Are you trying to use the psu that came with the machine? If so this is pretty normal most psu's that come with the machine are just barely over the line of being able to run the machine itself, smaller is cheaper and over the course of a run of building pc's even 10 cents per pc can add up.
no, this is kinda old 750w psu from server case, but it seemed it can't power radeon 280X + radeon 7970 at full load :/
If it's old, it may not have sufficient current at 12V. You shouldn't be looking at 750W, just the rated current at 12V.
I power a lot of my machines with a 300W PSU which can deliver 22A at 12v (270W max). When I add a GPU, I add a spare 175W PSU which can deliver 8.5A at 12v. I use this 2nd PSU to power just the 6pin GPU connector. A bigger 2nd PSU could be used if you had higher requirements for your GPU (ie 2 connectors).
i have another supermicro 865w psu. it easily can power two 7970, but now i'm waiting for the first fury x available in pc stores to built a refreshed crunching machine
That's sometimes true, but not universally necessary. If it does turn out to be the case, many (most?) modern monitors these days come with dual VGA / DVI inputs, and cables to match, and most GPUs come with either dual outputs or DVI / VGA adapters. With a combination of those components, you can often hook up the GPU needing the load to a spare monitor input - ugly, but quicker than a shopping trip.
I have noticed a strange issue when running 2 identical HD 7950 GPUs.
They are both running 1 BRP6 task, but the run times differ significantly: d0 takes about 6100 s, while d1 ~3100, which is almost double !'t explain the huge difference.
I didn't notice anyone addressing this directly.
You may already know this, have it all under control, etc., but I can't tell from looking at anything I can see about your host from here.
A lot of motherboards will have two, or even three PCIe x16 slots, but only one of them has enough "lane" connections to the CPU to run at maximum x16 speeds.
It is common to have one full-speed x16 slot and have a second x16 slot that is running with x4 bandwidth. It is also common to have a x16 slot and another slot that, if populated with a card, will cause the first slot AND the second slot to run at x8 speed (they share "lanes").
Some projects do not require a lot of CPU-involvement and you will barely see any difference in runtimes whether a card runs at x8 or x4 or x16.
Einstein@Home, at least the tasks I have run, does require a lot of "talking" between the CPU/main memory and the GPU. This motivated me to buy some expensive motherboards that have four x8 slots. (this is not strictly possible; the motherboard has a multiplex chip that allows any slot to run at x8 when needed but I don't think it is possible to REALLY have four x8 slots running full-speed at one time on an AMD board)
If more than x4 speed could be used, but only x4 speed is available (hard-wired on the motherboard), obviously things slow down.
A piece of monitoring software like rh-software's System Information Viewer would report the slots' speeds for you. If that isn't the issue, it might help point you to where the issue is.
EDIT: Apparently the CPU you have has the 40-lanes available. The question would be if the motherboard chipset does.
SECOND EDIT: It is also entirely possible that you just have too much going-on. If you stop processing on the CPU entirely, does the speed of the second GPU increase?
Thanks for the feedback guys. Indeed, you're right, I totally forgot about the PCIe lanes, so I checked and:
I prefer HWiNFO as I'm its author ;) So really shame on me that I overlooked this parameter!
But anyway, I wouldn't expect this to have such a huge impact on the performance.
I have reseated the 2nd GPU into an x16 slot, so now both have identical parameters. I was sure this was it, but it didn't fix it. Still there's an almost double difference in run time. The slower GPU has utilization of 80% (CPU 47%), while the faster one 90% (77%). No idea why...
RE: Any idea what might be
)
Do you mean this host crunched on both GPUs in the past and now one has slowed?
You don´t say what bus interface (PCI-e) each gpu is running. gpu-z should report this information.
RE: RE: oh, it works
)
no, this is kinda old 750w psu from server case, but it seemed it can't power radeon 280X + radeon 7970 at full load :/
server chassis
)
server chassis
RE: ... this is kinda old
)
If it's old, it may not have sufficient current at 12V. You shouldn't be looking at 750W, just the rated current at 12V.
I power a lot of my machines with a 300W PSU which can deliver 22A at 12v (270W max). When I add a GPU, I add a spare 175W PSU which can deliver 8.5A at 12v. I use this 2nd PSU to power just the 6pin GPU connector. A bigger 2nd PSU could be used if you had higher requirements for your GPU (ie 2 connectors).
Cheers,
Gary.
i have another supermicro
)
i have another supermicro 865w psu. it easily can power two 7970, but now i'm waiting for the first fury x available in pc stores to built a refreshed crunching machine
There's one other matter for
)
There's one other matter for running multiple GPUs.
Do you have a separate monitor attached to each graphics card? GPU cards must be under load to run.
A cheap and easy solution is to make a dummy VGA plug for the card that doesn't have a monitor attached. Then re-start the machine.
Here's a link that I found helpful.
http://www.geeks3d.com/20091230/vga-hack-how-to-make-a-vga-dummy-plug/
Regards,
Steve
RE: There's one other
)
That's sometimes true, but not universally necessary. If it does turn out to be the case, many (most?) modern monitors these days come with dual VGA / DVI inputs, and cables to match, and most GPUs come with either dual outputs or DVI / VGA adapters. With a combination of those components, you can often hook up the GPU needing the load to a spare monitor input - ugly, but quicker than a shopping trip.
RE: I have noticed a
)
I didn't notice anyone addressing this directly.
You may already know this, have it all under control, etc., but I can't tell from looking at anything I can see about your host from here.
A lot of motherboards will have two, or even three PCIe x16 slots, but only one of them has enough "lane" connections to the CPU to run at maximum x16 speeds.
It is common to have one full-speed x16 slot and have a second x16 slot that is running with x4 bandwidth. It is also common to have a x16 slot and another slot that, if populated with a card, will cause the first slot AND the second slot to run at x8 speed (they share "lanes").
Some projects do not require a lot of CPU-involvement and you will barely see any difference in runtimes whether a card runs at x8 or x4 or x16.
Einstein@Home, at least the tasks I have run, does require a lot of "talking" between the CPU/main memory and the GPU. This motivated me to buy some expensive motherboards that have four x8 slots. (this is not strictly possible; the motherboard has a multiplex chip that allows any slot to run at x8 when needed but I don't think it is possible to REALLY have four x8 slots running full-speed at one time on an AMD board)
If more than x4 speed could be used, but only x4 speed is available (hard-wired on the motherboard), obviously things slow down.
A piece of monitoring software like rh-software's System Information Viewer would report the slots' speeds for you. If that isn't the issue, it might help point you to where the issue is.
EDIT: Apparently the CPU you have has the 40-lanes available. The question would be if the motherboard chipset does.
SECOND EDIT: It is also entirely possible that you just have too much going-on. If you stop processing on the CPU entirely, does the speed of the second GPU increase?
Thanks for the feedback guys.
)
Thanks for the feedback guys. Indeed, you're right, I totally forgot about the PCIe lanes, so I checked and:
I prefer HWiNFO as I'm its author ;) So really shame on me that I overlooked this parameter!
But anyway, I wouldn't expect this to have such a huge impact on the performance.
-----
I have reseated the 2nd GPU
)
I have reseated the 2nd GPU into an x16 slot, so now both have identical parameters. I was sure this was it, but it didn't fix it. Still there's an almost double difference in run time. The slower GPU has utilization of 80% (CPU 47%), while the faster one 90% (77%). No idea why...
-----