Compile issue of GPU, and performance question
(too old to reply)
2015-12-16 10:28:37 UTC
Raw Message
Hi, all.

I recently came into the A51 world, and are facing the following questions which I hope someone can give me a hand

1) GPU complie issue:

Both for the linux kraken version and the kraken-win32 version,

the gpu code cannot be compiled due to the non-existence of "cal.h".

Does it base on older version of amd_app_sdk, below v2.0?

2) performance

Looking around the maillist, I found 3 typical implementation are given:

a) Frankson and krasken's version, 3xGPU(ATI 5950), 4xSSD

the performance is not sure. (some paper said It got 2 seconds for one burst)

b) Joel's practice with kraken-win32, 3xGPU(ATI Radeon 6990), 4xSSD

about 2 seconds for one burst, 17~18 for 20 bursts.

c) the recent update of Jan (Deka), 3xGPU(7970) 4xSSD

"5-60 seconds depending on network security, signal quality etc"

How many burst does the minimum 5 seconds correspond to ?

How much has it improved with new GPU and vector program?

3) GPU selection and new SSD

Which part of GPU is more important for A51? Thread number, Frequency, Display memory ?

With recent faster SSD such as Intel SSD (sequential read rate of 2200MB/S, 430K IOPS random read of 4KB.)

More old GPU (Radeon 6xxx) can be used to improve performance with or 2)b) configuration?