Dual Quad Core v i7

[Jock]

So I'm looking to build a second computer which would be used as a backup and also a render farm. Though please bear with me as i'm not overly clued up on the technical aspects.

 

The current computer I have has two quad core cpu's in it but I was wondering if its possible to have a rethink about the new one because the last one needed some quite expensive components like the ram that was needed, the psu and the mb itself.

 

So i'm wondering if I was to get an i7 which I beleive has 4 actual cores and 4 virtual cores, would that render just as quick as the dual quads if say we assumed they had the same processor speed?

 

I'm trying to ascertain which one would be used as the main computer and which as the backup.

 

Also, if you were to build a computer as a render node, does that mean you could just put a very basics gfx card into it, with the one good card in the primary computer? Or would the slave also need to have a decent card for some reason.

 

Thanks for your help.

[DavidR]

A dual-socket MB destined for server use often has integrated graphics, or you could just get the cheapest card available, the cheapest HD (since you're going to store data elsewhere?), and spend on the MB, CPU(s), and RAM. You need Xeons for a 2-socket MB, and you may want 6-core. AJ Lynn has a good column on this site that will give you a bunch of options suited to your budget, etc.

[MegaPixel]

Typically, Render nodes do not require fancy graphics cards as the rendering is primarily affected by CPU and Memory - Onboard GFX in this case would be fine. If you plan to utilize the new GPU based rendering technology, then you would need to carefully consider your Graphics Card (or Card's...)

 

Also, I've been working on a Dual Quad Xeon Dell XPS for the past 3 years as my primary workstation. I used to think that was fast, until my Boss recently purchased an HP system with a single i7 in it (Half the price of what we paid for the XPS back when..) and it performs roughly twice as fast.

[Macer]

Virtual cores are not equal to physical cores, a new dual quad will have 8 pys cores and 8 virt cores. ( unless you go for the 6 core cpus, then it would have 12 + 12.

 

Render nodes don't need good GPUs unless you are considering GPU rendering (which is starting to become mainstream), in which case you could get a much lower spec cpu and multiple GPUs.

 

AJLynn's blog covers a lot of it: http://www.3datstech.com/

[Jock]

Hmm thats interesting megapixel...single i7 might be the way forward then.

 

As for gfx card, i'd still have a decent one in the primary system, its just the render nodes thats in obeyance but looks like on board will do the trick.

 

Thanks for your help guys.

[AJLynn]

If the dual-quad Xeon is one of the newer ones that uses the same tech as the i7, it is equal to two i7's of the same clock speed.

 

What you need to do when comparing estimated render speeds is to determine the family of CPU (i7, Core 2, etc.) and the total GHz across all cores (don't count a core with hyperthreading as two cores - a quad i7 has 4 cores, not 8, so a quad i7 at 2.5GHz has 10GHz total).

 

If you are comparing two machines of the same family, you're almost done. Just divide the GHz of the faster one by the GHz of the slower one to know proportionally how much faster it is. (A computer with 24GHz total is 2.4x as fast AT RENDERING WITH AN EFFICIENT MULTITHREADED RENDERER as a computer with 10GHz total, IF THE CPUS COME FROM THE SAME FAMILY.)

 

If comparing a Core2 generation PC to an i7 generation PC (and we're including the Xeons of the same generation in this) you need a multiplier to account for the i7 being faster (because it's more efficient, and it has hyperthreading, and other tricks). What I did for this, and my friend who's a theoretical mathematician hates it because it's empirical but I'm an architect and I think empirical data is the best kind, was make a huge spreadsheet of Cinebench 11.5 results from many CPUs, normalize them and determine a number that expresses how fast cores are, relative to each other, MHz for MHz. Essentially, render speed per CPU MHz, which I normalized so that the Core 2 Duo Macbook Pro I had when I started the spreadsheet scored a 1.00. I then averaged the scores for everything I had in each family - which worked well because they're all in a few percent of each other.

 

In this system, the multiplier for a desktop Core 2 is 1.24 and the multiplier for a Core i7 is 1.90. So each GHz of Core 2 desktop chip is worth 1.24 GHz of Core Duo laptop chip, and each GHz of i7 desktop chip is worth 1.90GHz of Core Duo laptop chip.

 

By dividing 1.90/1.24 we find that each GHz of i7 chip is worth 1.53GHz of Core 2 chip. If you compared, straight up, a 3.0GHz Core 2 Quad to a 3.0GHz quad i7, the i7 is 53% faster for rendering.

 

This is pretty cool. Now with a bit more math we can compare, say, a dual-quad 3.0GHz Xeon box from a couple of years ago before i7 generation Xeons were introduced, to a 3.0GHz singe quad i7 desktop by using the 1.90 and 1.24 multipliers:

 

i7: 4x3.0GHz = 12GHz total, 12GHz*1.90 = 22.80 points

Xeon: 8x3.0GHz = 24GHz total, 24GHz*1.24 = 29.78 points

 

29.76/22.80 = 1.30, so the Xeon is 30% faster for rendering.

[Jock]

Thanks for the description..

 

So lets see if i can get this right.

 

My current sytem is 2 x Intel Xeon E5420A 2.5GHz

 

so 8x2.5 = 20GHz, 20x1.24 = 24.8 ponts

 

i7: as per your example if i went for 3Ghz = 22.8

 

So although the dual xeons are faster, its not by much, and yet if i went for i7 the ram would be cheaper the psu would be cheaper and the mobo would be cheaper.

 

Sounds like theres not much choice for what i need from it.

[AJLynn]

Right. So in order to make the Xeon worthwhile, it needs to be 5500 or 5600 series which use the i7 technology.