[balt]

Hi all,

I'm looking into investing a few grand into a new workhorse machine. I know what I need disk space wise... apart from that it's pretty open. I've been out of the hardware loop for a few years, and the CPU's out there these days are confusing! So, can anyone help me whether I should go down the server architecture CPU road or the workstation road? More specifically:

- How does a Xeon W5590 compare in performance to the i7-980X?
- How does the new 12-core AMD in dual setup (24 cores) compare to the above?
- Do I really need a Quadro FX5800 costing 4 grand for 4GB of video RAM? What other, less expensive but still well performing graphics cards are there?

I'd love to hear from some people out there that are using Maxwell on those processor architectures! I've recently bought a cheap Dell workstation with an i7-860 and 4GB of RAM for less than 700 euros, and that thing is bleeding fast compared to my old Q6600 and core 2 Duo machines... how much better are the high end machines still?

Cheers

- Balt

[Bubbaloo]

I just ordered a custom workstation with the new 980X. For rendering, invest in the processor and RAM. If you are planning to use the workstation to model also, invest in a quality card. Gaming cards can work well, but the Quadro cards really do well in modeling apps. I have used an expensive Quadro 4500 and a less expensive GTX 275, and there's a good performance difference. But any of the newer cards are going to do well.

Of course the Xeons are going to perform better, but you will PAY.

[balt]

Thanks, most informative! The Xeon's are not really much more expensive. It would be interesting to see the same graph with the Xeon CPU's added (as well as the Opteron's on the AMD side, although AMD is a bit back on the performance curve as I understand).

Cheers

- Balt

[dmeyer]

Since you are asking this thread on this forum I will presume your primary app is Maxwell.

Here are some benchwells from our machines.



Maxwell does very well with Hyperthreading. I haven't seen any benches yet but I'd guess the 980X around 1800-1900 benchwell. For Dual Gainstowns basically double it to ~3600 or so.

Considering that the Magny-Cours chips from AMD are nearly identical to Istanbul you can imagine them being around 1500 benchwell per 12-core socket @ 2.2 Ghz.

Other times here: http://www.benchwell.com

For the Quadro, whether or not its worth it will very much depend on what your primary 3D application is, workflow, etc. If you are building this, at $4k I might suggest getting a GTX260 first and then spending the $4K if you are not satisfied.

Also - Xeon's are not inherently better than their single-socket Core i7 counterparts. The only real difference is they have dual QPI links so they can be used in dual socket configurations. If you are going single-socket then save some money and go with the i7 series.

[balt]

That's fantastic advice, thank you very much! I know about benchwell, but couldn't find the W5590 listed there, must have mistyped my search!

Thanks again,

- Balt

[polynurb]

hi balt,

as it was mentioned i'd advice to get a quadro if you do use the computer as a workstation.
i am pretty happy with my fx4800.. enough power to drive 30"+21" display.

concerning the renderpower.. for a workstaion a xeon is nice.. because it can hold more ram (many open applications are possible) and you will get very fast material previews. etc.

if it is a pure rendernode, and you don't need more than 12GB ram for what you have to render, i'd go for an i7 920 and o'clock it to ~3,5 ghz.

two of these will still be cheaper and faster than a setup of a dual xeon (look at the price for the 3.33 ghz version!!), even if you need a node license more...

concerning magny cours, i could not yet find maxwell benchmarks for it.. but it starts to get very interesting with that many threads... see here:

http://www.anandtech.com/show/2978/amd- ... ore-xeon/6

on the following page, see the linux result for blender alpha.. magny cour beats them all.. but it has to do with how the os manages data distribution among the cores... i recently read an article that the windows kernel as it is now, is struggling with very high core numbers.. and that it might be modified on very low level sooner or later.