Old School is Best School Season 3 Round 2 Discussion

[I.nfraR.ed]

Nice scores everyone!

I wanted to participate, but had no time and also have more important things to sort out.

 

Hopefully I would be able to participate in some of the future competitions.

 

I wonder why noone ran 2x7900GTX. Was it limited by the platform, was it buggy, was it not scaling or smth else?

I have 3 of these, but didn't even had the time to mod 'em.

[phobosq]

Mine didn't run at all and they appear gone - both of them are heavily artifacting.

Someone planned to do 2x7950GX2, but I also miss it.

[ozzie]

congrats to all in the comp, well done guys, especially to stelaras, top runs man, and thx to the bot boys as well for running it, W9's Mr Paco for instigating the set up for the comp to start with

i tried 2 x 7900 gtx gpus in the venus but wouldnt see 1 card for some reason, tried everything, put the card in another board and no probs and away it went, but didnt like my board tho so ran with 1 card

[I.nfraR.ed]

Someone tried HyperSLI patch then? We were able to run SLI on a 790FX chipset, which does not support it officially.

Or some older driver?

 

I have SLI-DR and Expert boards. Also one Ultra-D with chipset modded to SLI-D, but haven't actually tried. Might do in the evening, since I had the board setup, just no time to run it.

[TASOS]

Just wanted to ask here ...

 

Did this stage ended for all at the same time ???

 

For me it ended early in the morning at 2 o'clock

 

But i see that other were submitting just a couple of hours ago ...

Was there an extension ???

 

I would have finished my Realbench

[I.nfraR.ed]

I noticed too there was a strange clock on the general page, like 12 hours more left compared to the counters on the stages.

Basically it seems we had half a day extra, but not sure if just for this stage.

[Dead Things]

Congrats everyone on another fun OSIBS round! Especially Stelaras - what a dominant performance! Good on you, sir! This round was haaaarrrd - Ice Storm and Realbench were beasts.

 

In response to InfraRed... for my part, tried to SLI Ice Storm with 2 x 7900GTX, 2 x 7950GT and 1 x 7950GX2 on an Ultra-D (with SLI mod), SLI-DR and A8N32-SLI. In every case, no matter the drivers used, SLI scores were lower than single card scores. The scaling of the benchmark as a whole remains a mystery to me. My 7900GTX scores just would not increase at all with any GPU core or memory increases. Only my 7950GT scores would go up, and even then, not by much. It seems to be more of a memory benchmark to be honest. And in that department, my TCCDs can't do the 2-2-2's like BH-5 or maybe stronger TCCD kits can do, so yeah...

 

Killed my Ultra-D in the process, but was able to bake my previously-dead SLI-DR back to life. So I guess I came away even-Steven. Think the Ultra-D needs to be recapped, so hopefully it's only mostly dead.

[TASOS]

I noticed too there was a strange clock on the general page, like 12 hours more left compared to the counters on the stages.

Basically it seems we had half a day extra, but not sure if just for this stage.

 

Thanks for the confirmation.

 

At least i know i'm not crazy ... seeing things on my own.

 

What more can i say ?

[websmile]

I did the moderation with handbrake on especially for stage 3, result is final from my side but as long as realbench gets no fix it will not be used again for any competition as long as I can stop it. No offline saving, only cpuz shown as mandatory by program when you save plus several bugs within the program need a fix or it will not have a future here, sorry that I have to say it this clearly..

On the clock issue, I already got a heads up yesterday and checked at 1 in the night, timer resetted to another 12 hours more, it was shown correctly at the lower side of the pages where remaining time also was sown. Still this is something that should not happen because it irritates people and can influence outcome of the competition.

Grats to the winners, it was a great race nontheless

[TASOS]

In total , it was an interesting round

 

Cinebench was a bit brutal , especially for the single-core cpu's

But it was do able.

 

Ice Storm , was something new (for the hardware used) and i would dare to say , that it was the most interesting stage (for me).

Almost every aspect of tuning , was giving it a tad more in final score.

 

RealBench on the other hand , almost killed all the fun of this round.

It was tooo lengthy for the capabilities of the hardware.

It was not easy to overcome the disappointment of the last minute failed bench session.

 

The non-saving score was also a big minus.

 

... and its final score , was not the result of hardware tuning.

It's software tweaking (the most) ... and i hate these kind of benchmarks.

 

//

 

I may have complaints about this weird competition ending time ... but all in all it was fun.

 

I didnt have the clocks to compete in higher level.

and i dont take risks (anymore) with sub-zero cooling , on old hardware , cause it's not easily replacable nowadays.

 

At least , i think i got the highest GT2 score of all , to remember.

Especially for my clocks and cooling.

 

... and i found , that i have a small X2 that can do 312 1/1 on mem (air-cooled).

 

Congrats to all !!!

There were some very good scores on top places !!!

 

See you next round.

[skulstation]

Congrats to all.

I did try to run 7950 gx2 skulstation`s 3DMark - Ice Storm score: 15777 marks with a GeForce 7950 GX2 and 2x 7950 gx2 skulstation`s 3DMark - Ice Storm score: 14439 marks with a GeForce 7950 GX2

but the 2x 7950 gx2 was slower than a singel 7950 gx2

Edited February 15, 2017 by skulstation

[TAGG]

I agree 100% on the realbench part, I think that benchmark should never be used again for hardware older than 1st i7 generation...

I spent at least 2 days worth figuring out the tweaks, and getting the file to save, and get the benchmark not to say files corrupted... Beyond that it still has the Pcmark feel to it, not as bad but i'd put it somewhere inbetween XTU and PCM05 in therms of beeing a mess that spits out scores based more on software than hardware (No offense to people who like that type of benchmarks, but I'm surely not one of them)

 

Cinebench was also a pain in the ass for singlecore, I had the benchmark run several times and couldn't screenshot it, because the process that was essential for interrupts (ie keyboard and mouse) crashed due to realtime priority, I then set the system up in a way, that it would shut down after 30 min, so that I had mouse and keyboard when fired up again

 

Ice Storm sure was fun, but i think it was too much of a memory benchmark, so in essence there was not a single "real" 3D bench in this round, which I think is quite sad...

 

Oh and can someone please tell me if there is any trick to fix hte coldbug on 939 CPUs, because my results got worse on every single chip below ~-10 to -20°C

Edited February 15, 2017 by TAGG

[lanbonden]

Oh and can someone please tell me if there is any trick to fix hte coldbug on 939 CPUs, because my results got worse on every single chip below ~-10 to -20°C

 

Well, other then high fsb = worse coldbug there aint much to do about it as far as I know. Thats why I had to run with 15x multi on my FX-60 even if it hurt memory performance alot.

 

Thinking about it after the runs I should have gone with higher fsb and lower clock for 3dmark ice storm. But I got kind of stressed from knowing that I was runing out of dice after to much time in realbench so I only had time for a few runs 3dmark.

[TAGG]

This unfortunalely only works on FX, i'd still like to know what Stelaras has done to get his Opteron 180 to 300 FSB... maybe the CPUs that are worse on air are better on cold... i'll definetly rebinn the 5 Opterons I got on SS

[I.nfraR.ed]

Sometimes bad air chips fly on LN2 with really unexpectedly low CB. You should test all your CPUs on clod, no matter if they are bad on air.

Well, this applies to less extent for lower multi chips like 3000+ and 3200+. I haven't found a single one that has an extraordinary low CB. And they need high FSB, so if they are really bad on air, then won't do much on LN2. By much I mean to break the records for these CPUs.

Generally a CB around SS temps (-45/-50) is considered very good.

 

One of these bad-air-good-on-LN2 is my 3800+ Venice. It could not even do 3GHz on air, but has -98C coldbug which is insane for these: AMD Athlon 64 3800+ (Venice) @ HWBOT

 

So if Stelios' Opteron has a nice coldbug around even -40, then it will fly. For example my 180 has a CB around -8 and does ~3.3GHz for CB11.5.

 

There are some other tricks for single core CPUs, but they don't help much. You can't lift that CB. You can only try to somehow increase temperature even on idle, because that's where it bugs mostly. When loaded it can usually do higher clock, due to the higher temperature.

Which you compensate with voltage.

 

PS: Unlocked chips (a.k.a FX) is best to be clocked with multi and low FSB (~225 max) and 250 divider for mem. This way you can get much lower on temperature, which translates to higher core clocks, but you all know that.

Edited February 15, 2017 by I.nfraR.ed

[TAGG]

Seems like it was a mistake to think air binning is the way to go with these chips...

3000+ and 3200+ I didn't even look at them, assuming they're crap... the only chips i took into consideration as far as singlecores go were 3500+ and opteron 148 and up

[I.nfraR.ed]

Ah, for this competition CPUs with 512KB L2 cache were not that good, although I bet my 3.7GHz Venice would do fine.

But 3000 and 3200 were a no go for sure, I was talking in general.

 

For this competition you were looking for nice scaling San Diego 3700/4000, FX-55/57 San Diego for Cinebench, for others a X2 was needed - basically higher bin Opterons and FX-60. Perhaps some Toledo, Manchester and Clawhammer would do fine as well, but they are slower clock for clock. FX Clawhammer and even 4000+ have lower CB and being 130nm can take higher voltage, but are a little slower than San Diego.

Some other CPUs would do good also, but these were the best choices.

[TASOS]

...

Oh and can someone please tell me if there is any trick to fix the coldbug on 939 CPUs ...

 

The trick to improve CB , they already told you here.

 

Now , the real thing to ask these guys here , is to tell you some exact FPO/BATCH , that fly in cold.

 

[sburnolo]

Congratulations to all, beautiful round, i saw great scores.I was lucky in the second stage thanks to the excellent driver suggested by Mat_Agnesi but unlucky in general because my opteron has bad cold bug around zero degrees.

[TAGG]

I don't buy single chips bay batch, I guess i'll just bin every s939 on SS to see where CB is, instead of going for max frequency on air...

Batches help only if buying single chips, not if you buy your chips per kilo, like I do, i'll just hope i find a decent one for the next time 939 comes up in an oldschool competition

[TASOS]

... buy your chips per kilo, like I do ...

 

That's a good tactic

 

I would have followed your method (if i had it as an option in my local market).

 

Can you send me a couple of kilo's (fx55 and fx57) ?

[TAGG]

Unfortunaly i have never found any etreme chips like FX or Extreme Edition P4 or QX... Most of the chips you'll get this way are either mainstream or server chips, that's why I only had Opterons for this round... and about 30 3200+ which are most likely crap

[Stelaras]

That's right . No magic trick except binning for cold . My first test is under SS (-50C) .

For example , the Opteron 180 I used for this round is able to sit idle at -55C and 350-355 Mhz FSB , while at -75C can reach up to -330-335 Mhz FSB .

I'm pretty sure it can bench every benchmark at 3600+ Mhz , but I didn't have time to find all the correct values ( Even drive strength values help with coldbug and stability at high clocks & low temps )

 

On the other hand , I have a question . I am 99.9% sure that the specific FX57 I benched is able to run 3dmarks at 3800+ Mhz and 32bit OS . It's the exact same one I used during the teamcup 2014 3dmark01 round . For some reason I couldn't bench cinebench even at 3700 Mhz . Anyone else noticed this difference ? I didn't have time to test my other 3 FX's and I'm still wondering what went wrong .

[Alpi]

My 154 looks awesome. Did Cinebench run easily at 3560 @ 1.5V with a eeak ss. Load temp was + 3-4 degree. I'm going to test it colder soon. My 165 is the opposite type. Had to keep the water temp over +10 unless it stopped.

[TASOS]

... My 165 is the opposite type. Had to keep the water temp over +10 unless it stopped.

 

Hah ... and i thought i had the worst cpu ever , with a CB at 0C