SunFire T2000 quick and dirty benchmark

Somehow we have recieved brand new SunFire T2000 in our laboratory last week. Many of our engineer are really eager to try out (we do test almost all software at Sun which being localized, so this could be an interesting target machine for all of us). But official real work put a side :-), I have run the very simple test on this machine. Before show you the number, I have to put the disclaimer,

in this blog, I will show you the number from my personal test, but non of the number does not mean to be official number at all. I ran this for just fun. Numbers and graphics may be not accurate nor may not represent what product would offer. The benchmark software I have used here are called c4, I have grabbed from somewhere but could not remember where anymore. You should not take number nor graphics without this disclaimer. Thanks,

So, test program I have run was called "c4" which I had copy on my home directory for ages, and did not remember where I got it. But it looks like relatevly small interger only program. I have wrote small script to ran this program for multiple instantace at same time, and measured total run time. Following are basic configration of the each machines:

• T2000: UltraSPARC T1@1,0GHz/6 Core, 8GB
• E5000: UltraSPARC II@400MHz * 8, 8GB
• SF480: UltraSPARC III+@900MHz * 2, 8GB

Those may be little older machines, but this is what I could accessed right away. Please note that T2000 are 6 core instead of 8 core, also running in 1.0GHz instead of 1.2GHz, you may consider this as smaller configuration of T2000. While all of machine are mostly idle, but not exactly (you may see some noise in data which may triggered by someone accessing the machine including T2000 itself, but I decided complete idle may not show significant differences).

Here goes result which run upto the 48 simultaneous instances (in this case same as process as thread). Lower the number are better/faster.

This looks quite impressive to me. I would have expected upto 24 thread/prorcess would be gradual increase in time/number, but even past 24 still looks gradual. Closer look may see some changes in about 24 to 25 processes where curve become little bit steeper.

Now take closer look into the up until 12 processes.

T2000 number are about same as E5000 (in this case of benchmark) until about 8 or 9 processes. Which match with E5000's configuration of 8 processor. SF480 shows single core's strength (about two time faster), but lead quickly fade away from 5 processes. In overall, I am impressed with T2000 ability to handle multi thread/process. Especially more than 24 thread/process handing which is more than hardware support for the multithread shows. Obviously your mileage varies, since this is very simple test case. But my speculation (in past blog written in Japanese) may suffice in some dgree (I was talking about 8 core UltraSPARC T1 may represent half of E10000's performance).