first double-CPU build

Moflo

n00b
Joined
Mar 10, 2014
Messages
27
This machine is going to be used for rendering CGI and post-production. There are some small details I'm not too sure so a little advice would be great.

Mobo: ASUS Z10PE-D8 WS
CPU: Xeon E5-2630 v3 2,4 GHz
Memory: Kingston ValueRam DDR4
PSU: Seasonic X-1250
HSF: NOCTUA NH-12DX i4

GPUs are already covered from my old build, as well as the Samsung 840Pro.

I would like to know if I need additional cables for connecting the PSU with the mobo, since there are 2 CPU power connectors (8-pin EPS 12v) Or would someone suggest a different PSU altogether?

Any suggestions on a case, can't really find on that explicitly says EEB the only one to specify is the Cooler Master Cosmos 2. I already have the Cosmos S but looking at the specs of the ASUS mobo I doubt the holes are going to align. Although EEB standards are the same than EATX but there are secondary placements that don't align.
Puget Systems use a Fractal Design Define XL R2 for the same build but again EEB not mentioned.

The only thing left are some HHDs for a RAID1 setup to store work. Does not have to be super fast or fancy. I foud some TOSHIBA 4TB MD04ACA400 cheaper than WD or HGST.

Any advice is appreciated
 
This machine is going to be used for rendering CGI and post-production. There are some small details I'm not too sure so a little advice would be great.
Mobo: ASUS Z10PE-D8 WS
CPU: Xeon E5-2630 v3 2,4 GHz

Let me play a role of a Nasty :D and question about basics of this design. Why using dual processor mobo and two processors of low basic frequency? One can use single processro mobo with 8-core single socket Xeon or i7-5960X and overclock it to get the same or even much better performance. For example, I have Asus X99-E WS mobo with Xeon E5-1680 v3 which overclocks without any problems to 4.3 GHz and I have 128GB ECC RAM in it. Simple calculation shows that my 8 cores @ 3.6 GHz is roughly equivalent to 2x6 cores of E5-2630 @ 2.4 GHz and it will leave it in dust at 4.3 GHz.

The dual Xeon processors with low frequency are made for data center servers where the number of threads is more important than speed. For your types of loads it is better to have a combination of bit less threads but much higher speed. Dual pocessor mobos are useful for workstation type of loads if it is clear that no single processors can do tasks and/or there is a need for more than 128 GB of RAM.
 
Sorry wirk, but the E5-2630 is 8 core hypertreding just like your E5-1680. So if you make your calculations I'll have 16 cores and you just have 8.

Although it is preferable for most applications to have a higher clock than the number of cores, this is not exactly true for a render engine. Right now I'm also taking a look at the 2660 10 core.
Btw, I don't want to overclock because I have done that and now I'm building a reliable Workstation to render high end images for print, which takes a few hours
 
True, I looked at the E5-2630 but not v3, the v3 has eight cores. Overclocking ot the E5-1680 v3 to 4.2-3 GHz is no problem at all. But is this really so that your applications are more dependent on the number of cores?
 
In my experience well optimized software that can utilize all the available threads and can give huge performance improvements over higher clock speeds.

This is not an area that I do much work in professionally however work loads that can be broken down into may parts and executed in parallel can use many cores very successfully and can yield huge speed improvements.

I see the benefits of a multi-core/multi CPU system when using software such as handbrake for video encoding and adobe premiere.

I suggest that you look online for some benchmarking results for the software that you plan on using to confirm how well it can utilize multiple cpu's

When you are looking around be aware that a multi-core cpu is different compared to a multi cpu system.
multi cpu system use a technology called SMP (symmetric multi processing) I think it stands for and software must be written to take advantage of this other wise the 2nd cpu is wasted. This limitation does not exist for multi core cpu's. This is because in a multi-core CPU the CPU itself handles the distribution of jobs over the cores in hardware independently of the operating system or software. Where as a multi cpu system the OS and software schedule the tasks over the CPU's. This means that unless the software is SMP capable you will see no advantage to a 2nd processor.
 
In my experience well optimized software that can utilize all the available threads and can give huge performance improvements over higher clock speeds. ....
This means that unless the software is SMP capable you will see no advantage to a 2nd processor.

It is obvious that well optimized multithtreading/SMP software will be always better. The problem is there is not much of such software. This is due to the fact that developers mostly produce software for the lowest common denominator hardware used by people. Typically one sees multicore/SMP as big advantage in servers/data centers where lots of threads are good since they run huge number of different tasks.

Let's take simple practical example of a PC with graphics cards in SLI. Adding second card results in almost doubling the graphics performance. Adding third and fourth card brings very little. Why it is so? The current model of graphics operation is such that graphics cards are served by single core in single thread which is nonsense for the SLI. Only the soon coming DirectX 12 will solve this problem by allowing multicore multithreaded model. This can be bring huge performance gains - if developers will use the tools they will get. But anyway, this will be a huge progress since until now they could not do anything to improve the SLI even if they wanted.

Here is another example in the direction of apps mentioned in this thread: Adobe Photoshop illustrating my point. Adobe is generally considered as one of the best in multithreading/SMP but results are mixed at best. One can see that performance increases up to 4 cores and gets flat, dual procesor test is just laughable since performance drops in many cases(!). Why? They optimize for the common 4 core single processors and do not bother about more since it is a niche market. I suppose that fully optimized software can be found among those which cost tons of money for ultrahigh professional segment.
 
Last edited:
I have this mobo and you should be aware that the ASUS Z10PE-D8 WS, has a combination of current and legacy mounts, specifically, H' Y' and J', so even mobo trays that are EBB might not support the mobo.

You definitely need a PSU that provides 2 CPU eps cables. That is why I ended up getting the Corsair AX1500. Overkill but oh well.
 
This article's a bit old, but still spot on.

http://www.fxguide.com/featured/the-state-of-rendering-part-2/

Again, while having gobs of threads is great, the fact is, many rendering packages simply don't take advantage of all of them.

Most deliver the best performance improvement with 4-8 threads and nothing after that. Seriously, you could get a bunch of 14-core CPUs, and 20-24 of them will effectively sit idle.

You really are better off with 4-8 physical cores running at the highest possible speed.

http://www.newegg.com/Product/Product.aspx?Item=N82E16819117499

A single 1650 v3 will wind up delivering better overall performance than the pair of 2630's.
 
This article's a bit old, but still spot on.

http://www.fxguide.com/featured/the-state-of-rendering-part-2/

Again, while having gobs of threads is great, the fact is, many rendering packages simply don't take advantage of all of them.

Most deliver the best performance improvement with 4-8 threads and nothing after that. Seriously, you could get a bunch of 14-core CPUs, and 20-24 of them will effectively sit idle.

You really are better off with 4-8 physical cores running at the highest possible speed.

http://www.newegg.com/Product/Product.aspx?Item=N82E16819117499

A single 1650 v3 will wind up delivering better overall performance than the pair of 2630's.

My workstation has plenty of cores, and I can confirm that 95%+-5% of scenarios, single application performance does not scale at all after 12 cores, and even 8 cores.

Unless you are running bunch of VMs it really is much better to have GHz.
 
My workstation has plenty of cores, and I can confirm that 95%+-5% of scenarios, single application performance does not scale at all after 12 cores, and even 8 cores.

Just wondering, is there any example of software which scales up well with the number of threads/cores in any reasonable amounts?
 
Basically unless you're running virtual machines or folding/cracking clients (Distributed.net, etc), after a certain point, the extra cores net you bupkiss.
 
basically except for soft image and light wave all of the professional render software can make use of multicore / hyper threading if you use the 2014 or later version. Renderman just came out with a new version that is actually faster on virtual cores than multiple cpus. The fastest setup currently is a blade server followed by dual i7 hex or octo core with a nvidia telsa setup. From there it breaks down to how fast you can feed the cores from ram. Most systems are not optimal at that. So 128GB of ram, an i7 hex core or octo core and a tesla or titan to offload the simple calculations make sure you get the old double precision titan if you go that route. Speed is much better when you are rendering as often times the later pieces need info from the early calcuations unless you makes you buckets like 5k each or a grid of a thousand pieces but then it can not reuse the lighting math and will look funny.

https://www.spec.org/gwpg/downloadindex.html
 
basically except for soft image and light wave all of the professional render software can make use of multicore / hyper threading if you use the 2014 or later version. Renderman just came out with a new version that is actually faster on virtual cores than multiple cpus. The fastest setup currently is a blade server followed by dual i7 hex or octo core with a nvidia telsa setup. From there it breaks down to how fast you can feed the cores from ram. Most systems are not optimal at that. So 128GB of ram, an i7 hex core or octo core and a tesla or titan to offload the simple calculations make sure you get the old double precision titan if you go that route. Speed is much better when you are rendering as often times the later pieces need info from the early calcuations unless you makes you buckets like 5k each or a grid of a thousand pieces but then it can not reuse the lighting math and will look funny.

https://www.spec.org/gwpg/downloadindex.html

Do you have specific render software that can make use of 30+ cores?
 
basically except for soft image and light wave all of the professional render software can make use of multicore / hyper threading if you use the 2014 or later version. Renderman just came out with a new version that is actually faster on virtual cores than multiple cpus. The fastest setup currently is a blade server followed by dual i7 hex or octo core with a nvidia telsa setup. From there it breaks down to how fast you can feed the cores from ram. Most systems are not optimal at that. So 128GB of ram, an i7 hex core or octo core and a tesla or titan to offload the simple calculations make sure you get the old double precision titan if you go that route. Speed is much better when you are rendering as often times the later pieces need info from the early calcuations unless you makes you buckets like 5k each or a grid of a thousand pieces but then it can not reuse the lighting math and will look funny.
https://www.spec.org/gwpg/downloadindex.html
Do you have specific render software that can make use of 30+ cores?

I made quick look and found that there is now available free noncommercial version of Renderman. Renderman could be then an extremely good test of performance of eight core machines plus graphics cards (unfortunately I have Titan X which does not fit as I see from above). Is there any demo of Renderman which could be used for this? How about using Renderman without graphics cards to test the CPU/multithreaded/multicore performance?

Nevertheless the question if Renderman scales well beyond 8 cores is still valid.
 
This machine is going to be used for rendering CGI and post-production. There are some small details I'm not too sure so a little advice would be great.

Mobo: ASUS Z10PE-D8 WS
CPU: Xeon E5-2630 v3 2,4 GHz
Memory: Kingston ValueRam DDR4
PSU: Seasonic X-1250
HSF: NOCTUA NH-12DX i4

GPUs are already covered from my old build, as well as the Samsung 840Pro.

I would like to know if I need additional cables for connecting the PSU with the mobo, since there are 2 CPU power connectors (8-pin EPS 12v) Or would someone suggest a different PSU altogether?

Any suggestions on a case, can't really find on that explicitly says EEB the only one to specify is the Cooler Master Cosmos 2. I already have the Cosmos S but looking at the specs of the ASUS mobo I doubt the holes are going to align. Although EEB standards are the same than EATX but there are secondary placements that don't align.
Puget Systems use a Fractal Design Define XL R2 for the same build but again EEB not mentioned.

The only thing left are some HHDs for a RAID1 setup to store work. Does not have to be super fast or fancy. I foud some TOSHIBA 4TB MD04ACA400 cheaper than WD or HGST.

Any advice is appreciated

Depending on the gpu's you are reusing I would say that that psu might be overkill, you would probably get away with a 850-900watt version of the above, or a corsair ax860. They will still have 2 psu connectors for the cpu's. I have your mobo's older cousin, the z9pe-D8 WS and I run it 24/7 with a pair of e5 2665 v1's on an ax750 no problem - I do not have any gpu's to speak on it though
 
Back
Top