has lots of i7s to try it out ?
Anybody willing to risk his CPU/s ?

[qurious63ss]

wish Intel would just allow us to remove those temps sensors

or if someone could figure out a mod so that we can run these CPUs at proper CPUs and none of this retarded below 5GHz on LN2

Don't worry I'm sure Fugger is taking notes. I'm sure he'll have a sticky up by the weekend on how to calibrate those pesky thermal diodes.

[Viss]

I bet somewhere in Greece someone is working on it already

[dread77]

wish Intel would just allow us to remove those temps sensors

or if someone could figure out a mod so that we can run these CPUs at proper CPUs and none of this retarded below 5GHz on LN2

I would also like to know how to go through this temp barrier.

When it comes to mods, Hipro is the man. Though, I am not sure if he is testing a Nehalem setup at all.

[qurious63ss]

Hey Fugger, tell Francois we'll all pitch in $20 if he spills the beans.

[Fr3ak]

I don't think you can reprogramm the CPU with "ordinary" tools. From what Francois posted so far, only Intel would be able to unlock the CPUs.

[SF3D]

I can run my other chip -110c and it is not helping at all. Temperature is not the limiting factor in some cases. If you have really good cpu, then it might help something. I think next stepping will be improved. Until then, those guys with handpicked amperage runaway chips will dominate this race.

I don't like the current situation.. Handpicking started with QX9650 and it is getting worse all the time. How to compete, if you don't have 100 chips to test I'm out for sure...

[dread77]

I can run my other chip -110c and it is not helping at all. Temperature is not the limiting factor in some cases. If you have really good cpu, then it might help something. I think next stepping will be improved. Until then, those guys with handpicked amperage runaway chips will dominate this race.

I don't like the current situation.. Handpicking started with QX9650 and it is getting worse all the time. How to compete, if you don't have 100 chips to test I'm out for sure...

I can't tell for sure, but I think that this issue is pretty much related with high temps. My retail 965 can take up to -125C, though when running a 32M SPi or WPrime 1024 at 4,85GHz I get bsods/freezes by the dozen. I took a validation at 4,9 and max post is 5,2. In 4,85 system can stay idle for hours.

I can't say for sure that it's not a bad setting (PLL, VCore, etc) because I only had the chance to test it for about 8 hours and then I run out of LN2. Don't forget that when saying -110C or -120C you mostly mean pot temperature and not Core temperature.

[saaya]

What are you talking mate

QPI is Quick path interconnect speed. It is marked as GT/s Gigatransfers per second. You see this value in CPU-Z front page "QPI link" At stock speed it is saying ~3200MHz, when it should say 6400GT/s(2x3200GT/s). When you raise BCLK this value will raise too and with lowest QPI multiplier (36x) it will go up to ~8000GT/s before it locks. Now if we would have lower multipliers than 36x we could get our BCLK higher before that ~8000GT/s limit.

GT/s is a marketing name and the actual qpi multipliers are 18x 21x and 24x.
18x133=2400Mhz=4800GT/s
21x133=2800Mhz=5600GT/s
24x133=3200Mhz=6400GT/s

gigiabyte calling the multipliers 36 42 and 48 is just confusing and its beyond me why they do this.

If you remember AMD 64 days and 1X to 5X HT multipliers, you will see what this issue is. Back then you had to lower HT multiplier to 4X or 3X to get your HT speed up to 200 to 250.

After all this processor is designed for servers and these really high QPI speeds were not what they had in mind. Maybe this will be fixed with new chipsets or 1160 processors. If QPI multipliers on X58/core i7 are limited to 36x on hardware level, there is nothing we can do. Just LN2 to CPU and Chipset and we might see some scaling

Are we on the same page now

i dont think so, intel is very aware that qpi multipliers are limiting the bclock, and that this limits how high you can overclock their cpus. they know it and they want it, otherwise why would anybody by a 965?

for every clock domain you have a set of registers to program the available multipliers, for the cpu, memory, uncore, and qpi... only qpi has a set of registers that defines a min qpi multiplier, so you can NOT use lower qpi multipliers than 18x. they didnt do this by accident or didnt think about how it limits overclocking...

and regarding slow mode, BloodRage has this as well, and weve been playing a lot with it, but it doesnt work... oh and the slow mode clock is actually not 100mhz, its 67MT/s which is 0.067GT/s wich is 0.033Ghz which is 33Mhz

thats the speed qpi runs at when the processor initializes, but as soon as thats done it forces a qpi multiplier of at least 18...

[saaya]

I don't think you can reprogramm the CPU with "ordinary" tools. From what Francois posted so far, only Intel would be able to unlock the CPUs.

im pretty sure even somewhat experienced power users could do it... if they knew the right buttons to push and the right bits to program...

[saaya]

There's surely a way, and with some good theory and tries we could make it.
But who has lots of i7s to try it out ?
Anybody willing to risk his CPU/s ?

ahum.... where do i start boss?

[SF3D]

Yeah Sascha, but I was right. We are QPI limited

Did tyou saw this comment? "That would mean higher BCLK clocks, but then the difference between basic and extreme edition cpu's would be thinner. Intel don't like that of course".

That's it.. New steppings must be improved or overclocking is not so interesting anymore cause everyone have same limit. (except 965 or new extreme edition cpu's)

[qurious63ss]

im pretty sure even somewhat experienced power users could do it... if they knew the right buttons to push and the right bits to program...

I'll bring my Flute multimeter and trusty logic probe. 1366 pins where do I start.

[T_M]

You wont get far using a Flute to measure voltage, or resistance for that matter.

[MidgetMariachi]

Hey Fugger, tell Francois we'll all pitch in $20 if he spills the beans.

LOL I'll throw down as well to get that bit of information!

Valdeam

[qurious63ss]

You wont get far using a Flute to measure voltage, or resistance for that matter.

LOL! It's a special kind of flute.

[BenchZowner]

I quote myslef

I got answer to my question about hardware level limits. QPI multipliers are locked in CPU's registry, so there is no way to unlock them. That would mean higher BCLK clocks, but then the difference between basic and extreme edition cpu's would be thinner. Intel don't like that of course.

This case is clear for me now

Bad news for 920/940 overclockers

GT/s is a marketing name and the actual qpi multipliers are 18x 21x and 24x.
18x133=2400Mhz=4800GT/s
21x133=2800Mhz=5600GT/s
24x133=3200Mhz=6400GT/s

gigiabyte calling the multipliers 36 42 and 48 is just confusing and its beyond me why they do this.

i dont think so, intel is very aware that qpi multipliers are limiting the bclock, and that this limits how high you can overclock their cpus. they know it and they want it, otherwise why would anybody by a 965?

for every clock domain you have a set of registers to program the available multipliers, for the cpu, memory, uncore, and qpi... only qpi has a set of registers that defines a min qpi multiplier, so you can NOT use lower qpi multipliers than 18x. they didnt do this by accident or didnt think about how it limits overclocking...

and regarding slow mode, BloodRage has this as well, and weve been playing a lot with it, but it doesnt work... oh and the slow mode clock is actually not 100mhz, its 67MT/s which is 0.067GT/s wich is 0.033Ghz which is 33Mhz

thats the speed qpi runs at when the processor initializes, but as soon as thats done it forces a qpi multiplier of at least 18...

It's funny how you too ( SF3D & you Sascha ) are disagreeing by agreeing
You're both saying the same thing, and reached the same conclusion

ahum.... where do i start boss?

Check your PM inbox