I have just learned that a GT 640 will work with my current set up, and it is probably the best graphics card out there that will work with my frankenstein system, which consists of parts salvaged from a University tech recycling program. I had to buy some things to make it runnable such as a hard drive, which is when I picked up my GT 440 OEM.

My question is how much of a performance boost should I expect from a 640?

My 440 has half as many CUDA cores, but finishes short runs between 7 and 15 hours ( depending on the project), and for long runs it usually finishes in under 48 hours, but always more than the 24 hour time limit. It currently is working on a massive Noellia project, which looks like it will take a hair over 60 hours.

Sadly I can not find the Gflops of my GPU as no tech site has gotten a hold of and rated the 440 OEM to compare to the 640. So I am hoping to compare in terms of average run times on projects.

---

Do you know which of the various 640s it is?
Look it up here: http://en.wikipedia.org/wiki/GeForce_600_Series
Within the Kepler range, performance should scale linearly with GFLOP rating, so you can estimate based on the performance other Kepler users report.

MJH

Sadly I can not find the Gflops of my GPU as no tech site has gotten a hold of and rated the 440 OEM to compare to the 640.

Take a look at the thread More NVidia confusion, this time with the GT440.

For calculation GFLOPS see the data ...
http://www.nvidia.com/object/product-geforce-gt-440-oem-us.html
http://www.geforce.com/hardware/desktop-gpus/geforce-gt640/specifications
http://en.wikipedia.org/wiki/FLOPS#Computing
... and you can get the theoretical performance.

My question is how much of a performance boost should I expect from a 640?

Perhaps you should examine those results, I listed one GT640 in my comparison with GTX 650TI.

GT640 on Vista, 1.14 GHz core clock if I remember correctly. A long-run with 90k credits (without bonus) took 143k seconds.

MrS

---

Scanning for our furry friends since Jan 2002

Thank you everyone.

I oddly never knew there was a somewhat simple formula that could be used to calculate the total Flops.

I am really thinking an upgrade is in order.

---

You might also want to consider the GTX650Ti. It's got twice as much raw horsepower than the GT640, yet the power consumption under typical load is still a modest 60 - 70 W, at max 85 W. That should fit into any modern system and can still be cooled silently.

And the formula for the maximum GFlops is really simple:
(operations per shader per clock) * (number of shaders) * (clock speed of shaders)

Operations per shader per clock are usually 2. Insert clock speed in GHz and you get GFlops directly. There's nothing more to calculating maximum GFlops.. what this number means, i.e. how much of this raw horsepower can actually be brought onto the street, is a different question. Though the answer is usually "the same percentage" in each chip generation, or more precise: for chips with similar CUDA compute capability level.

MrS

---

Scanning for our furry friends since Jan 2002