Masterverb Test Thread

[garyb]

it's a pretty plain 945g intel chipset.

i think the main difference is that the XITE-1 is on the pci-e bus, which has more bandwidth than the pci bus.

[dante]

I can only get 13 masterverbs on my XITE1D - exactly the same as what I got on my 3 x cards (12DSP). Mobo is ASUS p5KE Intel quad core 4 gig ram. I have my PCI cards still installed but disabled in device manager. Any ideas about how I can get more PCI(e) bandwidth (or indeed find out if theres a common bottleneck on both PCI and PCIe throughput) ?

Seems to me whatever resource was restricting system to 13 masterverbs on PCI cards could be the same as whats restricting it to 13 on XITE1D (?)

sysinfo

sysinfo.JPG (145.54 KiB) Viewed 4105 times

[siriusbliss]

Can you assign them to different DSP's in the D?

Greg

[garyb]

what exactly does the error message say when you reach the 13th(screenshot)?

13 masterverbs is plenty of bandwidth for most situations....

[dante]

Just DSP overload. Makes no diff if I manually assign. Today its maxing on only 11 for some reason.

masterverb test.

masterverb-test.JPG (110.52 KiB) Viewed 4098 times

[astroman]

imho the MV Test is misleading ...
It was designed to measure a specific property of mobo cipsets regarding access to main memory from a PCI card.
The XITEs try to avoid this by intercepting the original 'code' and remap the request to local memory on the card.
I have no idea about how they deal with onchip versus card DRAM, but they definitely handle it that way (I was told on the phone).
Hence you don't see a representation of the reverb capabilities of your new device but sideeffects of a specific method that makes few sense in this context.

MV Test is a quick and dirty way to estimate some mobos capabilities, but it's very unreliable regarding comparable results.
I've NEVER been able to get anywhere close to those 18 instances reported with the Asus TUSL boards - WITH exactly that mobo...
Regardless of Scope or PC-OS version, 1st or 2nd generation cards... there were all kinds of errors, many DSP memory related ('large modules don't fit' or so), even different results on subsequent installations on the very same hardware/OS combination.

It's a complex topic under the hood and something (or more) must have changed in the Scope software over the years.
I still have a PT reverb around that (often) acts totally strange...

cheers, Tom

[dante]

Yep, but one thing we do know is that its a restriction specific to the host PC. If GaryB can get 23 mverbs running on a $200 laptop then suggests to me if I want full power out of XITE then I would follow suit in the absence of well known tweak for specific mobos, taking into account ensuring there wasnt a huge trade in host power for VTS's etc.

[astroman]

well... see it from a different angle and it will become more clear.... that it's NOT related to the host PC.
Gary used the XITE-1, which has 3x the amount of chips compared to your box, the new ones that is
which would lead me to the estimation that Gary's system would fail somewhere around MV #50

You cannot extrapolate that so called calculation power in a linear way if you compare an old to a new Sharc.
Those chips are clocked higher (5x) which doesn't mean you can automatically load 5 instances of an old module into them.
In particular if there's a kind of 'fixed timing' involved (delays).
So it's more likely that the 'scheduler' spreads them over several chips - and you have 'only' 4 of those.
Additonally the new DSPs can (effectively) double the operations per cycle on the same data set (SIMD), which usually requires recoding (imho), but possibly SC has provided some general 'automation' (dunno).
5 times the clock and 2 times the instructions gives that virtual 10 times the power frequently mentioned with those DPS.
As this is code dependant, your real mileage DOES vary
But at least I wouldn't expect any major efforts for the classical Masterverb, as there are 'more prospecting' devices from various sources...

cheers, Tom

[maky325]

How so? Didn't some of you guys posted back then when xite was announced at some show that one can run more then 80 masterverbs etc..

Here is Jimmy posting how he is able to run 43 Ambience reverbs. AFAIK Jimmy know what he is doing and Ambience is much much heavier device then simple Masterverb. If he is able to run 43 Ambience i assume about 60 Masterverb plugins should be possible on XITE big.

http://forums.planetz.com/viewtopic.php ... hilit=xite

[garyb]

you don't have dsp overload.

read the mesdsage. you are out of dsp resources for the reverbs. this seems to be pretty much the case for the XITE-Ds. it's possible that in the future, there might be optimization for the Ds that allows better use of resources, but it has nothing to do with motherboard resources, afaik. my guess is that it has something to do with the old and new dsps. if you load the masterverbs and assign them to other dsps, you may find that you can load more....

[astroman]

How so? Didn't some of you guys posted back then when xite was announced at some show that one can run more then 80 masterverbs etc....

so what ?
I'm not interested in numbers and 'mine is bigger than yours' bla... just commented Dante's report and gave a couple of (imho) good reasons.
If you read with (more?) attention, you may have noticed that Dante has the 'small' XITE-1D
and you may have noticed that I already pointed to the apple-versus-oranges nonsense of power comparison...
So why do you bring it up again with absolutely nothing to contribute ?
You're welcome to criticise and question my arguments, but please back it up by some content.

cheers, Tom

[dante]

I realise the XITE1 has way more DSP than the XITE1D, but then in turn the XITE1D has way more DSP power than my PCI (12DSP) cards (again a factor of around 3 x) so I suppose I thought it to be a strange 'coincidence' that both the cards and the XITE1D flaked on exactly the same number of instances ( 13 ) on the same host system. And in general, I can run much more of everything else on XITE1D than PCI.

At this stage I'm open to it being either host or the XITED itself, as I also tried to test on a HP laptop using the PCIExpress card I got with the XITE. Unfortunately I got other errors opening Scope on the HP laptop, even though the installer recognised the XITE1D. It could have been that PCiexpress socket on the HP didnt easily engage into place, due to the housing. But if I get a laptop with a decent connector I'll try the test again.

[dawman]

Yes it loaded that many on the first revision May '09, but it wasn't until further testing that hooking them up, as opposed to just loading them is an entirely seperate matter.
I use 7 Reverbs now and a hardware Verb in an AUX.
Haven't even tried any cock measuring since then actually.

We should route this thread over to the GS forum thread where dozens of experts are all red faced arguing over nullification theories.
Perhaps this thread can nullify the GS thread and we can all get back to making music.

Ankyuvarymush.

[the19thbear]

[maky325]

How so? Didn't some of you guys posted back then when xite was announced at some show that one can run more then 80 masterverbs etc....

so what ?
I'm not interested in numbers and 'mine is bigger than yours' bla... just commented Dante's report and gave a couple of (imho) good reasons.
If you read with (more?) attention, you may have noticed that Dante has the 'small' XITE-1D
and you may have noticed that I already pointed to the apple-versus-oranges nonsense of power comparison...
So why do you bring it up again with absolutely nothing to contribute ?
You're welcome to criticise and question my arguments, but please back it up by some content.

cheers, Tom

Dude i don't know what is your problem? I wasn't talking to you at all. I was referring to Dante result. I didn't quote any of your words? I thought it's pretty clear since i am talking about performance results. I don't know what the heck are you talking about and what do you mean with " 'mine is bigger than yours' ". To you it can be "so what" but in case you didn't noticed there are a lot of potential future XITE users.

If one user reported pretty different performance result (dante) then other (jimmy in this case) i would really like to know what caused it and is there any kind of configuration difference which is wise to avoid(for example different Motherboards). You know there are a lot of people which would like to get any possible info before buying some gear. It is not necessary always childish co** measuring if that is what you mean with your mine is bigger then your thing sentence. I feel sorry for you and your very distorted moment and i hope it's just a tickle in time. Co** measuring...a what?? Dude..

Please calm down and be helpful if you can contribute with some results. So far i am aware that Jimmy has stronger XITE unit then dante but difference is pretty huge. Jimmy posted info (unlike you) that his numbers are from some early Scope software build but then i am curious to know why these results decreased with newer software version? Shouldn't this be opposite?

[garyb]

i believe it's a difference in architecture between the XITE-1 and the XITE-D that makes the discrepancy. there are other resources besides bandwidth that control how many reverbs can be run. again, Dante did NOT have PCI overflow.

[dante]

When I asked Ralf about the slotting arrangement on the 'D', he also told me that inter-DSP issues should be reduced, due to thier being less of them :

>> regarding the first 6 communication Sharcs, its the same.
>> On XITE-1 D you have also 4 slots, but each slot has only one fast DSP,
>> which make the communication problems less and distribution more
>> easy than on XITE-1. So in most cases XITE-1 D should be more easy to
>> configure.

So I dunno, must be something else. I will test on another host, I'm considering asking S|C if I can buy another PCIe card making the 'D' easily swappable from one host to another (or find a laptop with the right slot). But if anyone else has a 'D' and can do the test, Im definately interested in the results. I've asked Ralf if he can do the test, if they have a 'D' lying around.

[astroman]

Please calm down and be helpful if you can contribute with some results. So far i am aware that Jimmy has stronger XITE unit then dante but difference is pretty huge. Jimmy posted info (unlike you) ...

well... I was indeed kind of p*ssed and I feel right that again
you ignore a thorough and precise explanation WHY those results differ which I gave in my post above.
While that explanation makes sense - I'm not SC, so I don't claim it's ultimate truth.
Yet I've given a prediction around which number of instances it's supposed to fail on the big XITE.
Anyone curious enough could enter those devices...

Additionally and more important (!) I've explained WHY the test is plain nonsense on the XITE or XITE 1D
and should NOT be considered any measure of performance AT ALL
it was created for a specific part of an 'old' chipset architecture that doesn't even have a connection to any XITE

I gave examples that my own test results with the same mobo (in the supposed environment) were almost 50% off from others, making the test non-trustworthy on it's own.
Shroomz once had some 40 instances loaded on a Pulsar II and could keep loading and loading until he noticed the system was actually hanging, though it pretended to be still operable.

cheers, Tom
(pretty calm btw)

[astroman]

So I dunno, must be something else.

sure it is - you're flogging a dead horse ...

THAT reverb was written with what was there at it's time - today you'd use a completely different strategy.
The 'loader' cannot optimize the code, it can just move it to a location that pretends to be it's familiar environment

cheers, Tom

[garyb]

When I asked Ralf about the slotting arrangement on the 'D', he also told me that inter-DSP issues should be reduced, due to thier being less of them :

>> regarding the first 6 communication Sharcs, its the same.
>> On XITE-1 D you have also 4 slots, but each slot has only one fast DSP,
>> which make the communication problems less and distribution more
>> easy than on XITE-1. So in most cases XITE-1 D should be more easy to
>> configure.

So I dunno, must be something else. I will test on another host, I'm considering asking S|C if I can buy another PCIe card making the 'D' easily swappable from one host to another (or find a laptop with the right slot). But if anyone else has a 'D' and can do the test, Im definately interested in the results. I've asked Ralf if he can do the test, if they have a 'D' lying around.

yes, your results are normal for the D from what i've seen. as i said, it's not a bus bandwidth thing.
yes, you can order an additional pci-e card.