Have Prime 95 round off errors with my POS duron at High fsb

[el]

I am running my duron750 at 6.5x138 cuz any higher fsb and I start having round off errors in prime95. Now are these errors in the chip or within the transfer of data to and fro? I am am running my ddr at 2.6volts should I up it with a chance for better stability or can the duron not handle the higher fsb. I did try different multipliers and I start getting the round off errors at around 137-138! I think I will sell teh duron and upgrade to the new Palmino! ;)

[WillysNut]

I believe it has to do w/ the associated heat that results from a voltage/fsb increase. If the alum/copper interconnects in the cpu get too hot..signals essentially get "lost" in transit. The increased temperature of the tiny wire reduces its ability to properly conduct or transfer the signals.

Can anyone verify this? ...gonna do some searchin'

[.. AE 86 ..]

try upping the ram voltage

[Luis]

I´m having errors too at as low as 137MHz, but my system is watercooled (about 36-38C now) so I don´t think it is because of heat. I´m at 2.6Vmem too, timings 16,8,6,2,5,2,2, and had to reduce fsb to 136. Maybe raising Vmem helps...

[el]

my temp is 37 under load. and some people have temps around 50 and don't get any errors. i think it is the chip but I wonder if bender and dimmreaper with the 600's at 1100+ are having round off problems and how did they get rid of them.

[Fiz]

It is probably the higher fsb, try raising the VIO to something around 3.4 or 3.5, that may help out some, also try using more conservative settings on the ram and see if that helps at all.

Fiz

[Luis]

Today I raised the Vcore to 1.95V and could make 140MHz fsb with no errors. Vmem is still at 2.6V (same timings as before) and cpu temps climbed to 39-40C...

[el]

my ddr is running at 140 with 16 8 8 2 8 2.5 2 settings adn voltage at 2.7. prime95 isn't showing errors so I guess my problem was with the ram not the chips ALU unit not being able to handle the speed.

Also I have ram sinks on teh ddr to help keep it cool.

[el]

up to 6.5x142 stable and still need to test 143 adn 144. 145 passed the first five and then had a round off error. darn it!!! will try 2.8 and 143 up and probably leave it at that.

[Alex99]

From Prime95 readme:

"Round off checking. This option will slow the program down by about 15%. This option displays the smallest and largest "convolution error". The convolution error must be less than 0.49 or the results will be incorrect. There really is no good reason to turn this option on."

Just turn off the Round off checking.

[Crash893]

if your worried about errors then go to the prime.log and look for SUMOUT Errors ( it logs the time)

[el]

you guys seem to know your prime95 stuff so will the rest of my apps have similiar problems or just prime95? I want a fast machine but I don't want to sacrifice quality for quantity.

[Luis]

I agree with you, El. I do like big numbers too but I don´t like false results in seti, etc., so yesterday after the 3rd. system crash I decided to go back to 135 @ 1.85V (temps climbing above 40C). My chip is still locked at 7x, should I try lower multipliers? I also couldn´t make any difference with rhe ram settings (still don´t like the idea of giving more than 2.7V...) ;)

[el]

Another update! at 144x6.5 for 936mhz with the same ram settings.

3dMark2001 runs fine and so does Prime95 but I still would like to go a little higher. I think the next thing I will do is test the fsb at multiplier of 7 and see if 144x7 is possible! hell even 143 would get me to 1001 :-)


We will see and man this EPOX board rocks! so many tweaks! I need a better chip cuz I feel that all my effort is kinda wasted but I must wait for the 1.53 Palminos or XP or Athlon 4 or A4 or Model 1900. Man AMD needs to cool it with the different names me thinks.

Luis you need to unlock that 700 I think you would go higher and faster at 6.5x??? rather than 7x138! maybe your chip could handle it better at the lower multiplier.

[Luis]

Thanks for the advice! right now I´m in the task of finding AS II here in Buenos Aires, so as soon as I get it I will try. And again I agree with you, This mobo is MASA!!!:D

[Crash893]

there is a new version of prime that may give you less errors becuase its made for p4 and athlon

[el]

cool thanks crash! do you have a link to the newest version.

[Crash893]

sorry about the delay.....

"Version 21 offers about 20% speed improvements for Celeron 2, Pentium 3, and all Athlon processors. Pentium 4 processors get about a 3x improvement. Of lesser importance, there are other new features."


windows=ftp://mersenne.org/gimps/p95v21.zip
windows 2k,nt,xp service=ftp://mersenne.org/gimps/winnt21.zip
linux ( dynamic link)=ftp://mersenne.org/gimps/mprime21b.tar.gz
linux (static link)=ftp://mersenne.org/gimps/sprime21b.tar.gz


if you have a diffent os than up there then let me know ill find you a version that works

hope that helps out

here is the bench mark page for the work units
http://mersenne.org/bench.htm see how you do compared to every one else.