But on my Intel quad system, overclocked from 2.4 gig stock to 3.51 gig, most crashs occur when I use the mouse while Rybka is still running, especially after hours of use.
I would imagine that you may be just on the edge of stability, and that as Alan says, a mouse movement at the wrong moment just pushes it over the edge. I would try slightly increasing the voltage to the CPU, just a notch. Watch the effect it has on temps, just to be sure. That might well resolve the problem.
A CPU will last around 10 years in normal circumstances, even if you were to halve the life, most people would have upgraded long before then.
For me overclocking makes sense, the alternative is to go and buy a CPU, run it at stock speed so I am slower than everyone else, but at least know I will be getting slower and slower reliably for some years? Sorry but I don't see the point in that.
9 X multiplier (fixed), 405 FSB, showing 1077 Mem Frequency. (My OCZ RAM is rated at 1150 MHz). I would like to get Mem Frequency closer to 1150, but the next step up from 2.66D for the System Memory Multiplier is 3.00A. That gives Mem Frequency of 1215 (too high?)
My Standard memory timing controls: 5,5,5,12.
Advanced Timing controls: Auto, Auto, Auto, 35, Auto, 8 for Static tRead, 5, and 2T. The CPU/PCIEX Clock Driving Control at 900 mV.
The System Voltage controls: DDR2 OV =.40 (I can't change it--would prefer + .35), PCI-E OV +.20; FSB OV + .20; (G)MCH OV + .175.
The key CPU Voltage control is currently at 1.50625 where I can keep my water-cooled temps to a self-imposed limit of 65 degrees C, but unfortunately not 100% stable at 405 FSB: can use Rybka for hours but then it (though not Windows) shuts down.
On the other hand, if I increase CPU Voltage to 1.5125 (and certainly to 1.5187), I am stable at 405 FSB but eventually get temps up to 67 degrees C. Perhaps I can live with 67 degrees as a peak? If I stop Rybka from time to time, which I often do, then temp. drops in 3 minutes from 67 to 45.
The RAM voltage can significantly affect the performance of the memory, so I think you could try a couple of things. First I would raise the DRAM volts to +0.5v and try the multiplier at 3.00. There is a reasonable chance it will work, I have heard of that RAM running at 1300mHz although that was on a 680i chipset.
The other thing worth trying is to lower the timings a little. This will also result in significantly improved performance. Raising the volts to 2.3v as before, leave the multiplier at 2.66 but try manually setting the timings to 4-4-4-12. The remaining timings can be left on auto. Also don't forget you have two options under the High Speed DRAM DLL settings. If you have stability issues with one setting, try the other before giving up on it.
That review of the RAM you have might give you an idea of it's capabilities.
I would hope that you could achieve that FSB speed at 1.5v or a little less on the CPU, although as has been said before, every chip is slightly different. I am not sure why Rybka would shut down and not windows, a system crash due to instability usually takes windows out with it.
The only way to fine tune is patient experimentation, you have got pretty good results so far, and I hope you are pleased overall with your computer
That is a fantastically impressive review of the OCZ 1150 RAM that I bought on your recommendation. It looks like that RAM can run up to 1300 mem frequency safely, or certainly in the low-mid 1200s. I will try your two ideas.
However, I could not find "the High Speed Dram DLL settings." Are they also in the M.I.T. submenu?
Are you referring to "DRAM Timing Selectable"? The two choices are Manual and Auto.
/* Steinar */
/* Steinar */
What you need is:
2 x Xeon X5355 2x800$
Mainboard e.g. Tyan Tempest i5000XT 400$
4 x 1 GB FB-DIMM 4x 80$
PSU min. 550W 100$
any graphics adapter - from 40$ up
case for EATX e.g. Chieftech Mesh 100$
optical drive 40$
Are there any 2 socket boards that (or does this board) allow overclocking of the less expensive xeons?
i.e. for a quad I would buy a q6600 g0 and overclock to 3-3.2ghz rather than buy the more expensive q6700 etc.
I am looking at the possibility of having an easy overclock of Xeon chips if 800 Mhz FBDIMMS become commonplace. At the moment I think the Xeon chips run at a memory of 667 Mhz .... if a motherboard comes out that can support 800 Mhz then you have a quick and simple overclock by simply using a faster memory on a chip that is rated for lesser memory. In theory a 3 Ghz Xeon would then be able to run at 3.6 Ghz if the CPU can handle it ... and I am almost positive it can with proper cooling.
yes I know the q6600 will not go in these Xeon boards but the chip above is 771 socket the E5310 or similar are also 771 I just wonder if these can be overclocked in a similar way to the q6600 on the 775 socket.
i.e. are there any overclocker friendly Xeon boards - I am looking for bios options not hardware mods - my days of painting bridges on chips is over - I just don't have the time.
If I find anything I will post however I am not seriously looking at the moment as any more PC purchases and I might be divorced....
All the best
thank you for the reply - sad news
And thanks M ansari and phil for making this thread more interesting.
@ M Ansari:
4,6 Ghz o/c sounds nice to me :-) I will run my beforementioned setup with Vapochill and that x38 asus board on DDR3 - ths cpu:s will either be QX6850 or a brand new engineer sample of yourkfield quad with 12MB cashe. Any ideas if this yorkfield is o/c able in any interesting way?
I would say extremely promising....
Hi Alan. Always nice to see you here. Well that statement forces me to use the King Crusher account :-) I dont know how many kn/s a 5ghz yorkfield setup will generate, but it should be stronger than thoose intel 8 way 5345 that I've been running for a long time now. 500-650 kn/s on 8 way and thats almost like a quad thats o/c as i understand it. I tested the 232a on an opteron 16 way some months ago and it was higher but not that much, kn/s was between 700 and 900. If this new system would reach 800 kn/s as an average then i would be satisfied :-)
I will come back asap with benches and results and also many new problems to solve, i suppose..
Yes if we are lucky they will become much more easy to o/c. In 1-2 weeks i will also have a pair of harpertown to play with on a Supermicro board(so far the best o/c abilitys from all socket 771 boards that I've tested). Lets hope this will be the real killer setup for us chess comp freaks :-)
See this link:
Updates about my vapochill system:
To reach o/c like 5-6 Ghz on that new QX6950 or my ES sample we will need at least ability to deliver 250 to 300 watts vapocooling during full cpu load and overvoltage. Original vapos just delivers 180-200 watts wich is between +-0 and -25 C. So today i contacted a compamy that will mod my xe II from 180 w to around 300 i hope. We need at least -100 degreeds C to be stable @5,5 ghz and the vapochill systems can in no way deliver that.
Pros is that the new 45nm process will be much better to o/c our o/c guru "Fugger" (the new world record owner in 10 diff fields) said this after the new record:
"10x reduction in gate leakage = more speed with less volts
we will soon see 6ghz 45nm overclocked parts"
Thats soo niiice :-)
Yes the Vapochill LS only holds 200 watts the other version is rated 180 watts. I figure you need around 250 watts to 300 watts on a QX6850 to be stable at 4.6 Ghz to 4.7 Ghz. I can get my Vapochill (200watt) to speed up my 6850 to around 4.6 Ghz but it is not stable with chess. The CPU at that oclock simply generates too much heat for the rating of the LS compressor to handle. It runs very well if a high cpu cycle application is not thrown at it which means that the only problem to those high oclocks is controlling the heat. Obviously the new 45nm chips will be dramatically better because they generated much less heat ... times are looking good. But you know ... I think that after watching Zappa vs Rybka ... and seeing how much of an advantage it is to have efficient scaling ... I think that Vas will be on a heavy duty work effort in getting Rybka to be efficient at MP. This is maybe the incentive that is needed for that. So probably going for a high oclock on single sockets will no longer be necessary since with efficient scaling more cores are more important than more Ghz.
When i looked deeper in to it i realized that all Vapochill products is out of date, they are designed for single and dual cpu:s. They cant just deliver whats required for extreme o/c:s of intel quad systems. Yes, at least 250 w @-40C on a full cpu load of 4 cores is needed for 4,5ghz+ stable. The LS just delivers around 100watts on -40C and when temps are going up to -25 and higher we have to decrease the speed dramatically :-( Good is that i realized that before paying my XEII its now canceled :-)
I will order a private solution built in a Lian LI 2000 case wich will deliver at least 250watts @-40C. And the pricetag forces me to win the next freestyle :-)
Intel Nehalem will have 8 cores and 16 threads and will be able in 2008 for socket 775 - its just 1 cpu.. so lets see about if this setup will not be a good investment still.. :-)
I know that there are some guys out there that easily could mod your LS to more then double perfomance and that pricetag does not force u to win the next freestyle - but.. you could do it just for fun :-)
Been away for a while and my x38 things has been delayed. But thats good, meanwhile I've discovered that the p35 chip will give at least the same o/c results. Ok there are some theorethical advantages with X38 but thats more interesting for traditional benchmarkers running 3dmark, superpi etc. For us compchess freaks the p35 with native support for 1333 FSB will give the same enhancements for new QX series cpu:s. Also the X38 DDR3 requires much more power.
I've changed my order to a Gigabyte GA p35 Qd6 and its seems to be the oveclockers dream board as far as intel quads are concerned. The cpu will be a QX6950. I will come back with results if its interesting.
My computer is in this list, too.
But Fritzmark doesn't tell much about Rybka's performance on a computer.
So i think, your 4x4.32 GHz computer is a bit faster than my 8x3 GHz computer when using Rybka 2.3.2a.
how good is it for chess?
Apparently Deep Junior is using the same machine at Leiden.
I have no idea weather that is over kill for Junior or just what he needs to give him a competitive edge. It certainly isn't giving him a sterling showing.
in past I've been following the truly worthy hints put by Phil Harris and Co. regarding sticking together an individual quad system. :-)
But getting into the very details of overclocking I'm missing some background information.
Some experienced bud advised as follwing:
"Also you have to know that running a fast FSB provides much higher performance than a similar Ghz with lower FSB ... for example a CPU running 8x400 = 3.2 Ghz will perform much better than a 9.5x337 = 3.2Ghz [...]"
Please, what is the background? Is that regarded to the RAM frequency? What had I to pay attention to in this way? And what about the
"Static tRead Value" in this context?
In the GigaByte forum I got only one answer: set the cpu clock multiplier as high as possible?!
Thank you so much.
So it's certainly true that if the CPU speeds match, the 8x400 would offer a slight edge over 9.5x337. The only thing to bear in mind is that some motherboards are less stable than others at high FSB's so I would try to balance memory performance against CPU performance. So for example, if your motherboard allows up to 400 FSB at x8 (3.2gHz) but will go to 360 at x9.5 (3.42gHz), then the higher CPU speed would negate any memory latency advantage.
Adjusting Trd requires a little bit of trial and error, unless you really know what you are doing. I would suggest using Memset, which at least allows you to test RAM settings without rebooting every time. In Memset Trd is known as "performance level" and in general should be set as low as your system will allow. If you go too low your system will simply crash, so then you need to start again. As a rough guide on your type of setup, a setting of 6 or 7 is usually stable, 5 can be a little edgy...
> The only thing to bear in mind is that some motherboards are less stable than others at high FSB's...
It's the GA P35-DS3 v1.0 with 2x Mushkin 2 GB with 5-5-5-15 at 2.1V. CPU is running with 9 x 366 MHz = 3.33 GHz (running stable for many hours under ful load). What could you advise here?
> Adjusting Trd requires a little bit of trial and error,
I do know but wish to avoid many trials and errors.
> I would suggest using Memset, which at least allows you to test RAM settings without rebooting every time.
Does that mean I was able to set different RAM latencies under Windows? :-o
Thank you again! :-)
Assuming you have a Q6600 G0 on good air cooling, your result is pretty good. Remember to keep your CPU temperatures at a max of about 65c or a little less if possible. The P35 chipset is very good for over clocking, at very high CPU volts (over 1.525v), the DS3 might not have quite the stability in power delivery that you get from the DQ6 version, though I am pretty sure it will be fine up to that point.
You don't mention the specification of your RAM, or what speed it's running at. Assuming it's running well and within it's specified speed, it's worth trying to lower the latencies, aim for 4-4-4-12. Do this in memset first, if it proves stable, you can transfer the settings to the BIOS. If it doesn't work first time, try increasing the DDR volts to 2.2v. Not all RAM responds well to lower latencies, but it's certainly worth a shot.
When OCing with the newer motherboards there is a big difference to how fast you can run the FSB between quad and dual cores. As a guide dual cores can be run at 500Mhz plus (some 600+) where as quad cores usually run in the 400-500mhz FSB range although there are always exceptions to the rules. From my own experience getting passed 475mhz on the FSB has proved tricky for my quads ie: On this simple air cooled machine I run 10x420 (will do 10x440 but not chess stable) where as our benchmarking tweaked machine with all its cooling runs 11x470 ... If you do not set the memory yourself in the BIOS or by memset on the fly then as you ramp up the FSB the BIOS automatically tends to slow the memory down by raising the latency :(
I get better memory OCing results from the P35C chipset compared to the much newer X48 (more tweaking needed I guess) as it looks like the memory internal settings have been lowered and Rybka loves fast memory (Rybka 3 will prob be the same)
I think that Vas said that one of the differences between Rybka 3 and Rybka 2.3.2a is that fast memory will not make a difference anymore.
>I think that Vas said that one of the differences between Rybka 3 and Rybka 2.3.2a is that fast memory will not make a difference anymore.
I must have missed that and wonder what has changed? .... Anyway the faster latency will give you more bandwidth which can make the system fell snappier overall but faster benchmarks does not always mean stronger chess ie: An 8 core Skulltrail with Rybka is no better in the real world than the quad at 10x420 even though the Fritz benchmark may well be a lot higher, Zappa Mexico 2 on the other hand scales really well on the Skulltrail in comparison, that said in tests skulltrail it is well behind Nehalem :) which is where Intel have been meaning to get to be able to match AMD in the memory controller department etc
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill