We are buying new workstations and want to put as much memory in them as we can affort. We can get 32 GB 667 MHz moduls for the same price as 16 GB 800 MHz so right know we are going for 32 GB moduls.
However, dos anyone know if it will affect the renderspeed much choosing 667 MHz moduls instead of 800 MHz moduls?
Workstation specs:
Dell T7400.
- Dual quad Xeon X5450 with memory riser (3.00GHz)
- ATI FireGL V7700 512 MB
- SAS Hard Drives (15000Rpm)
- Vista Business 64 Bit
- 30" Monitor Dell 3008WFP
In my experience I found that a faster CPU makes a bigger difference than faster RAM. The difference between 667 and 800 is not much unless you’re doing scenes with a gazillion polys you’re looking at a few seconds if that much. Well in my experience anyway…
from my understanding it works something like this:
if your cpu frequency is lets say 1333 mhz, and your ram frequency is 667, then your cpu will be forced to work at 667, which will be a significant performance loss when a cpu will try to address ram to read/write data.
I had tried an overclocking test, where i took my modules from 800 to 1333, and i could see about 10-15% performance gain during render time.
@Morbid Angel
It seems right - however if the perfomance bust is “only” 10-15% going from 800 to 1333 I guess it will be less from 667 to 800, right?
I dont know much about OC but doesn’t have to be a special type of ram moduels to take the heat?
Our work is mainly large stills with a lot of high-res maps applied. Right now we only have 1 GB ram per core and it´s a daily struggle so we think that 2 GB of RAM in a new workstation is not enough and therefore choose 4 GB. But are there others who work with 2 GB of RAM without any problems?
Are there any other thoughts/comments to the workstation specifications?
yes you would need a specialized ram modules for overclocking, and also overclocking can get unstable, overall I dont think its worth it, unless you have a pro doing it, and he can guarantee success.
But I can tell you that at work we are running 8 core 8 gb ram machines with clock speed of 2.0 at 1333 and ram of 667, and at home I got a 4 core 4 gb machine, with proper tuning of cpu and ram I can get it to render nearly as fast as the 8 core, perhaps at about 15% slower.
In the long run though, the more ram the better. If you guys work with extreme resolutions, then you can probably sacrifice speed but gain simplicity and stability.
Sure. Though its a little old as well, maybe 3 years old, but still woops the dual xeon’s butt.
cpu = qx6700 extreme default clock @ 2.66 ghz fsb 1333 (I overclock to 3.3ghz)
ram = corsair 4 1 gb ddr 2 @ 800 mhz (overclocked to 1.333 but not very stable at that)
mother board = asus p5ws 64 pro.
Few notes on this setup:
Overclocking works well only for rendering. So for example if you are doing some other heavy processing like realtime rendering which addresses the SSE instruction it will reboot the pc.
Also you cannot get the intel quad core in dual configuration, thats why I only have single 4 core.
Actually you can know but only with the Intel Skull Trail Mobo
A few months ago I was looking at a setup with a Gigabyte Mobo, Intel Q9650 and some Patriot PC6400 DDR800 ram to build a cheap renderfarm. But I have to read up some more on Over Clocking so I don´t burn them
I also wanted to build the workstations that we know are buying from Dell but we don´t have the time as we needed them a week ago.
Interesting. Yes when I was buying my machine, there were no known motherboards who could support that.
Also asis p5w64 pro is very easy to configure for overclocking. There are direct options for non experienced users, for example as easy as just choose - overclock to 10%, 15%, 30%.
Or if you are advanced user you can manually configure the timings.
I was looking for something I stumbled on this. Its a comparison between DDR2 and DDR3 ram with 800 vs 1333 clock speeds. The article is in russian, but you can see by the images the performance benchmarks.
well there is more tests to be conducted. I would try to go with the fastest ram available and the most reasonable cost
eventually, over a period of time, this will play a role, I mean if your render takes say 30 seconds with 667 module and 25 seconds with 800 module, then it does not matter really. But if you render takes 2 hours with 667 module and 1.8 hours with 800 module and you need to render 1000 frames of animation…you can see where the difference will play.
But you are right, our main goal is stability.