Chipset Tests

subhuman · Post by **subhuman** » Tue Jan 22, 2002 9:30 am

I have, over the past 2 years, tested many motherboards on my own to find the 'perfect motherboard for DAW.' Since I had been doing this for myself, I wasn't as concerned with hard numbers, but now that I've mentioned some chipsets/motherboards are better than others, and people have been wanting proof, I am feeling obligated to provide some hard numbers for those people to look over.

So, I need to find a scientific way of quantifying my results - they must repeatable, and easily tested by anyone. Of course, the easiest way for us to test, is using our Creamware DSP cards to saturate the PCI bus. Another way would be to run a RAID array off a ie: Promise ATA100 RAID card, or a SCSI RAID array, and measure the amount of PCI traffic.

For the Creamware DSP card test, I used a fresh clean install of WinME (although, any version of Windows gives similiar, if not identical, results), on the VIA boards, the latest VIA 4-in-1 were installed (4.25 or higher), latest BIOS, and no other tweaks applied. A G450 card was used on the AGP bus, with the latest drivers at the time of testing, no tweaks done with it however.

I then installed a Pulsar1 board, installed it's V3.01 driver, made sure it had it's own IRQ, then installed Pulsar V3.01. You will get similiar, if not identical, results using Luna V3.

Now, drag a masterVerb into the project, which uses the PCI bus to access main memory for the delays it uses to make an accoustic 'space.' Each one of these eats a little bit of PCI bandwidth. Load as many masterVerbs as possible until "PCI Overflow" messages pop-up. Sometimes if you "retry" or "optimize" your project, you can get one more verb in. We kept our Sample Rate dialog open, and our DSP Load dialog open, to make sure we're getting clock, and the DSPs are showing load. The number of reverbs you can load here is the "load number."

Now, run audio through each of the reverbs using something like WinAMP. You might have problems trying to get them to work - remove a reverb until you get "stable" reverbs. This is the "run number."

I'll post my results for various motherboards (and therefore, chipsets), in a bit.

dblbass · Post by **dblbass** » Tue Jan 22, 2002 7:11 pm

Love this idea. Not an easy issue to measure simply, though, is it.

for example: motherboard/chipset X might do better in "load number" and "run number" than motherboard/chipset Y, both untweaked, and thus X would look like a winner. But then, on of us might discover that with just the right BIOS or OS fiddling, Y could be made to perform much better than X, no matter how tweaked.

But no matter, your idea is great. I hereby propose we name the benchmark after you!!!
We can all publicise and compare whatever board/tweak configurations we develop with a relevant, reasonably consistent benchmark, e.g.:

"I got a SubTest Load/Run count of 7/6 !!"

You would think someone would write a very simple utility proggie which would run at low level and monitor, analyse, and report on PCI traffic. I have used similar such things to look at threads or registry activity, or HD performance, why not PCI bus? Maybe there is one out there somewhere?

at0m · Post by **at0m** » Tue Jan 22, 2002 10:02 pm

Here's a nice tool to check your pci load

I'm going for the Subtest now. I could load 8 Masterverbs, not under the same circumstances. Let's get the test standardised.

Sorry, snapshot's only for yahoo group members.

I got DSP meter in red band, 44.1kHz, 3ms latency, 8/6 from wav to micromixer to analog out. XingPlayer and MediaPlayer continue to play, even heavy programs open without delay. My Acorp6A815EPD pc 1 Pulsar2.
When I added a Pulsar1 to that, I could get to 9/10. And I had only 80% DSP use.
This shows that (in current setup) my 6A815EPD pci bus can handle 9 MasterReverbs.
It has a separate pci bus for IDE and USB.

My Soyo BX with 2 Pulsar1 cards (usualy;) can get 3/8 to run stabile, 4/8 when nothing's done on the pc. If I work a bit, 4/8 is not doable, I can only do 3/8. Pci slots, IDE and USB are on the same PCI bus.

[ This Message was edited by: atomic on 2002-01-23 11:49 ]

[ This Message was edited by: atomic on 2002-01-24 03:48 ]

musquash · Post by **musquash** » Wed Jan 23, 2002 4:18 am

i could load and run 8 masterverbs @ 44.1khz 3ms latency using a pulsar2
i used winamp, played by another soundcard, connected to analog-in, through all 8 reverbs in a chain to analog-out
(btw: it's a chilly psychedelic bell-sound, i feel like a monk on lsd now

Neutron · Post by **Neutron** » Wed Jan 23, 2002 8:56 am

On 2002-01-22 22:02, atomic wrote:
Here's a nice tool to check your pci load

he is getting one

I have one allready

JoeKa · Post by **JoeKa** » Wed Jan 23, 2002 7:48 pm

3 questions:
-What is the effect of the "snoop ahead" setting in the Bios on Pulsar?
-As Pulsar has its own quartz mounted which clocks the DSPs to 60MHz, does overclocking of the MoBo affect the DSPs in any way?
-Can it be that Masterverb is much more PCI-hungry than the MV Classic?

On my P3B-F(BX440) with P3/500 I can run 8 from 10 loadable "MV Classics" stable @ 44.1, 3ms, but with the same settings I get only 6 of 7 "normal" Masterverbs running stable... How come?

[ This Message was edited by: JoeKa on 2002-01-23 19:49 ]

kimgr · Post by **kimgr** » Thu Jan 24, 2002 12:03 am

On my P3B-F(BX440) with P3/500 I can run 8 from 10 loadable "MV Classics" stable @ 44.1, 3ms, but with the same settings I get only 6 of 7 "normal" Masterverbs running stable... How come?

Because the Classic doesn't have the same early-reflection engine.
Less delaylines=less PCI traffic.

Kim.

eliam · Post by **eliam** » Thu Feb 07, 2002 8:04 pm

Anyone have test-results with a P4 mobo? I wonder what is the pci limit of these...

at0m · Post by **at0m** » Fri Feb 08, 2002 4:08 am

i wonder what's the pci limit of yours...

marcuspocus · Post by **marcuspocus** » Fri Feb 08, 2002 5:56 am

I don't know exactly about P4, but i'm pretty sure it better than the one on Intel i815ep, which is one of the best with at least 8 masterverb without PCI Overflow...

I suspect 10 or more Masterverb on a good Asus P4TE with P4-1.8ghz +

at0m · Post by **at0m** » Fri Feb 08, 2002 6:04 am

A little change in config. Acorp 6A815EPD, 512MB CL2 Apacer, 2 P3 @ 1087MHz. All busses at 145MHz. (Other frequencies give me the same results)

Load Number: 10,
Run Number: 9 Stereo Masterverbs.

To EarlyFirst: How fast did your Acorp go when it got fried? Some advisory would be appreciated

[ This Message was edited by: atomic on 2002-02-23 20:31 ]

Neutron · Post by **Neutron** » Tue Feb 19, 2002 9:01 pm

sorry to sound dense but are they mono or stereo masterverbs you are measuring with

Immanuel · Post by **Immanuel** » Wed Feb 20, 2002 10:29 am

ECS k7s5a:

With a retry, I got 4 masterverbs running errorless parallel thru a dynamixer (windows media player as source). I do not have DSPs for any more masterverbs, but considdering I never hit 4 masterverbs without a retry, I guess, there is not much more go for anyway.

Immanuel

Neutron · Post by **Neutron** » Wed Feb 20, 2002 12:12 pm

the intel i815 and i845, and i850 all use the same ich2 chip for the PCI bus so i would expect results to be close.

Rob van Berkel · Post by **Rob van Berkel** » Fri Feb 22, 2002 4:21 am

To Sub: when you mention 'load a MV' do you only load, or do you connect them as well?
Is there a (PCI-usage-)difference in connecting all MV-inputs to the same audio-src and leave the MV-outputs open? What if you mix down the outputs. How about putting the MV's in cascade ('elephant-mode')?
Imho, to succesfully setup a benchmark and being able to compare its results, one should describe a strict environment.
I think it's a great idea and I will do some tests on my P4/1.7 system (I850 chipset) this weekend.

Rob

[ This Message was edited by: rvberkel on 2002-02-22 04:22 ]

at0m · Post by **at0m** » Fri Feb 22, 2002 11:33 am

Connect the Analog source to all MV's Inputs, and mix them down using the simple mixer, to Analog Dest.

Connected, the MV's are activated.

subhuman · Post by **subhuman** » Fri Feb 22, 2002 11:37 am

No elephant mode as that seems to give different results - do it as atomic menions, each getting a signal from WAV source, to a mixer, output mixer so you can verify its working. Use 'Stereo Masterverb' device.

[ This Message was edited by: subhuman on 2002-02-22 11:37 ]

at0m · Post by **at0m** » Sat Feb 23, 2002 7:34 pm

This links to 2 example projects I made for the subtest.

Another attempt to standardise it. Let us know how much MasterVerbs you can load!

Check this one for my Sisoft Sandra Memory benchmark during the Subtest.

[ This Message was edited by: atomic on 2002-03-02 21:05 ]

JoeKa · Post by **JoeKa** » Sat Mar 02, 2002 5:45 pm

Ass-kicking new system I´ve got...

Celeron 1200 on Asus TUSL2-C, standard BIOS PCI settings, I get 12 of 13 Masterverbs running now.
After loading the 9th MV I felt my system running slower, but I was able to load 13 MV´s and start Winamp, but after a few seconds, my system froze. With 12 MV´s, my system still works, ok, very slow, but it works.
NO PCI-OVERLOAD MESSAGE AT ALL!!!
Nice MoBo, I must say... *g*

Rob van Berkel · Post by **Rob van Berkel** » Tue Mar 05, 2002 2:42 pm

Just want to check if I'm doing the Sub-chipset-test right: I loaded <A HREF="http://www.berkel.net/mv.gif" TARGET="_blank">this</A> project. As you can see there's 11 MV's without any PCI message. With 12 DSP's all that I run out is horsepower (partly due to the mixer). Hmm, doing something wrong here or is the Abit TH7-II (no RAID) and P4/1.7 an ideal combo?
BTW, when using WAV or ASIO source instead of direct inputs like SPDif and Analog-IN the signals travel one more time over the pci-bus: from soundsource (program in memory) to pulsarboard. Same goes for WAV or ASIO output. This should be taken into account when comparing, should't it?
So, wouldn't it be better to use physical ins and outs to really have the # of MV's be a measurement for the PCI throughput?

Rob

[ This Message was edited by: rvberkel on 2002-03-06 04:33 ]

[ This Message was edited by: rvberkel on 2002-03-06 04:33 ]