Different results on tesla s1070 and GTX8800 why?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, all,
I got the different results running on tesla s1070 and gtx880 with the same code.
And I wonder why? What’s the probable reason.
With the env as fllowing:

Tesla S1070 using one C1060 + nvcc 2.3
GTX8800 + nvcc 2.1

So confused!!

Any suggestion?

2

What does “different” mean? There are plenty of things that could be wrong - building for the wrong architecture, double precision floating point, differences in the hardware itself.

3

avidday,

Thanks for your reply.

The different is the result under tesla s1070 is not the same with that under gtx8800, only a little of differenence.

The result is an integer that count something.

I will consider the tip you brought forward carefully.

Thanks again, and also hope I can find what’s up.

4

Race condition? Are both results acceptable correct results?

On the other hand, you can run some memory test to see if there is any problem with the GPUs.

5

I reinstall toolkit 2.1 and sdk 2.1 on tesla S1070, and with the driver having no change, but I get the the second result, now I focus on the float point and 32bit-64bit OS(tesla server).