I.nfraR.ed

You're not doing anything good, but taking from my points! Nice result, but I want to see a real efficiency (at least in 32M) from you Wish you good luck with the qualifications

I'm able to run 1080+ 7-7-6-20 1T with 1.7125V on the GTX2's, but took me quite a while to figure out what was the problem before. I could use it for everything, but 32M, it was freezing either on the initial or the first loop. What helped me was changing tWWDD to 4. Put all subtiming one or two notches up from their auto values, then started to tighten them one at a time. Oh, lowered CHA and CHB Data ref voltages to 0.495. 1100 7-7-7-21 1T might be possible too. Board is M4G-Z with 0603 bios and 2500K on water. I wanted to make a comparioson of GTX2's at 1081 7-7-6-20 1T and BBSE at 1120 7-9-6-20 1T/1140 7-9-7-21 1T, but have no time for tweaking correct subtimings. I see some guys are running hypers at 2T with 3D, but not sure how big is the impact in 32M. Maybe I can try same settings with the only difference being CR 1T/2T.

Thanks guys. Just found some spare sinks that fit, so I'm going to play it safe. It would be my second LN2-cooled vga and I've never played with 8800 before. I don't expect great results, but still it's a good experience.

Would it be safe if I run 8800 on LN2 without heatsinks on the vrms, but just a fan blowing over them?

Not sure. I remember HD4870 Matrix, but don't remember about HD4890. However it looks like enhanced HD4870 Matrix. Don't remember seeing it retail in Bulgaria.

Asus non-reference. I remember there was a news back then that they will produce HD4890 with a non-reference PCB and better PWM. Can't find the picture right now. Edit: Oh, here it is http://www.ampletech.net/recensioni/2577/thermalright-t-rad2-gtx-nuova-aria-per-le-schede-video/?pagina=3

Beautiful! Asus non-reference, right?

I can sell you a 5.2GHz Sandy, ok? Nice competition and nice time for me to get some 8800 scores.

Damn you Now I should run it again and try to optimize things. Nice score

A fun competition, although I lost one NF7-S. Still have one working (out of 4), so all is good. Nice scores in all stages. Congrats to the winners.

You probably missed the Vtt/Vref mod and the result was that Vtt didn't follow VDIMM/2 ratio. At 3.2V on the ram it had been stuck @ 1.45V.

Congrats! Very nice I went up to 2.35V on water, but FSB didn't go up. Had no time to get LN2 this week.

To me MVP is more like ramdisk and disk cache programs. It's definitely good for daily usage and speeds things up (at least in theory), but doesn't represent the actual speed of the hardware.

Nice! I've found one of the other NF7 working, so I will try again

Loose the leading "M"

Sorry to hear that What was the reason?

Well, cas 1.5 should not be real according to the specs, but Oskar Wu did something (maybe related to internal latencies or idk) and infact it's a little bit faster than real CAS2. There's almost no difference for stability, but CAS2 goes a little higher if we speak about max benchable frequencies. Anyway, nice cpu there. I'm glad to see people are still benching these cpus...

Yep. I don't know why they all want removing the scores and continue digging into that direction. Plus, current rules does not force using the hwbot app for the screenshot, so these results fully comply them.

I kinda agree with rbuass, although he's skewing things towards his opinion, without providing enough arguments. Probably it's the language barrier. So, there's perfect sense in forcing the user to have a valid screenshot included in the benchmark. This way, we eliminate the possibility for submitting a photoshopped result. I think it's good to change the rules. Ofcourse, I don't doubt the validity of the forementioned scores. On the other hand, if the mechanism validates the file, it doesn't matter if it's bugged or not (could that happen at all?). I don't know what the mechanism is, but I giess it doesn't validate against the clocks used and the relative performance expected from those clocks. Opening required cpuz and gpuz tabs after the validation file is saved doesn't change anything. It would eliminate only a photoshop patch work. However you still need a valid file to submit.

If the score/screen is real (not edited), then for me it doesn't matter. As long as the score is the same. What's the difference between both cases? 1. Finish benchmark, open cpuz tabs, gpuz, make a screenshot from the hwbot app and save the score (it will contain the screenshot) 2. Finish benchmark, save the score from hwbot app, open cpuz, gpuz and make a screenshot manually You can manipulate frequencies in both cases, so I don't think that's a valid point. And you can't save the file, then open it within the program (like with 3DMark -> submit offline result). This means the system didn't freeze after the benchmark. The only difference I see is that with the first case you don't need to make a correct screenshot manually, it's already in the hwbot validation file.

Thanks, Massman! Kinda not the idea I had, because initially I wanted some of the air/water guys get that board (or ideally one that competed with a low-end board, regardless of cooling). All top-5 guys, except me (970A-UD3), had used top-of-the-line UD7 board. So, basically they don't need UD5 if we assume UD7 is better. For example saint19, Strunkenbold, fgi, MrPaco, kirbster have competed with older boards and probably don't have the funds to buy FX990 board or can't justify spending so much money for it. They could have been happier, imo. Anyway, congrats to froxic. You're pulling really nices scores lately. Keep going in that direction!

60+ on the cpu

Why not upload it then? It's not about prizes.

tWWDD = 4, otherwise it just freezes on 32M.

It could do 281. I have pencil mod on this one, but had somewhere another NF7-S v.2 with a hard-mod. It doesn't scale much with volts. Up to 2V is ok, but scaling after 1.8 is small and over 2V makes it even worse for my board. Yours might be different. Here's one good guide http://www.techarp.com/showarticle.aspx?artno=106&pgno=0 . Some of the boards need Vref mod as well. It depends on the board rev (numbers on the back of the pcb), but I don't remember which one, it was looong time ago these boards were new. Other than that, you must actively cool the SB as well, it gets very hot. If you choose the resistor method. never change its value while the board is running. You'd better put a small heatsink on the voltage regulator, too. And ofcourse...triple-check everything before firing it up. I've had no problems running with vdd mod for a long period of time, but the leak killed it .