13k ppd on GTS450

[Toconator]

Well I couldn't resist. Local shop had a weekend only sale on lots of stuff and I grabbed a MSI GTS450 1GB Cyclone for $99.99 CDN, was gonna grab a GT240 for $29.99 but they sold the last one 1/2 hr before I walked thru the door. I ripped the 1300-1500 ppd HD4850 out of my 2nd Quad rig as soon as I got home and got the GTS450 up and running with the GPU3 client (won't run the GPU2 for some reason, maybe the force g80 flag mighta worked?) and HFMnet reports 13k ppd !? TPF is 1:01 on core15 P10946. Shaders are clocked @ 1700 out of the box on this card which is pretty quick. Anyone else out there seeing similiar numbers or is something freaky on my setup? If this is about right then I suspect I know how jimbowdang, Beaxch, Cypher, and Wolff suddenly boosted their stats. Ha, Ha, gonna have to wait awhile longer before catching me now boys...

 

[Jon855]

GPU2 won't run on it because GTS 450 is a fermi card and requires the GPU3 client - and no, that flag won't work on fermi.

Nice PPD though

 

[APOLLO]

Incredible deal, what province are you in?

 

[Toconator]

Vancouver, B.C. Got the Just Cause2 and Civ5 upgrade bundle as well as Starcraft2 Liberty. Jeez this Bears-Giants game is slow. I'm tempted to load one of these up but I gotta work early in the morning...

 

[flyz]

I also saw that card on sale but I couldn't be arsed to go to the store. That and I've already spent too much this month to build another i7 rig for folding.

Nice PPD btw and I hope those prices will come down more when the 6xxx series comes out in a couple weeks.

 

[jebo_4jc]

(looks at two GTX470 cards that only produce 14k in disgust)

 

[Toconator]

jebo, with 121k ppd that disgust must be fleeting. I have also lost about 1500 ppd from the SMP on the PhII quad but uh, I think it is worth it. My Q6600 OC / GTX 260 combo is only getting 12k ppd right now so I HADTA do something

 

[APOLLO]

Also, by OCing you'll likely hit 14k PPD or more, just be careful how high you set the shaders. I can't believe the deal you got. Second hand GTX 260s are still asking for well over that.

 

[jebo_4jc]

Yea, not complaining. The disgust is fleeting, just like these 470s power efficiency.
Posted via [H] Mobile Device

 

[Toconator]

Also, by OCing you'll likely hit 14k PPD or more, just be careful how high you set the shaders. I can't believe the deal you got. Second hand GTX 260s are still asking for well over that.

Hmm, the MSI Afterburner on the CD didn't work. New version 2.0 from their site works tho. reports only 49C with 51% fan speed loaded with GPU3 running. MUCH nicer than the space heater HD4850 it replaced. cores and shaders are linked on these 400 series cards I see. this is already a factory overclocked card so I don't know if I'll get much more...

/crosses fingers on left hand while loudly proclaiming 'For the Horde!', aims mouse pointer at OC slider ...

 

[APOLLO]

/crosses fingers on left hand while loudly proclaiming 'For the Horde!', aims mouse pointer at OC slider ...

Don't go too high for now. We don't know what kind of OC these cards can take yet. Some people have reported issues. That could be just the luck of the draw but better to err on the safe side. 13k out of the box is very nice.

 

[sirmonkey1985]

(looks at two GTX470 cards that only produce 14k in disgust)

overclock your shaders more.. current gpu3 WU's are shader clock dependent.. its why the g92's did so well on the fermi WU's even though its not allowed..

 

[Toconator]

Don't go too high for now. We don't know what kind of OC these cards can take yet. Some people have reported issues. That could be just the luck of the draw but better to err on the safe side. 13k out of the box is very nice.

not to worry, not going too radical right now. tried 875/1750 and it took no prob. now at 900/1800 and temp is 50C and fan speed is 55%. only 12.9k ppd but on a new protein. not going to unlock voltage or anything. fan did get a little louder at 55%. going to leave it here overnight and see what happens. did not auto load on Windows startup so if it locks and crashes the reboot will be back at defaults.

 

[APOLLO]

not to worry, not going too radical right now. tried 875/1750 and it took no prob. now at 900/1800 and temp is 50C and fan speed is 55%. only 12.9k ppd but on a new protein. not going to unlock voltage or anything. fan did get a little louder at 55%. going to leave it here overnight and see what happens. did not auto load on Windows startup so if it locks and crashes the reboot will be back at defaults.

OK, great you're getting higher stable clocks. Permanent problems were reported by OCs over 925MHz, IIRC, so you should be OK with your settings. BTW, what PPD are you up to so far as a maximum you've seen for this card?

 

[Toconator]

OK, great you're getting higher stable clocks. Permanent problems were reported by OCs over 925MHz, IIRC, so you should be OK with your settings. BTW, what PPD are you up to so far?

current 13355 ppd on core 15 P11220. downloaded at 10:50 pm, ETA 12:30 am. this sucker folds [H]ard !!!

 

[10e]

Dammit, I missed the sale totally, and I have two locations five minutes from where I live.

Chump = me

 

[jebo_4jc]

overclock your shaders more.. current gpu3 WU's are shader clock dependent.. its why the g92's did so well on the fermi WU's even though its not allowed..

Trying, but these cards can't go all that high on the shader clock. Is there a way to set independent clocks for a dual cards setup?
Posted via [H] Mobile Device

 

[sirmonkey1985]

Trying, but these cards can't go all that high on the shader clock. Is there a way to set independent clocks for a dual cards setup?
Posted via [H] Mobile Device

evga precision.. press the sync button where it shows the 2 cards in the center.. i believe you can do it in afterburner as well..

 

[Zero82z]

Is there a way to set independent clocks for a dual cards setup?

Go into settings and uncheck "Synchronize settings for similar graphics processors".

 

[Camaroz06]

So a gts450 out folds a gtx460?

 

[mrbigshot]

no, its a 912pt unit so it throws the ppd calculations off.

a heavly overclocked 450 gts gets around 9100 pts if you look at the actual amount they make, not using the estimator. bone stock there around 7500. there are a few bigger point units like you see in the P112xx's that will get you a little more but not 5k more. it would out fold a 460 (11k),465(12k), and 470 (14k) if that were the case, that would be quite an accomplishment with what? 35% less shaders (just a guess at the percentage, they have 192 vs the 460's 336).

 

[jebo_4jc]

Go into settings and uncheck "Synchronize settings for similar graphics processors".

that's in afterburner right? sorry I didn't specify.

Afterburner is clearly superior to evga precision from what I've seen so far.

And, to answer Camaroz' question, it does appear that the 450 out folds every other fermi card out there, as a matter of fact.

 

[W.Feather]

i though ti read / heard somewhere that you cannot unlink the shader from clock speed on FERMI, but hell i'm probably wrong

 

[jebo_4jc]

i though ti read / heard somewhere that you cannot unlink the shader from clock speed on FERMI, but hell i'm probably wrong

you're right

 

[Zero82z]

that's in afterburner right? sorry I didn't specify.

Afterburner is clearly superior to evga precision from what I've seen so far.

And, to answer Camaroz' question, it does appear that the 450 out folds every other fermi card out there, as a matter of fact.

Yes, it's in Afterburner. Afterburner and Precision are pretty much the same program; aside from Afterburner being able to adjust Vcore, I don't see how it's really superior.

 

[jebo_4jc]

yeah they're both based on rivatuner obviously, but the vcore adjustment is huge, particularly for the 450 and 460 which appear to respond very well to voltage bumps

afterburner's graphical control over fan speed is another big plus IMO, although that feature may also be present in Precision in later versions. I might have been using an old version of Precision.

 

[Zero82z]

I just set my fan speeds to the highest percentage with a noise level I can stand, so Precision and Afterburner both work equally well for me in that respect.

 

[jebo_4jc]

I just set my fan speeds to the highest percentage with a noise level I can stand, so Precision and Afterburner both work equally well for me in that respect.

that has been my approach over the past couple years as well. However, since i do actually live in/around the areas of my GPU folders, I've found it very helpful to be able to set specific (and aggressive) speed profiles for my fans. Turns out a 9800gx2 doesn't need to be at 100% to stay below 90c

 

[Zero82z]

Well, by highest noise level I can stand, I mean I set them as high as they go without being audible over the other noise around them.

 

[Toconator]

getting 14271 ppd on core 15 P10971 with 900/1800 clocks on this GTS450. haven't tried voltage bumps yet. i'd like some feedback on what other peeps have been successful with b4 I do. reason I opted for the 450 instead of a 460 was budget restraints so I don't want to fry this thing and revert to the (shudder) HD4850

 

[APOLLO]

no, its a 912pt unit so it throws the ppd calculations off.

a heavly overclocked 450 gts gets around 9100 pts if you look at the actual amount they make, not using the estimator. bone stock there around 7500. there are a few bigger point units like you see in the P112xx's that will get you a little more but not 5k more. it would out fold a 460 (11k),465(12k), and 470 (14k) if that were the case, that would be quite an accomplishment with what? 35% less shaders (just a guess at the percentage, they have 192 vs the 460's 336).

Well, irrespective of how production is achieved, bottom line is what credit you receive for work submitted to Stanford and an OCed GTS 450 is capable of 14k-15k PPD, which is right on par with what most people have seen with their GTX 460s. Granted, these are the -advmethods WUs and not the standard GPU3 WUs released earlier, but I wouldn't think of folding anything else with Fermi ATM.

getting 14271 ppd on core 15 P10971 with 900/1800 clocks on this GTS450. haven't tried voltage bumps yet. i'd like some feedback on what other peeps have been successful with b4 I do. reason I opted for the 450 instead of a 460 was budget restraints so I don't want to fry this thing and revert to the (shudder) HD4850

Good that you're holding back a little. I understand your concerns and would consider a 450 myself since the savings over a 460 get compounded with more than one card. Also to be considered is the long-term power consumption savings of the 450 over the 460. I would probably bump the clocks up a bit more if you're still running at 900 but would be wary going much higher than that.

If your card is running at reference volts you might be able to increase that a bit. I would not increase the voltage on a factory over-volted card. There have been reports about damaged 450s on several forums already. It's just too early to know if this is a manufacturer-related issue or OCing damage. That's why I posted about exceeding ~925MHz, which admittedly is an arbitrary figure on my part. The Asus and a few other cards come with those clocks out of the box, but not the MSI. Anyway, for those interested in the GTS 450 and which cards have what factory frequencies, look at this chart: http://www.guru3d.com/article/geforce-gts-450-review-roundup/26

 

[Agent_N]

no, its a 912pt unit so it throws the ppd calculations off.

a heavly overclocked 450 gts gets around 9100 pts if you look at the actual amount they make, not using the estimator. bone stock there around 7500. there are a few bigger point units like you see in the P112xx's that will get you a little more but not 5k more. it would out fold a 460 (11k),465(12k), and 470 (14k) if that were the case, that would be quite an accomplishment with what? 35% less shaders (just a guess at the percentage, they have 192 vs the 460's 336).

Actually, shader clock makes the most difference in terms of PPD, not so much the number of stream processors, though that can increase output, but not like a shader clock increase. If I bump my shader clock from 1836 to 1890 on my 9800GT I can net another 100+ PPD. With the 450's shaders running about the same or higher than mine, I am not surprised that 14K is possible.

 

[Toconator]

just to compare, I have 3 other GPUs folding right now:

9800GT @1620 = 4K ppd, GTX260-216 @1512 = 6.3K ppd, GTS250 @1512 = 4K ppd
all are folding core 11 - P10109 in combo with SMP on all CPU's

 

[APOLLO]

just to compare, I have 3 other GPUs folding right now:

9800GT @1620 = 4K ppd, GTX260-216 @1512 = 5.6K ppd, GTS250 @1512 4K ppd
all are folding core 11 - P10109 in combo with SMP on all CPU's

Yeah, the G92s and GT200s are starting to show their age with these P10109 WUs, and they're the only thing I've been seeing the past few days..

 

[jebo_4jc]

Well, irrespective of how production is achieved, bottom line is what credit you receive for work submitted to Stanford and an OCed GTS 450 is capable of 14k-15k PPD, which is right on par with what most people have seen with their GTX 460s.

...and 470s and 480s

 

[Zink]

So spending $500 will get you 50k ppd in gpus? If Standford keeps WUs like this then adding 50k ppd to an SR2 might make sense... (with 1kw psu of course)
The cards should be right around 100ppd/w so they aren't that inefficient.

 

[APOLLO]

So spending $500 will get you 50k ppd in gpus? If Standford keeps WUs like this then adding 50k ppd to an SR2 might make sense... (with 1kw psu of course)

Unfortunately, these cards would still be a poor choice for SR-2 or other high end mega-core systems running -bigadv SMP. The -advmethods GPU3 WUs are extremely high in CPU cycle use, and I have read about the GTS 450s being the worse of the Fermis in this respect. So, best to use these in low end processor systems.

The cards should be right around 100ppd/w so they aren't that inefficient.

They're the only Fermi so far to have a single 6-pin connector like the G92 cards. If anything, that should give a rough idea about power consumption.

 

[Toconator]

The -advmethods GPU3 WUs are extremely high in CPU cycle use, and I have read about the GTS 450s being the worse of the Fermis in this respect. So, best to use these in low end processor systems.

I didn't set the GPU3 with the -advmethods flag but maybe they all are anyway? I did lose some SMP ppd off my Ph II X4 945, approx 25%. Lose 1500 ppd SMP, gain 11.5K+ ppd GPU. I'm not in the same league as SR2 tho, I don't think I can efficiently run bigadv units with my stuff anyway. I think GPU is the best way for me to get points for the Horde for now. Was toying with the notion of maybe a 1075T if I can scrape up enuff for it and a mobo and RAM and go headless with it. Can these take on bigadv? If not I'll opt for a 460 to add with my 260. Hmm, my lab/office/HT is already 80F with all the gear in here. I might need to add A/C too!

/proclaims " [H]eating Canada for the [H]orde! " as he uncorks the bottle...

 

[jebo_4jc]

Short answer, no, x6 can't do bigadv.

Long answer, well, is a long answer.....
Posted via [H] Mobile Device

 

[APOLLO]

I didn't set the GPU3 with the -advmethods flag but maybe they all are anyway?

It could very well be. I stopped running GPU3 when PG made it clear they were off limits to pre-Fermi hardware. That was a while ago...

I did lose some SMP ppd off my Ph II X4 945, approx 25%. Lose 1500 ppd SMP, gain 11.5K+ ppd GPU. I'm not in the same league as SR2 tho

There have been some threads on GPU+SR-2 and it is very tricky to see an advantage because the losses in -bigadv bonus are enormous for the gains, even from Fermi. Best case scenarios people reported were only a slight advantage in PPD but a great increase in power draw. Mind you that was with 460s, 470s and 480s not GTS 450.

Can these take on bigadv?

No, there's a minimum of 8 cores requirement to run -bigadv.

If not I'll opt for a 460 to add with my 260.

Just get another 450 unless you play a lot of games. The price difference here in Canada between the 450 and 460 is greater than in the US., and there doesn't seem to be a huge PPD advantage for the latter. The power consumption should be less as well.