Critter 1.0 64-bit SSE4 getting the most out of my Cpu

[mhalstern]

I thought the members and especially Richard Vida, would find this interesting:

My Pc is built around an i7-920. I just fixed my cooling system and overclocked it to 4.1 GHZ. In order to prevent overheating the cores, I want to use the least amount of voltage that is necessary for the overclock. The default voltage for my motherboard to run my chip at the stock speed of 2.67 GHZ is 1.27 v. For the first overclock attempt, I started at 1.3 and incremented in the smallest steps possible, until I would reboot or bluescreen with 101 error, indicating the need for more vcore. Everytime I thought the setup was stable, Critter, whether in IA, or simply playing a short length engine match, would show the instability of the overclock, and cause the PC to reboot, with the 101 error code in the event viewer. Other top engines were not causing this. I current have my minimum voltage set at 1.325 v and I'm hoping this is 100% stable. Today, Critter has been in IA for the last 9 hours without any issues.

I keep Vdroop off, which means that the bios Vcore setting is the minimum. It will not drop below that level, but will go higher, depending on the load. At very little loads, CPU-Z shows my chip volted at 1.341. This is the interesting part: At full loads, engines that use all 4 cores (ht off) cause the voltage to go up to 1.352 v. For some thinks, especially in IA, some engines get the voltage up to 1.364 but not all of the time, and not continusly. In engine matches, it gets the voltage to 1.364 more often, and in long runs of IA, it gets the voltage to 1.364 continuously.

It looks like Critter is getting more out of my chip than other top engines.

[bob]

I thought the members and especially Richard Vida, would find this interesting:

My Pc is built around an i7-920. I just fixed my cooling system and overclocked it to 4.1 GHZ. In order to prevent overheating the cores, I want to use the least amount of voltage that is necessary for the overclock. The default voltage for my motherboard to run my chip at the stock speed of 2.67 GHZ is 1.27 v. For the first overclock attempt, I started at 1.3 and incremented in the smallest steps possible, until I would reboot or bluescreen with 101 error, indicating the need for more vcore. Everytime I thought the setup was stable, Critter, whether in IA, or simply playing a short length engine match, would show the instability of the overclock, and cause the PC to reboot, with the 101 error code in the event viewer. Other top engines were not causing this. I current have my minimum voltage set at 1.325 v and I'm hoping this is 100% stable. Today, Critter has been in IA for the last 9 hours without any issues.

I keep Vdroop off, which means that the bios Vcore setting is the minimum. It will not drop below that level, but will go higher, depending on the load. At very little loads, CPU-Z shows my chip volted at 1.341. This is the interesting part: At full loads, engines that use all 4 cores (ht off) cause the voltage to go up to 1.352 v. For some thinks, especially in IA, some engines get the voltage up to 1.364 but not all of the time, and not continusly. In engine matches, it gets the voltage to 1.364 more often, and in long runs of IA, it gets the voltage to 1.364 continuously.

It looks like Critter is getting more out of my chip than other top engines.

Doesn't mean that at all. Critter may well beat on a particular instruction (or instructions) that require just a touch longer to complete than others, and which will become instable as you ramp up the clock without ramping up voltage...

[mhalstern]

Thanks for the clarification. I was about to edit my post before I saw the reply. It would have been more appropriate to simply ask why Critter needs additional voltage, rather than claim it gets more out of my chip.

[bob]

Thanks for the clarification. I was about to edit my post before I saw the reply. It would have been more appropriate to simply ask why Critter needs additional voltage, rather than claim it gets more out of my chip.

This is a risk of overclocking. No matter what you run, it is unlikely you can test every critical path thru every CPU pipe to be sure that the clock edge does not fall prior to the circuit properly settling... I've only seen this kind of problem about a zillion times, which is why I don't overclock. A common statement "But it ran all the recommended tests for 24 hours with no reported errors..." That only means that it ran all of _those_ tests. Not everything your chess engine will try. And it is not easy to prove that a chess engine is running 100% correctly. I once ran on a Cray T932 where one of the 32 CPUs developed a fixed point math error. We tested for a couple of days before we began to notice that occasionally "everything was not quite right."

[mhalstern]

That is helpful.

I have something interesting to add:

A lot chess engine enthusiasts like myself overclock. It is a pain to get stable, but the results are noticeable. It is easy to point to nps or benchmark gainsfor evidence of overclocking helping engine performance, but playing results are more impressive. If I log on to playchess at 4.1 GHZ, and play enough games, my results are always better than if I log on with the stock 2.67 GHZ.

Here is my main point. It appears that the majority of overclockers are "Gamers." They recommended 2 popular tools to check the stability of an overclock:

1 - Prime95
2 - Linx


In working to obtain 2 instances of stable overclocks, I have had many instances where both utilities passed all tests and reported a stable overclock, but Chess Engines would crash the pc with the 101 (need more vcore) bsod error. Maybe my experiences are related to SSE, as Last year, Rybka, which utilizes SSE was crashing my early overclock attempts as well.

Your points are very informational. Perhaps at my last voltage setting that crashed Critter, all apps not utilizing SSE would be 100% stable at full load.