I looked at dual 270’s (due to drop in price soon) but I think you get more power for the money out of the 4200’s. For less than $4000 you can get 3 4200’s with a total of 6 gig ram. I think that would significantly out perform 2 270’s mounted on a single board sharing 4 gig ram and resources. The rack mount thing is cool but I couldn’t justify the additional cost (one man operation). I own two Boxx workstations that have been great though.
I posted this question X2 or dual Opteron on the Autodesk hardware forum and got an informative response that includes power consumption.
To get an idea of the performance metrics of each CPU, we can base this using
the 3ds max 7.0 scores as reported by GamePC:
Benchmark scores:
1 x X2 4200: 161
2 x Opteron 275: 144
(lower times are better)
Given this, we can see that a dual Opteron 275 system kinda spanks the single X2
in rendering tasks, being around 10% faster. No surprise, as you are pitting 4
CPUs of computing horsepower against 2, and 3ds max/VIZ is one of the few apps
that can actually use these extra CPUs. But is that 10% enough to make your
rendering farm better? Especially since there seems to be a point of diminshing
returns after 2 CPUs in max.
At the hardware level, the Opterons give you 3 boxes to deal with instead of 5.
That means a savings by not having to pay for 2 extra systems, not to mention
lower shipping costs. This savings is somewhat offset by the added cost of each
dual Opteron system, however. We can probably call the price a draw. Network
connections are essentially free and not part of the cost issue.
In terms of power consumption, GamePC’s benchmarks show the Opteron systems
consume 282 Watts under full load, while the X2 only consumes 181 Watts. A
3-system Opteron farm is 846 Watts, while the 5-system X2 farm is 905 Watts.
Thus, the Opteron farm will cost about 7% less to operate, which is also about a
wash.
But to make a final determination we must compare IF the additional 2 X2 systems
make your distributed rendering system more efficient and make up for the slower
rendering times against fewer Opterons.
Let’s assume a single X2 will render a scene in 10 hours. Given the benchmarks
above, one dual-CPU Opteron system will render the same scene based on the
equation
[(10 / 161) = (x / 144)]
Solving for X = 8.94 hours for the Opteron system.
Assuming the additional systems in the rendering farm divide a task equally, 5
X2 systems would render the scene in only 2 hours. However, 3 Opteron system
would render the scene in 8.94 / 3 = 2.98 hours. 4 Opteron systems would finish
in 2.23 hours. It would take 5 Opteron systems to see any time savings!
Conclusions:
The marginally faster rendering times provided by the Opteron do not scale
against using more but slower X2 systems in this scenario. At this level it
takes the same number of Opterons to beat the same number of X2 systems.
Go with the 5 X2 systems.
