Ram Selection

[Andrew1]

Just bulding a new rig, mainly designing in 3dsmax and rendering with Vray,

what ram should I use to better over all viewport and processing performance??

I am looking at corsair 2400 Mghz-Cas10 and 1866 Mghz-Cas9, looking at for right now to get 2x8 - 16 gigs for my new 6 core rig.

any suggestions?

 

thanks,

 

Andrew

[notamondayfan]

I've used Corsair in my builds, it's always reliable and never any problems for me. I used these when I upgraded http://www.amazon.co.uk/Corsair-CMZ8GX3M2A1600C9-Vengeance-Performance-Desktop/dp/B004CRSM4I/ref=sr_1_7?ie=UTF8&qid=1381419291&sr=8-7&keywords=corsair+16+gb+ram

 

Viewport and rendering performance will not really be affected by ram, CPU and GPU are the key here.

 

Dean

[Dimitris Tolios]

I agree with Dean...Intel chips won't get meaningful performance gains past 1866, and fast than 2133 you will get nearly nothing.

Those are "next gen" speeds maybe, or simply aimed towards a "got to have it" crowd that is gearing towards showing off benchmark results that might be dealing with tenths of 1% differences.

 

Some compute tasks do get better with faster memory (ofc timings are important there, and maybe a 2133 CAS 9 will be better for many stuff than a 2400 CAS 11 etc), but that's neither viewport or general CPU rendering with VRay / Metal Ray.

For example I have seen it when Folding for F@H (very very small RAM utilization by comparison - maybe 100MB - but very complicated calcs), and I have seen it with GPU Rendering, where overclocking a fast card's VRAM does yield decent speed gains.

[Andrew1]

thanks for the reply,

 

I am thinking of getting the 1866 but the 2400 is for about 25 dollars more but the latency is better in the 1866 so my question is is it worth it

to go to the 2400 spending the extra couple bucks or should I just get the 1866 with quicker latency and be over it...,

I am buying all my parts today after work, I want to eventually go to at least 32 gigs of ram or perhaps more I just want to know

how it is going to affect the performance of my new computer , I also do video editing and large format printing using photoshop and dealing with images that are at times 10ft x 40 ft on the screen this is the reason I am asking as ram is quite important dealing with large files...

[Dimitris Tolios]

Well...the faster the mem the "better" in general...just not by much to justify the cost increase.

$25 is little difference for a s2011 system, so, just go for 2400, so that you don't "wonder" what you would be missing...in reality you won't be missing much. Maybe nothing.

 

According to Anandtech's results (and others, before it) 2400 CAS 10 will be in the region of 2-3% faster than 1866 CAS 9 (he was using Haswell i7).

 

These figures are for CPU compute and "closer" to a real world scenario, not memory benching where ofc the 2400 will dominate more profoundly.

 

In general - as wrote above - easy tasks on the CPU side, make it rip through them fast and keep asking more from the RAM, so faster RAM makes a difference. The more complicated the tasks (both for compute and viewport), the less important the RAM speed is: "what if I need 10ns more to get a packet, when I will be spending whole seconds on it"?

 

With more complex tasks, the time each "packet" spends in transit from RAM to the CPU becomes irrelevant, as it is disproportionally smaller than the time it spends going through the CPU.

Returns are diminished once you reach the "fast enough" mark, which for the last 3-4 generations of intel CPUs is DDR3 1600. 1866 has become as cheap as 1600, so it is a no-brainer...just get it...as 2133 and faster kits will start dropping, ofc we can pick them up with little remorse, but that doesn't make them a "must have".

 

Same goes for heavy viewports: the CPU is "preparing" frames for the GPU to render and display. For low resolutions and small/simple models, the CPU is the bottleneck, as the GPU is rendering the requested frames faster than the CPU can provide them (that's in both CG viewports and games).

Start adding large textures, complex geometry and multiple light sources = complex shadows, and the GPU becomes the bottleneck. The CPU doesn't do much more work than it was doing before "reporting" the frame contents to the GPU, so both the CPU and the main RAM end up "waiting" for the GPU, and absolute performance doesn't really decrease once you've reached decent IGP from the single thread that prepares the frame for the GPU - usually that means that as far as viewport goes, an i3/i5/i7 - regardless of cores/threads - will give you similar results in high-resolution and complex viewports, while the fastest clocked chip will give you proportionally better results in low-resolution, simple viewports.

 

So there is this "funny" situation, where the benefits from faster RAM are noticeable only where performance is already above "perceivable" levels (i.e. having 140 over 90 FPS in a simple viewport is undetectable by our eyes). Ofc there are benches, and video transcoding tasks and unzipping large files and other auxiliary tasks which will get "better" (some more than others), but in general rendering tasks and viewport acceleration in high resolutions won't really benefit.

Edited October 10, 2013 by dtolios