It is currently Fri Mar 29, 2024 2:56 am


Cores v. Speed

Buying or building computers for Hauptwerk, recommendations, troubleshooting computer hardware issues.
  • Author
  • Message
Offline

sutherland

Member

  • Posts: 186
  • Joined: Fri Jan 12, 2007 11:50 am
  • Location: Scotland

Cores v. Speed

PostTue Jul 21, 2009 5:09 am

I'm sure I've read things about this on here before, but can't find it now.

Am I correct in thinking that Hauptwerk is optimized to make maximum use of the the number of cores rather than speed, so it's better to have more cores than high speed (e.g. I'd be better with a quad core E5320 1.86GHz than a 5150 2.66GHz)?

I wanted to be sure about this as I was reading a test that showed some software would only use 4 cores even when 8 were present so was faster with the 5150 than E5320 (http://www.techwarelabs.com/reviews/pro ... ndex.shtml).

Thanks
Offline
User avatar

mdyde

Moderator

  • Posts: 15444
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Cores v. Speed

PostTue Jul 21, 2009 6:17 am

Hello Philip,

Hauptwerk is fully optimised for multiple CPUs and multiple cores (at least up to 8 cores, since we've never tried it on computers with more cores than 8).

With all other factors kept constant (clock speed, L2 cache size, bus speed, microarchitecture/revision, etc.), performance doesn't quite scale linearly with number of cores (since there is some overhead in synchronising and load balancing between the cores), but it isn't far off. So given a 4-core CPU and 2-core CPU that were identical in all other regards, the 4-core should outperform the 2-core by slightly less than a factor of two (pehaps 1.8x).

Higher clock rates also give a direct, and usually relatively linear speed increase. E.g. a 2.66 GHz CPU will probably given around 40 percent higher polyphony than a 1.86 GHz CPU with all other factors being the same (number of cores, L2 cache size, bus speed, microarchitecture/revision, etc.).

Looking at Intel's specs for the two processors you mention:

- Intel® Xeon® Processor E5320 (4 cores, 8M L2 cache, 1.86 GHz, 1066 MHz FSB): http://ark.intel.com/Product.aspx?id=28031

- Intel® Xeon® Processor 5150 (2 cores, 4M L2 cache, 2.66 GHz, 1333 MHz FSB): http://ark.intel.com/Product.aspx?id=27218

... the E5320 has four cores compared to the 5150's two, and twice as much L2 cache, but about 30 percent lower clock speed and a slightly lower bus speed than the 5150. I understand that the 5320 is also a later model (microarchitecture) than the 5150, which in itself is likely to give better performance.

So given a choice between just the E5320 and 5150, I would definitely go for the E5320, since I would expect it to give considerably higher performance.

If other models in the E53xx range are an option:

http://ark.intel.com/ProductCollection.aspx?series=30807

... then for future-proofing of those I would probably opt for the E5345:

- Intel® Xeon® Processor E5345 (4-core, 8M L2 cache, 2.33 GHz, 1333 MHz FSB): http://ark.intel.com/Product.aspx?id=28032

... as a good balance between cost and performance. It has a 25 percent higher clock speed (should give approx. 25 higher polyphony) and a faster bus speed for about 42 percent higher price.

However, you might also want to consider some of the more recent processors, e.g. the 54xx series or 55xx series:

http://ark.intel.com/ProductCollection.aspx?series=33905

http://ark.intel.com/ProductCollection.aspx?series=39565

http://en.wikipedia.org/wiki/Intel_Nehalem_(microarchitecture)

E.g. the E5420 (4-core, 12 MB L2 cache, 1333 GHz FSB) seems quite cheap for its spec:

http://www.dabs.com/category/components-and-storage,motherboards-and-processors,processors/11147-50520-40380000

You mentioned in a previous post that you're intending to use the (extremely large) Sonus Paradisi Caen sample set. If you're planning on using the surround version (which uses two voices of polyphony per pipe, i.e. needs about twice the CPU power compared to other comparable sample sets) with large registrations, then you'll need a serious amount of CPU power. From other users of the Caen surround set (I don't have a copy of it myself at the moment, so I can't give accurate benchmarks) 8 cores might even be needed (although of course the non-surround version would be half as demanding, and you can always avoid the largest registrations and thus then get away with less).

If you want something that should give absolutely reliable audio/MIDI performance without compatibility issues, a Mac Pro running OS X might also be something to consider:

http://www.apple.com/uk/macpro/

http://store.apple.com/uk_smb_67752/browse/home/shop_mac/family/mac_pro

(There's no VSTi version of Hauptwerk yet on OS X.)

The Mac Pros use the latest Intel Nehalem Xeon processors.

Hope that helps.
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline

sutherland

Member

  • Posts: 186
  • Joined: Fri Jan 12, 2007 11:50 am
  • Location: Scotland

Re: Cores v. Speed

PostTue Jul 21, 2009 7:43 am

Thank you, Martin, for your comprehensive response. That's made things clearer.

It's actually the Zwolle set I'm using - just wet at the moment, but I'm considering upgrading to surround so I want to have my system in readiness. I've got 24GB of RAM which copes well, but my current 5150 processor tops out with polyphony of just over 3000. I've just this minute picked up an E5320 on ebay for £65 and looking at past auctions my 5150 should make a similar amount, so it'll be interesting to see how much more polyphony the E5320 will give me. I do have a space on the motherboard for another processor so if I spot another E5320 and can find a heatsink I'll go for that too.

Switching to Mac would probably be a bit more of a challenge than I could welcome at the moment, so I will continue to upgrade my current system as good deals come along. I can go up to two X5355 2.66Ghz processors. At least that was the highest spec when the machine was new, but perhaps it can accept any socket 771 such as the X5365 3.0Ghz, I'm not sure.

I'll report back on the polyphony increase once I've installed the E5320 in case the information is of use to anyone else.

Thanks

Philip
Offline

Johannes Sørensen

Member

  • Posts: 228
  • Joined: Sat Jan 20, 2007 10:25 am
  • Location: Copenhagen, Denmark

Re: Cores v. Speed

PostTue Jul 21, 2009 8:17 am

Another question in relation to the number of processor cores.
In the way Hauptwerk generates the sound and uses the CPU cores, are there any benefits of more processor cores in respect to the chorus in the sound with large registrations as long as the processor power is big enough to generate the polyphony?

Best regards
Johannes
Offline
User avatar

mdyde

Moderator

  • Posts: 15444
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Cores v. Speed

PostTue Jul 21, 2009 8:48 am

Hello Johannes,

Apart from the total amount of CPU power available (polyphony etc.), the number of cores shouldn't make any difference to the sound.
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

micdev

Site Admin

  • Posts: 2099
  • Joined: Sat Feb 09, 2008 9:24 am
  • Location: Canada, Quebec

Re: Cores v. Speed

PostTue Jul 21, 2009 12:27 pm

CPUBenchmark.net offers graphical results of CPU tested with the Passmark software. By running the software on your actual Hauptwerk system, you will be able see how your system compare itself to other systems and from there you should have a good idea how your new system will perform.

Graphical results at: http://www.cpubenchmark.net/
Testing software at: http://www.passmark.com/products/pt.htm

Regards
François
Best regards
François

Virtually sharing my enthusiasm and experience with you
Worldwide technical assistance, consultation and ready to play system.

http://www.HauptwerkConsultant.com

AND Hauptwerk Support Manager
Offline

sutherland

Member

  • Posts: 186
  • Joined: Fri Jan 12, 2007 11:50 am
  • Location: Scotland

Re: Cores v. Speed

PostTue Aug 11, 2009 6:33 am

I thought I should update this thread with my CPU experiences. First, I went from a dual core 5160 3.0Ghz (not 5150 as I mistakenly said earlier) to a quad core E5320. This didn't make a whole lot of difference to what I could do with the Zwolle organ and for some strange reason I had to lower the polyphony to 2700 to avoid problems.

Since then I decided to upgrade again so I've gone for two E5345 processors which gives me eight cores and on the polyphony testing organ I can hold down all 61 keys with strips of wood and no break up occurs. I don't know how much higher it would go than 7625 as that is the highest multiple the testing engine will allow (125 pipes). 70% of that is 5337, but in fact I get little moments of breakup even if I set it at 5000 while using 25 stops (i.e. the equivalent of 50). I hasten to add that I am not routinely using 50 stops, but trying to establish what I need to get the surround Zwolle running smoothly on a large registration. I want to use the surround version instead of the wet. Maybe this means I need to go up to X5355 processors, but I'm not sure that processing power is a problem. I notice that most cores rarely go much beyond 10% in their use apart from one that goes to around 40%. Is this normal?

As for other specifications, I now have 32GB of ram and can load the surround organ mostly at 24 bit uncompressed, with some stops at 16 bit uncompressed. This fills up 31GB and allows me to use processor power for polyphony rather than decompressing samples.

I have disabled network connections and so on as recommend in the guide go getting rid of audio glitches. I am sure I can solve the problem by throttling back the polyphony, but according to the figures I shouldn’t need to do that.

Just now I reduced the dB in Hauptwerk to -12, increased the polyphony to 6,000, and increased the latency to 2048. This seems to be giving click free sound for now, but I haven’t played it enough to be sure. I notice the latency is reduced for 96Hz as opposed to 44Hz. I take it this refers to the settings on my EMU control panel. That is something I’ve generally ignored and gone for the 44hz default. Should I be making a different choice in the EMU control panel?

One other question, the E5345 have Enhanced Speedstep and I notice the processor speed flips between 1.8 and 2.3. Is it better to have this disabled for Hauptwerk so that the speed remains at max?

Thanks.
Offline
User avatar

mdyde

Moderator

  • Posts: 15444
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Cores v. Speed

PostTue Aug 11, 2009 7:36 am

Hello Philip,

I notice that most cores rarely go much beyond 10% in their use apart from one that goes to around 40%. Is this normal?


When you have 8 or more cores Hauptwerk will always dedicate one core to running the background models (tremulants, wind supply model, etc.) and relay, and the others to the audio engine, so yes - that's normal and not indicative of a problem.

I have disabled network connections and so on as recommend in the guide go getting rid of audio glitches.


What's the maximum value that the DPCLat utility shows on your PC? If that has high values then it indicates a driver/hardware issue that's likely to cause audio glitches.

Just now I reduced the dB in Hauptwerk to -12, increased the polyphony to 6,000, and increased the latency to 2048.


What buffer size were you using previously? You shouldn't normally need to go above 1024 (and 512 should also be fine, although you'll get a slightly lower polyphony). Try setting the buffer size to 1024 both in Hauptwerk and in the E-MU's control panel (I can't remember off-hand which one takes precedence with the E-MU driver).

I notice the latency is reduced for 96Hz as opposed to 44Hz. I take it this refers to the settings on my EMU control panel. That is something I’ve generally ignored and gone for the 44hz default. Should I be making a different choice in the EMU control panel?


Hauptwerk automatically sets the sample rate according to the sample set, so 48 kHz will be used anyway for the Zwolle. You just need to leave the sample rate unlocked in the E-MU control panel, so that Hauptwerk can change it when it needs to.

One other question, the E5345 have Enhanced Speedstep and I notice the processor speed flips between 1.8 and 2.3. Is it better to have this disabled for Hauptwerk so that the speed remains at max?


Definitely disable that, as well as any other similar power-saving features in the BIOS. If the clock speed gets reduced then that could cause audio glitches if the load increases suddenly (e.g. if you play a large chord after a period of inactivity).

Also try disabling the 'snoop filter' in the BIOS, if there's an option for it. A gentleman told me he needed to do that on one PC to get good audio performance.
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline

sutherland

Member

  • Posts: 186
  • Joined: Fri Jan 12, 2007 11:50 am
  • Location: Scotland

Re: Cores v. Speed

PostFri Aug 14, 2009 5:53 pm

Thanks again Martin.

Here's an update on the issues.

mdyde wrote:What's the maximum value that the DPCLat utility shows on your PC? If that has high values then it indicates a driver/hardware issue that's likely to cause audio glitches.


I'll download that program and report back.

What buffer size were you using previously? You shouldn't normally need to go above 1024 (and 512 should also be fine, although you'll get a slightly lower polyphony). Try setting the buffer size to 1024 both in Hauptwerk and in the E-MU's control panel (I can't remember off-hand which one takes precedence with the E-MU driver).


It was 1024, but there is certainly a direct connection between the clicks/break up and latency. At 512 there is a lot of breakup playing scales quickly even with just three or four stops pulled. At 1024 it is less frequent and at 1280 it goes away (but my fingerwork is not the fastest).

Definitely disable that, as well as any other similar power-saving features in the BIOS. If the clock speed gets reduced then that could cause audio glitches if the load increases suddenly (e.g. if you play a large chord after a period of inactivity).

Also try disabling the 'snoop filter' in the BIOS, if there's an option for it. A gentleman told me he needed to do that on one PC to get good audio performance.


This is proving to be rather trying because despite disabling Speedstep in the BIOS I can see that the processor speed is flipping between 1995Mhz @ 1.1625v and 2332 @1.3v. This appears to be a common frustration and I've tried disabling all power saving, but as for others on various forums this doesn't seem to stop it happening. Any idea how I can kill this and keep the speed at 2.3Ghz?

Another very odd thing is that Vista’s device manager was reporting that I had two Xeon 5160 3.0 Ghz processors, two Xeon E5320 1.86Ghz processors, and four E5345 2.33Ghz processors. I uninstalled the first two and rebooted so it is now reporting the correct eight E5345 processor for the dual quad cores. I’m not sure if this could have affected anything.

On the subject of Snoop Filters I have the Intel 5000X chipset and came across this article (http://www.dell.com/downloads/global/po ... ishnan.pdf) on the effect of having the filter enabled or disabled. It seems that whether it helps or hinders is partly application dependent. I’m going to try it off, but do you know if HW should be affected by it either way?
Offline
User avatar

mdyde

Moderator

  • Posts: 15444
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Cores v. Speed

PostSat Aug 15, 2009 6:07 am

Hello Philip,

t was 1024, but there is certainly a direct connection between the clicks/break up and latency. At 512 there is a lot of breakup playing scales quickly even with just three or four stops pulled. At 1024 it is less frequent and at 1280 it goes away (but my fingerwork is not the fastest).


My guess is that some driver/hardware component is causing excessive DPC latency. With a buffer size of 1024 shouldn't be getting audio glitches.

Any idea how I can kill this and keep the speed at 2.3Ghz?


No idea off-hand - sorry.

On the subject of Snoop Filters I have the Intel 5000X chipset and came across this article (http://www.dell.com/downloads/global/po ... ishnan.pdf) on the effect of having the filter enabled or disabled. It seems that whether it helps or hinders is partly application dependent. I’m going to try it off, but do you know if HW should be affected by it either way?


I'd suggest just trying it on and off and seeing which gives best performance.
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

polikimre

Member

  • Posts: 676
  • Joined: Tue Sep 12, 2006 7:39 pm
  • Location: USA, NC, Cary

Re: Cores v. Speed

PostSat Aug 15, 2009 6:33 am

sutherland wrote:This is proving to be rather trying because despite disabling Speedstep in the BIOS I can see that the processor speed is flipping between 1995Mhz @ 1.1625v and 2332 @1.3v. This appears to be a common frustration and I've tried disabling all power saving, but as for others on various forums this doesn't seem to stop it happening. Any idea how I can kill this and keep the speed at 2.3Ghz?


Try setting the power plan to "High performance" instead of Default or Balanced. (Control Panel, Power Options). At the same place you can change the plan settings, then click on the advanced settings link. Scroll down to Processor power management: you can set the minimum and maximum clocks there.
Offline

sutherland

Member

  • Posts: 186
  • Joined: Fri Jan 12, 2007 11:50 am
  • Location: Scotland

Re: Cores v. Speed

PostSat Aug 15, 2009 8:37 am

polikimre wrote:Try setting the power plan to "High performance" instead of Default or Balanced. (Control Panel, Power Options). At the same place you can change the plan settings, then click on the advanced settings link. Scroll down to Processor power management: you can set the minimum and maximum clocks there.


Thanks. I had been trying that, but the processor options did not show up in the options. I discovered they only appear if Speedstep is enabled in the BIOS and they are set at 100% when you choose high performance. The thing is that whether Speedstep is enabled in the BIOS or not and whether the clock speeds are set at 100% or not the processor speed and voltage keeps changing as above. It seems there are a lot of people (mostly gamers) suffering from the same problem, so there is lots of discussion about it, but no one seems to have a fix.

On a more satisfactory note, I've been using the organ this with Snoop switched off, latency at 1024, and the correct 8xE5345 processors showing in the device manager and I haven't been having any problems. I've had more than a few false dawns with computer problems so I'm not rushing to conclude this has solved it, but it's been trouble free for an hour or two.

Return to Computer hardware / specs

Who is online

Users browsing this forum: No registered users and 1 guest