GTS 450 PPD

[ADay2Long]

Woohoo GTS450 is on the way!

Now that bigadvs are getting rare, I feel GPU folding is just easier.

[Soultribunal]

Originally Posted by boggtrotter59

Excellent... my current favorite folding card is the GTX260, they run very cool... rarely break 70 degrees C while folding (don't have to babysit them) ... was thinking of upgrading a few of them to the GTS450's. If I can get the numbers you guys have posted I think it will be well worth it.

It really is, heck even consider how a card that requires no PCI-e Connector (GT430) can pull 8800ppd.
These cards really do well on the new GPU front, I can't wait till i get a rig worthy of them this december/jan


ST

[sswilson]

Since this thread is titled "450 PPD" my latest observations would probably fit best in here.....

I've now got a 450 running on both a Q6600 @ 3.6G and an i3 530 @ 4.4G. Both cards are running the exact same OC (975/1950/2100) and there does not appear to be any difference between the GPU3 PPD for the lower clocked CPU vice the GPU3 PPD on the higher clocked CPU. This suggests that in spite of using more CPU resources, the higher PPD GPU3 WUs don't seem to be affected by CPU clock speed.

We'll see how this works out as it looks like I might have more GPU OC overhead left on the lower clocked Q6600 rig. More to follow.

[Sagath]

Considering my 460 on the 9xx units uses about 3-4% of my CPU with a 9xx WU (6-7% for both cards obviously), I highly doubt that a Q6600 will bottleneck the 9xx pointers on a 450. That being said rather then subjectively debating this, why not objectively give us some core usage of both rigs?

This is not an attack, I'm honestly curious of the info Steve, because right now, folding with GPU3 limits my SMP folding too much to really make it worth it imo in ppd/watts. It just kills my TPF having my CPU at a constant 8% usage, so I'm stuck in the old GPU1 and 2 era choice when they used a lot of CPU cycles; Do I SMP fold, or GPU fold?

[Xeven]

@Sagath

Just curious, are you limiting the number threads for your bigadv? In my 920 I had to limit to -smp 7 since the beta gpu3 units (2X460) eat so much cpu cycles I can't even make the deadline on the bigadv. Coupled with setting the affinity of the gpu3 processes to a single core (I haven't actually verified if this makes a difference), I normally achieve a 25-27K PPD on good bigadv units and a measly 17K on the PITA P2684. If the GPU clients are processing normal gpu3 units however, it doesn't seem to matter and the bigadv unit runs fine with the default setting.

[sswilson]

Core usage:

Single GTS450 985/1970/2100 / Q6600 @ 3.6G / 10936 (Run 1, Clone 28, Gen 6) 925 pointer / 6.2% total CPU usage reported by Realtemp

Single GTS450 975/1950/2100 / i3 530 @ 4.4G (2 Core HT) / P10970 (R1, C26, G11) 925 pointer / 8% total usage CPU usage reported by Realtemp


The main reason I commented on the 3.6G vice the 4.4G was that we'd been thinking that perhaps the high clock speeds that I had was boosting PPD on the units that use more CPU cycles. That doesn't appear to be the case.


edit: Heh... scratch the 985 on the Q6600 rig, just EUE'd on a 611 pointer. Back down to 980..... ;)

[Sagath]

Thanks for that info Steve, interesting the 530 uses more horsepower then the Q6600!

As to the SMP questions, Negative on the -smp 7 flag.

I ALWAYS get SMP errors when doing this on my rig. There is some threads about it on the official F@H forums, basically they say any odd numbers are bad because of the way Folding is set up to run, and its not recommended. Now, that being said, lots of people run it successfully that way, and a few cant. For whatever reason, I'm one of those few.

[sswilson]

Originally Posted by Sagath

Thanks for that info Steve, interesting the 530 uses more horsepower then the Q6600!

As to the SMP questions, Negative on the -smp 7 flag.

I ALWAYS get SMP errors when doing this on my rig. There is some threads about it on the official F@H forums, basically they say any odd numbers are bad because of the way Folding is set up to run, and its not recommended. Now, that being said, lots of people run it successfully that way, and a few cant. For whatever reason, I'm one of those few.

I guess that's what "virtual" vrs real cores gets you when it's not a lot of threads at minimal usage.

On another note.... just pulled two 611 WUs. Looks like it's the same PPD on both rigs (both @ TPF of 53 seconds, so the higher clocked one is probably 53.2 while the other is 53.5, but HFM reports the same 3% average because of the 53).

[boggtrotter59]

Originally Posted by sswilson

The MSI cyclone on an open bench, and stock fan profile runs around 55C with max extra core voltage. The ASUS TOP (max extra core V as well) runs under 50C with a custom profile and fan speeds between 80 - 90%.

Does the cyclone take op more than 2 slots? I checked pictures of the card online and it doesn't appear to take up more than 2 slots... It has a weird fan on it... these sometimes don't fit when the spacing is tight, just thought I'd check before I pull the trigger (they are currently $109 at NCIX)

[sswilson]

Nope, same width as a traditional 2 slot plastic shrouded card. Grab one quick... it's already OC'd to 850, the cooler is phenomenal, and hopefully you'll have as much luck with overhead on the OC as I did!!!! :)