Contribute
Register

Need advice on building Dual Xeon

Status
Not open for further replies.
Joined
Feb 27, 2011
Messages
21
Motherboard
<< need manufacturer name and model # >> See Forum Rules !!
CPU
i7 2600K
Graphics
GT640, GT520
Mac
  1. iMac
  2. MacBook Pro
  3. Mac mini
Mobile Phone
  1. iOS
Need advice on building Dual Xeon or IB custo for audio production

Hello Tonymac friends.

It's been 2 years since my last build, and I believe it is time for another one. I've been running Tonymac Customacs in my recording studio for about 4 years. I kept my G5 sitting in the corner for safety for the first year or so, but when it became obvious how stable these machines can be, I ditched it. I used to run Pro Tools HD with 2 HD cards, but now I simply run 2 different native systems on custo macs. I also have an imac and a mac mini laying around in case things get real hairy.

Anyway, my current machine which is a Z68 2600K just isn't cutting it when it comes to some of my surround mixes. I do a lot of post, and have recently had 7.1 mixes as well. When running QT video out of my avid Mojo, things start slow down considerably, and I run out of processing power.

I'm thinking of going the Dual Xenon route. I have been reading all of the threads on these builds from the last 2 years, including the ones by Braindeadmac, Warelike, and of course Pinknugget.

So my question is: Is an SR-2 based 1366 machine still the way to go at this point? Would I be best going with a 2011 based machine? I'm looking at a 2-3 year commitment for the machine, and I need to significantly upgrade my processing power now. My current machine only scores a 11,800 on geek bench, so please let me know if I'm barking up the wrong tree, and need simply to upgrade my current custo. Spending 12K on the new Pro Tools HDX is not an option.

Thanks.
 
Hello Tonymac friends.

It's been 2 years since my last build, and I believe it is time for another one. I've been running Tonymac Customacs in my recording studio for about 4 years. I kept my G5 sitting in the corner for safety for the first year or so, but when it became obvious how stable these machines can be, I ditched it. I used to run Pro Tools HD with 2 HD cards, but now I simply run 2 different native systems on custo macs. I also have an imac and a mac mini laying around in case things get real hairy.

Anyway, my current machine which is a Z68 2600K just isn't cutting it when it comes to some of my surround mixes. I do a lot of post, and have recently had 7.1 mixes as well. When running QT video out of my avid Mojo, things start slow down considerably, and I run out of processing power.

I'm thinking of going the Dual Xenon route. I have been reading all of the threads on these builds from the last 2 years, including the ones by Braindeadmac, Warelike, and of course Pinknugget.

So my question is: Is an SR-2 based 1366 machine still the way to go at this point? Would I be best going with a 2011 based machine? I'm looking at a 2-3 year commitment for the machine, and I need to significantly upgrade my processing power now. My current machine only scores a 11,800 on geek bench, so please let me know if I'm barking up the wrong tree, and need simply to upgrade my current custo. Spending 12K on the new Pro Tools HDX is not an option.

Thanks.

There are several successful 2011 single socket builds in User Builds and even a couple in Golden Builds you might want to take a look at. If you go with dual Xeons, I hope you have a huge bank balance :lol:, 'cause them suckers ain't cheap!
 
There are several successful 2011 single socket builds in User Builds and even a couple in Golden Builds you might want to take a look at. If you go with dual Xeons, I hope you have a huge bank balance :lol:, 'cause them suckers ain't cheap!

Yeah, I've been pondering simply upgrading my mobo and processor to one of the new 6 core 2011 chips. However in Pro Tools you can pick the amount of processors you want to use for plug-ins, and the more the better.

I've seen matched pairs of Xeon 5650 chips on the flea bay for under a grand. However, would a new 6 core get me close to the processing of a newer 6 core machine?
 
You might want to wait until Apple releases a new Mac Pro, if they do at all. Reason being is that if they do, the market will be flooded with lots of used Mac Pro's and I imagine you could pick up a dual processor 2010 Mac Pro for a nice price. I am not generally a fan of investing in old technology, but I don't see a lot of dual CPU hackintosh's around and with dual six core CPU's you'd have 24 cores, plenty to play with. The SR2 socket 1366 are getting a little dated themselves anyway, just a thought.
 
It's a complicated question. It depends on how multithreaded your particular apps are (I'm not familiar with them), whether they can take advantage of SB-E (lga2011) improvements, how much you're willing to pay for small incremental improvements, and how much you like to gamble.

Most current commercial apps that I'm familiar with do benefit significantly from SB-E improvements, such that mid-level SB-E's are as fast as extremely overclocked Westmere's. And if they use AVX, or SB-E SSE extensions, SB-E will be significantly faster.

lga1366 (in SR-2) is end of life. There is no future upgrade path. lga2011 has at least 18months (Ivy bridge update to SB-E will use lg2011), and while Intel has not commented, there's reason to think the next version, a haswell update, will also use lga2011 (it's speculated that the next xeon socket change will occur when the switch to pci4.0 occurs, which will largely be driven by adaptation of 100GbE).

The gambling part has to do with OSX and SB-E. If there is another Mac Pro, and it uses lga2011, then a 2P lga2011 hackintosh becomes much more attractive. Currently there is no OS power management. For Xeons that means no turbo, and no idle. Server chipsets (even if they are in consumer boards, such as the Asus Z9PE) rely on OS signals, so if the OS can't send them, the only thing you can do is set a static speed. So, until there's a lga2011 based Mac Pro (and there's no guarantee there ever will be), 2P xeons are significantly handicapped (limited to base speed, and unable to drop speed/voltage when idle...so more power draw).

This is not the case for X79 based boards (consumer 1p lga2011). You still can't get speedstep, but you can get max speed (and overclock) and idle (idle steps are handled in hardware). There's not a lot of practical difference between this and full power management (all those extra steps don't really matter much anymore given the speed that current processors can change bins).

There's a good chance that an overclocked (a nice, easy, safe overclock) 3930k would be as fast or faster than a pair of mid-range E5-26xx's (since the 3930 could have 6 cores at 4.2 say, vs. say E5-2660's having 8 cores each, but only running at 2.2) unless your applications significantly benefit from more than say 8 cores (which is a good approximation of a 6c/HT cpu).

So my take on your options are...

Get a used sr-2. It will be end of life, and to get comparable performance you need to be comfortable with a fairly significant overclock. It will be cheaper than a 2P lga2011.

Go with a X79 based system. Cost would probably be similar to the sr-2, maybe a little more depending on how good a deal you can get on a sr-2 combo. While not supported in OSX, performance and power draw will be essentially the same as if it were. You won't have sleep though. Compared to the sr-2, will use a lot less power, and depending on how well threaded your apps are, and how well they take advantage of SB-E enhancements, could be a little bit slower to a little faster.

Go with a 2P lga2011. Will cost the most. If your applications are well coded for >16 threads, will likely be significantly faster than the x79 system, even if it is never natively supported in OSX. Power draw will be significantly higher than the X79 system, compared to the sr-2 it will depend on your load ratio. You also won't have sleep. If A new Mac Pro is released with lga2011s, this system has the potential to be significantly faster than the X79 (would still require well-threaded apps, but less so as having native turbo would narrow the Mhz advantage of the X79).

If I were you, I'd try to get a very concrete answer to how many threads your particular apps can effectively utilize (in particular where dimishing returns really hits...i.e. it may be able to 'use' 32 threads, but if 32 thread performance is 5% better than 8 thread...). I'd also try to get some real numbers of how your apps run on Westmere vs SB-E (doesn't have to be xeons, can be i7-970 vs 3930).

Note that GeekBench is not a good measure of how these will stack up. You need some real numbers from your apps. As an example, my highly overclocked sr-2 x5680 set up had a slightly higher GB (linux) than my 2P E5-2680, but in the applications I use, the lga2011 set up is 2.5-3x as fast.
 
You might want to wait until Apple releases a new Mac Pro, if they do at all.

I believe them that they will. I was hoping it was going to be early this year, but now I'm convinced it will late in the year, if this year at all. I have a huge mix coming up in the spring, and I'd really like more power for it. Believe me, I've wanted a new Mac Pro for a while. Amazing that a company refuse to take my money, but I guess that's why we are here.

I've been reading up on how to properly set my host processor settings in PT. When I watch my core usage in iStat, I NEVER see the virtual cores doing anything. I'm realizing now that setting Pro Tools to using one less core, not one less virtual core provides better stability, and more plugs. Also, from research I've done it appears that synths and samples benefit most from FSB speed and higher CPU clock, and plug-ins benefit more from more cores.

I need plug-ins. Lot's o plug ins baby.

The SR2 socket 1366 are getting a little dated themselves anyway, just a thought.

Agreed, which is why I'm curious if there are other options for me that might get me above 20K geek bench. That would be double what I've got now, which is what I would like to do if I'm going to invest (above 30K would be nicer). Stability and help resources are also key, i.e. having a solid base of other builders to be able to lean on.
 
If a new Mac Pro comes out it looks like it my use either a socket 2011 or 1356 based on the newest Xeon's that are out now, my gut tells me it will most likely be 1356, but that's just me, I have nothing to base that assumption on. If you have the money saved, then wait, I expect that if they are going to release a new Mac Pro that it will be announced in June along with the new iPhone and whatever else is going to get updated this year. It will likely either be ready the day its announced, or a few months later. I expect it to be priced in line with what they have now with entry level single processor models being around or just under the two thousand dollar mark and the full package dual processor monsters coming in at over four and half thousand dollars. I also think they'll be dropping AMD and going with Nvidia GPU's along with new additions of thunderbolt, USB3 and maybe even a case redesign.
 
It's a complicated question. It depends on how multithreaded your particular apps are (I'm not familiar with them), whether they can take advantage of SB-E (lga2011) improvements, how much you're willing to pay for small incremental improvements, and how much you like to gamble.

Right, this cuts to the core of questions, and I've been pondering/researching them for a while. To complicate things, Pro Tools 11 is right around the corner, and will FINALLY be a 64 bit program. I'm convinced that PT being a 32bit program (with admittedly old code) hand cuffs the efficiency of the program. However, PT11 is due to arrive in the next 6 moths, and could be out as early as next week (winter NAMM conference).

If I were you, I'd try to get a very concrete answer to how many threads your particular apps can effectively utilize (in particular where dimishing returns really hits...i.e. it may be able to 'use' 32 threads, but if 32 thread performance is 5% better than 8 thread...). I'd also try to get some real numbers of how your apps run on Westmere vs SB-E (doesn't have to be xeons, can be i7-970 vs 3930).

Trying to get concrete answers about this out of Avid hasn't been easy.

Basically, the best systems seem to be machines that can dedicate one entire processor to Pro Tools, and one to everything else. There is a dialog box in Pro Tools where you set this. Alternatively, many users report the best performance when allocating half of the cores to PT. I myself have witnessed large sessions where I see 4 of the 8 virtual cores having heavy usage, and the other 4 not doing anything. This is with Pro Tools set to use 7 of the 8 cores. It's the same when I tell it to use 3 of the 8. I take this to mean that it does't utilize hyper threading very well (at all)?

So, from real world experiences, the happiest users are the ones with dual processor machines. On the other hand, that's a lot of cost for not much improvement when you are talking about an entire processor for taking care of the OS and email while PT is running.

Thanks for your knowledgable and detailed response!
 
Getting back to my point about other users experiences. I have NOT found that allocating Pro Tools to use half of the cores gets the best performance. Maybe these users have dual processors. I'll have to investigate.

If I make a session where I max it out I can get the system to play back at 7 of the 8 cores allocated, and it chokes using one less core. This says to me that does use hyper threading. When I look at activity monitor, it appears as if all 8 of them are being used pretty heavily, regardless of what I have Pro Tools set to. This is puzzling.

BTW, I need to get my system to run cooler. CPU temps were spiking at 96C.
 

Attachments

  • Screen Shot 2013-01-18 at 10.06.09 PM (2).png
    Screen Shot 2013-01-18 at 10.06.09 PM (2).png
    22.7 KB · Views: 203
Based on that graph, PT is not very multi-thread friendly. By setting the core count, your forcing the scheduler to include that many (logical) cores. The fact that your averaging ~60% utilization means it's really just using 4 threads, that are being spread among the specified cores. My guess is if you had 12 cores, utilization would go down to 35-40%.

So, unless the update involves a major recode (which 64bit doesn't necessarily imply), I don't think you'll see much improvement in PT itself by having more than 6 REAL cores. I can't say anything about the plug ins, it depends on what they actually do. In it's current iteration, I'd suspect running PT with cores set to 4 (and a utilization of 100%) would perform as fast as any other arrangement. Assuming the plugins are not computationally intensive, there should be a significant speed up going from 4 to 6 real cores, as ancillary processes (plug ins, OS, IO, etc) wouldn't have to share resources with the PT threads (since you'll have 2 real cores available to handle them, not impinging on the PT thread pipeline flow).

I'd also guess that SB-E is probably significantly faster than Westmere for PT. It could be other things, but the utilization graph looks like there are a lot of redirections. Since two of the biggest improvements in SB-E over Westmere are cache architecture and memory bandwidth, you might see significant improvement...but I'd imagine someone has compared them.
 
Status
Not open for further replies.
Back
Top