Ivy Bridge vs Sandy Bridge for computer chess

[syzygy]

Short bursts are generally "not interesting". I've tested on most, and they seem to run at their nominal clock speed after a few seconds of full-bore chess. Except for the Dell case where they screwed up the BIOS. I won't claim to have run on all processors however, but I have run on a bunch, at least through the most recent 6 cores which happen to be in our new cluster...

I presume you could tweak most anything if you use exotic cooling and such. I'm obviously always talking about "stock boxes" as I don't have any interests in overclocking or over-stressing a CPU beyond what Intel (or AMD) recommends.

Well, just see what Larry wrote. Apparently his processors benefit from turboboost for extended periods of time. The clock rates he reports are correct.

[diep]

Short bursts are generally "not interesting". I've tested on most, and they seem to run at their nominal clock speed after a few seconds of full-bore chess. Except for the Dell case where they screwed up the BIOS. I won't claim to have run on all processors however, but I have run on a bunch, at least through the most recent 6 cores which happen to be in our new cluster...

I presume you could tweak most anything if you use exotic cooling and such. I'm obviously always talking about "stock boxes" as I don't have any interests in overclocking or over-stressing a CPU beyond what Intel (or AMD) recommends.

Well, just see what Larry wrote. Apparently his processors benefit from turboboost for extended periods of time. The clock rates he reports are correct.

You assume that someone who has problems distinguishing between what a proces is and what an instance is and what a match is, that he's capable of reading out the frequency settings of a cpu based upon

"we from McDonalds advice hamburgers from McDonalds" ?

I'd take Bob's opinion here any time of the day instead above anyone who is diong hearsay.

If i look at diep's nps-es i get back from testers who tested at similar machines, all of them turboboost turned on, the nps simply isn't reflecting it manages to boost itself.

Only if you enforce it in bios an overclock of the cpu it will 'boost' to that frequency in general and then you need very good cooling.

[bob]

Short bursts are generally "not interesting". I've tested on most, and they seem to run at their nominal clock speed after a few seconds of full-bore chess. Except for the Dell case where they screwed up the BIOS. I won't claim to have run on all processors however, but I have run on a bunch, at least through the most recent 6 cores which happen to be in our new cluster...

I presume you could tweak most anything if you use exotic cooling and such. I'm obviously always talking about "stock boxes" as I don't have any interests in overclocking or over-stressing a CPU beyond what Intel (or AMD) recommends.

Well, just see what Larry wrote. Apparently his processors benefit from turboboost for extended periods of time. The clock rates he reports are correct.

If it runs at a overclocked speed for extended periods, makes no sense to not just set that as the basic clock frequency, since there is no need to lower the clock if the CPU is not overheating at the faster rate...

None of mine behave like that, however, but we don't (as I said) have a machine with every Intel chip that was made.

I have it disabled on every box I use to keep times stable...

[diep]

Short bursts are generally "not interesting". I've tested on most, and they seem to run at their nominal clock speed after a few seconds of full-bore chess. Except for the Dell case where they screwed up the BIOS. I won't claim to have run on all processors however, but I have run on a bunch, at least through the most recent 6 cores which happen to be in our new cluster...

I presume you could tweak most anything if you use exotic cooling and such. I'm obviously always talking about "stock boxes" as I don't have any interests in overclocking or over-stressing a CPU beyond what Intel (or AMD) recommends.

Well, just see what Larry wrote. Apparently his processors benefit from turboboost for extended periods of time. The clock rates he reports are correct.

If it runs at a overclocked speed for extended periods, makes no sense to not just set that as the basic clock frequency, since there is no need to lower the clock if the CPU is not overheating at the faster rate...

None of mine behave like that, however, but we don't (as I said) have a machine with every Intel chip that was made.

I have it disabled on every box I use to keep times stable...

I didn't figure out all the legal disclaimers, but AFAIK intel isn't garantueeing bugfree operation of cpu's when they turboboost.

Most HPC centers have turboboost turned off simply.

It's of course a terrible waste of system time to play matches at such a machine of 15 seconds all with 0.15 increment.

At a 2 socket machine the switch latency is determined by the speed at which the runqueue fires which is a lot slower than 10 milliseconds. Count at it losing over 30 milliseconds. there are always heated debates on how much you actually lose and whether windows loses you more than linux there. That's not an interesting discussion here.

Realize all this works different on a 2 socket machine than a single socket machine...

So you lose a LOT of the system time to the OS trying to handle 64 processes at 32 cores.

then it takes some milliseconds for the caches to adjust as the statistical odds that an idle proces gets scheduled at the same logical core is near zero.

In many cases it will occur that 1 proces gets for the entire period of time of a move, the full system time of a PHYSICAL core and the next move a proces has to share 1 logical core with another proces, so it gets less than half the system time. In short it it like running it on a 100Mhz cpu then for the period of time for just this move.

Realize that linux is a very primitive monolithic operating system, just like windows is.

The only 2 good things we can say about it, is that it is for free and that Linus is doing big effort to get drivers for every possible hardware toy.

Yet it's a total outdated kernel for modern HPC machines that have 16 physical cores and 32 logical cores with NUMA properties.

Most HPC workloads run for months nonstop at many cores. The operating just needs to not mess up then. That's a total different scenario than here where the operating system, basically something that works total single centralized core, has to steer nonstop 64 processes at 32 logical cores.

An absolute nightmare.

Small revisions in kernel and libraries might make huge differences then. Not because it runs 'so much better' then, but because others will just total mess up.

[syzygy]

You assume that someone who has problems distinguishing between what a proces is and what an instance is and what a match is, that he's capable of reading out the frequency settings of a cpu based upon

It's not very difficult to read what's on one's screen. (And btw, I did not notice any of those problems you mention. I think it's at most you that isn't reading.)

I'd take Bob's opinion here any time of the day instead above anyone who is diong hearsay.

I won't try to stop you, but several of Bob's claims are verifiably incorrect.

If i look at diep's nps-es i get back from testers who tested at similar machines, all of them turboboost turned on, the nps simply isn't reflecting it manages to boost itself.

Look, there are very easy to use programs that report the real frequency that the cores are running at. Larry obviously has one such program installed. For his E5-2670 It is reporting 3.0Ghz when all 8 cores are in use and 3.3Ghz when at most 2 cores are in use, which is exactly what it should be if turboboost is on.

I don't find it useful to spend more time on this with people that haven't done their homework.

[syzygy]

If it runs at a overclocked speed for extended periods, makes no sense to not just set that as the basic clock frequency, since there is no need to lower the clock if the CPU is not overheating at the faster rate...

Not all 100% cpu loads produce an equal amount of heat. For example running prime95/mprime produces far more heat than a chess engine using all cores. I can easily believe that Larry's system running Komodo on all 32 hyperthreads manages to do that with turboboost on. Running instead 32 instances of mprime might force the processors to run at their base speed of 2.6Ghz.

I have it disabled on every box I use to keep times stable...

I have it enabled, but set to the same frequency for any number of active cores.

[diep]

You assume that someone who has problems distinguishing between what a proces is and what an instance is and what a match is, that he's capable of reading out the frequency settings of a cpu based upon

It's not very difficult to read what's on one's screen. (And btw, I did not notice any of those problems you mention. I think it's at most you that isn't reading.)

I'd take Bob's opinion here any time of the day instead above anyone who is diong hearsay.

I won't try to stop you, but several of Bob's claims are verifiably incorrect.

If i look at diep's nps-es i get back from testers who tested at similar machines, all of them turboboost turned on, the nps simply isn't reflecting it manages to boost itself.

Look, there are very easy to use programs that report the real frequency that the cores are running at. Larry obviously has one such program installed. For his E5-2670 It is reporting 3.0Ghz when all 8 cores are in use and 3.3Ghz when at most 2 cores are in use, which is exactly what it should be if turboboost is on.

I don't find it useful to spend more time on this with people that haven't done their homework.

I believe testdata i get back rather than someone who is handwaving.

If the chip would be systematically turboboosting under full load, intel would of course not sell it as the budget cpu's he has, yet as a much higher clocked and more expensive CPU. In fact if you work out that simple commercial insight a tad more you'll figure out that under no circumstance under full nonstop load, the cpu's will both be able to turboboost at all.

Only their testmachines do.

As a reminder to you on how 'happy' intel allows reading modern cpu data under linux, mind explaining to me how in linux you can see the cpu-temperature and cpu frequency of motherboards that run MODERN intel Xeon processors?

Because you will be in for a nasty surprise...

[lkaufman]

It's of course a terrible waste of system time to play matches at such a machine of 15 seconds all with 0.15 increment.

If you want to test two ideas per 24-hour day, and you expect the ideas to be worth no more than two elo, you need more than 20k games to have a fairly high probability of making the correct decision. Apparently this sort of fast testing works for us, as every Komodo version has been clearly stronger than every previous one on nearly every rating list, including the 40/20 and 40/40 lists, and we now clearly outrate all engines except Houdini on single core. I would like to ask you how you would choose to test given our hardware and the desire to test two changes per day?

[Joost Buijs]

I have it disabled on every box I use to keep times stable...

I have it enabled, but set to the same frequency for any number of active cores.

You are completely right. I have two 6 core machines overhere, one Nehalem and one Sandy-Bridge, I have both set them to the same boost-fequency for any number of active cores. It is very easy to demonstrate that the cores really run at their max. boost-frequency for an indefinite period of time when under load and properly cooled.

Changing the boost-frequency of locked server CPU's is not possible I suppose.

[diep]

I have it disabled on every box I use to keep times stable...

I have it enabled, but set to the same frequency for any number of active cores.

You are completely right. I have two 6 core machines overhere, one Nehalem and one Sandy-Bridge, I have both set them to the same boost-fequency for any number of active cores. It is very easy to demonstrate that the cores really run at their max. boost-frequency for an indefinite period of time when under load and properly cooled.

Changing the boost-frequency of locked server CPU's is not possible I suppose.

There is indeed a big difference between proper cooling and enforcing single socket cpu's at a specific frequency and server CPU's which also have a cache coherency to solve between them, amongst others.

What sort of cooling do you use Joost?

Over here i'm aircooling under which 100 CFM fans for the clusternodes, and heatsinks with heatpipes, yet no liquid/water cooling.

What frequency do you run them on at tournaments?

It seems though ivy bridge cpu's are less consistent in possibility to overclock than sandy bridge, unless you manage to get a good chip and modify the cpu's heatspreader. That would make overclocking ivy bridge a lot more complicated. Any experience there?