OK, time to get a new rig
Am thinking of populating my new rig with 32GB (8 X 4GB) of RAM.
Want to know if anyone know how much RAM speed matters in rendering time.
No, I am not talking about over-clocking (OC).
What I mean is, if I populate my new rig with 8 X 4GB of 2133 (or faster) DDR3 DIMM, how much faster would the rendering speed be, compared to 8 X 4GB of 1333 DDR3 DIMM?
One more question: Is the "CAS Latency" thingy important in the speed of DDR3?
The fastest DDR3 DIMM on the market now is rated at 2800, but it has a CAS Latency of 11
On the other hand, DDR3 DIMM that are rated at 2133, CAS Latency at 9, and DDR3 DIMM that are rated at 1333, with CAS Latency of 7
How much effect the CAS Latency has on the speed of rendering?
Neither will have a significant effect on render time. RAM is not a significant factor in rendering *speed* in general, you just need to make sure you have enough of it to render your scenes at your intended resolution. As long as that condition is met, slightly faster RAM will not matter much. 32GB is a great amount to have. CPU speed and number of cores/threads are the biggest factors.
- Oshyan
Here's a couple of articles:
http://techreport.com/articles.x/20377
http://www.anandtech.com/show/4503/sandy-bridge-memory-scaling-choosing-the-best-ddr3/1
generally the difference is very marginal if you dont OC at all.. If one does OC, then its a whole different matter
Forgot to mention:
The mobo is LGA 1155 based, and the CPU I'll be using is i7, quad-core (8-hyperthreads).
CPU speed is yet to be decided.
Most probably Sandy Bridge since there are online reports of Ivy Bridge overheating.
I havent read reports of ivy bridge overheating but what I do know is the fact that its runs a bit hotter by default but it's not an issue. In the end it's not even that much faster, just some slight improvement...clock to clock speed is a bit better. If you are not planning to OC then Id suggest Ivy, but if you are then Sandy.
(http://plaza.fi/s/f/editor/images/sb_vs_ivb_povray2.png)
(http://plaza.fi/s/f/editor/images/sb_vs_ivb_cinebench.png)
Hmm... interesting !
Thanks for all the tips !!
Umm ...
http://www.tomshardware.com/reviews/ivy-bridge-benchmark-core-i7-3770k,3181-10.html
(http://media.bestofmicro.com/C/I/334818/original/mem%20scaling%20sandra.png)
Looks like higher speed DDR3 RAM does enable a higher memory bandwidth
While in the real world application, using WinRAR ...
(http://media.bestofmicro.com/C/K/334820/original/mem%20scaling%20winrar.png)
... looks like DDR3 1866 is the optimum
Wonder if it's the same for TG2 ?
The closest comparison you're going to get to tg in terms of performance on Toms Hardware is the 3d studio max ratings. Buckets and clock speed with enough ram to hold all your mask/texture/model/heightfield/resolution is IMO the pecking order to performance.
On a side note, we're getting a 32 core system in the next week to handle real time interactives the geek department want to churn out. I can't wait to do a tg benchmark on it!!!
Quote from: Hetzen on May 09, 2012, 05:09:41 PMOn a side note, we're getting a 32 core system in the next week to handle real time interactives the geek department want to churn out. I can't wait to do a tg benchmark on it!!!
Oh wow !!
Care to share with us the specs of that 32-core system, it that's possible, I mean?
Sure. It's a dual Zeon Octo-Core which has 16 sandybridge cores (32 threads) running 3.1Ghz and a turbo frequency of 3.8Ghz. It has 64GB Ram and a Fire Pro 9800 4GB RAM graphics card. It has 1 SSD OS drive and 4X SSDs raided to achieve 2GB per second transfer rate. *drool*
I've only had a quick play with it and got a benchmark of 6.40 minutes, with what looked like only 16 buckets. I need to investigate further why I haven't got all 32 cores rendering, and unfortunately I won't be able to look at that until next Wednesday.
sounds like one hell of a beast machine ;)
Quote from: Hetzen on May 11, 2012, 07:59:32 AM
Sure. It's a dual Zeon Octo-Core which has 16 sandybridge cores (32 threads) running 3.1Ghz and a turbo frequency of 3.8Ghz. It has 64GB Ram and a Fire Pro 9800 4GB RAM graphics card. It has 1 SSD OS drive and 4X SSDs raided to achieve 2GB per second transfer rate. *drool*
I've only had a quick play with it and got a benchmark of 6.40 minutes, with what looked like only 16 buckets. I need to investigate further why I haven't got all 32 cores rendering, and unfortunately I won't be able to look at that until next Wednesday.
Funnily enough it's already been christened "The Beast".
ENVY !!!!
Ok , i feel better now ;D
Hi,
I'm not sure that RAM speed would make a noticeable difference to rendering. Those benchmarks are probably really well optimised for memory access and are probably performing specific tasks. TG2 on the other hand is not terribly well optimised for memory access and it's doing all sorts of stuff at once. There are bound to be other inefficiences which would outweigh the increase in memory speed. At this stage I would say you're probably better off having more RAM than faster RAM.
Regards,
Jo
Quote from: jo on May 11, 2012, 10:10:56 AMHi,
I'm not sure that RAM speed would make a noticeable difference to rendering. Those benchmarks are probably really well optimised for memory access and are probably performing specific tasks. TG2 on the other hand is not terribly well optimised for memory access and it's doing all sorts of stuff at once. There are bound to be other inefficiences which would outweigh the increase in memory speed. At this stage I would say you're probably better off having more RAM than faster RAM.
Regards,
Jo
Many thanks for the headsup !! :)
Quote from: Hetzen on May 11, 2012, 07:59:32 AMSure. It's a dual Zeon Octo-Core which has 16 sandybridge cores (32 threads) running 3.1Ghz and a turbo frequency of 3.8Ghz. It has 64GB Ram and a Fire Pro 9800 4GB RAM graphics card. It has 1 SSD OS drive and 4X SSDs raided to achieve 2GB per second transfer rate. *drool*
I've only had a quick play with it and got a benchmark of 6.40 minutes, with what looked like only 16 buckets. I need to investigate further why I haven't got all 32 cores rendering, and unfortunately I won't be able to look at that until next Wednesday.
Looks like you're not the only one who can't wait until next Wednesday :)
When you can get all 32 buckets to run, be sure to drop us a line on the speed of that "beast" you got there
4:30. With a bucket size of 64.
I'd had maximum threads set at 16 for the 6:40 time.
Quote from: Hetzen on May 17, 2012, 05:47:31 AM
4:30. With a bucket size of 64.
I'd had maximum threads set at 16 for the 6:40 time.
As of TG 2.3, the 64-bit builds of TG have the default max-threads set to 64 and subdiv cache size varies depending on number of cores detected. If you're using a custom default project, you might want to remove the thread and subdiv cache size parameters from the XML with a text editor, so that TG inserts its own defaults when it loads.
Oh, I see that you're using the benchmark scene. In that case the benchmark needs to be edited!
Matt