Prime95 performance suffers after installing GT 640

[TimSorbet]

Originally Posted by Rodrigo

What to try next?

What happens if you take out the graphics card and go back to the built-in one temporarily? If there's no change, also uninstall the Nvidia drivers.

[Rodrigo]

Originally Posted by kracker

Just curious, what is your PSU?

The specs page for this machine is hardly a model of clarity and organization, but here's what I found:

MT: 300-watt ATX Power Supply – PFC/non-PFC with a 115v/230v line switch (varies by country/region)

• MT: ATX Power Supply – Passive PFC
• SFF: Custom Power Supply – Passive PFC
• Passive Power Factor Correction (PFC/NPFC) – with line switch set to 230V – No PFC in 115V line switch position
• 90 to 140VAC, or 180 to 264VAC operating voltage range
• 100 to 127VAC, or 200 to 240VAC rated voltage range
• 50-60 Hz rated line frequency
• 47-63 Hz operating line frequency range
• MT: 300 watt maximum rated power
• SFF: 250 watt maximum rated power
• 80-mm power supply fan – variable speed for optimum acoustics

(I have the microtower [MT] model.)

Let me know if this info tells you anything we can use.

Rodrigo

[Rodrigo]

Originally Posted by Mini-Geek

What happens if you take out the graphics card and go back to the built-in one temporarily? If there's no change, also uninstall the Nvidia drivers.

Thanks, I'll try this next if nothing else turns up.

Rodrigo

[Rodrigo]

I pulled out the GT 640 (didn't do anything, myself, about the Nvidia drivers), then rebooted and relaunched Prime95.

The per-iteration times are back to about where they were before installing the 640 -- 73 ms on one exponent and 67 ms on the other.

I don't understand this. It didn't happen with the 630 on a nearly identical machine.

Any ideas?

Rodrigo

[chalsall]

Originally Posted by Rodrigo

Any ideas?

Something weird is going on... Might simply be airflow and the resultant thermodynamics (as was previously suggested).

Weird is not necessarily bad -- but it should be investigated further to fully understand.

A test -- if you can, put in a card of the same size as the GT 640 in the same slot on the same machine, but don't "light it" (read: power it or load it). What happens?

[Rodrigo]

What can we do next, as a follow-up?

Rodrigo

[chalsall]

Originally Posted by Rodrigo

What can we do next, as a follow-up?

Rodrigo

If you really want to get empirical...

Cut out a bit of cardboard in the exact shape of the board in question (less the card's slot insert). (With the machine off) electrical tape it into the same position.

Then monitor the temperatures and the performance of the machine.

Some of us do this kind of thing for a living....

[swl551]

Originally Posted by Rodrigo

I pulled out the GT 640 (didn't do anything, myself, about the Nvidia drivers), then rebooted and relaunched Prime95.

The per-iteration times are back to about where they were before installing the 640 -- 73 ms on one exponent and 67 ms on the other.

I don't understand this. It didn't happen with the 630 on a nearly identical machine.

Any ideas?

Rodrigo

Others suggestions may lead to other decisions based on those results, but I'm sure you are open to ideas.....

Based on what I read so far...
If the degradation is immediate then I think power supply is insufficient and preventing the processor from reaching max potential. Thermal degradation takes a bit of time to impact the processor as long as the heatsink is properly installed. If you kept the case open with a fan blowing into it to saturate the interior with cool air and you still have the problem then I would do a PS swap-out test next.

[Aramis Wyler]

Agreed, though It hardly seems possible for an x40 to draw down that much power. I thought the 600 series was more efficient.

[kracker]

Originally Posted by Aramis Wyler

Agreed, though It hardly seems possible for an x40 to draw down that much power. I thought the 600 series was more efficient.

Not *all*. 640 is Kepler's minimum right now.

[Xyzzy]

http://www.geforce.com/hardware/desk...specifications

65W