@cm said:

 DG, please keep in mind this is on an i7 system and reduction of latency increases CPU load, not necessarily in a linear manner though.

more powerful systems (eg. using 2 x XEON W5580 = 16 virtual cores) will help, but i', not sure if 128 would be even possible ... maybe wait for the 6 core XEONS announced for early 2010

christian

EDIT: i just noticed that obviously it _is_ possible to get down to 128 on a 2 x XEON 5580W machine - this is more than i dared to hope for ...

Hi guys,
Don't know if anyone still remember me.

I've finally settled down for graduate. So finally I can get back to VSL community. Also finally I ordered VI for myself - although only Solo Strings I and Appassionata Strings I. 2500 dollars is a lot of money in China, espeically for a student:)

Looks like you guys are talking about supercomputer now! MIR is definitely a mir-acle:), this will render my research useless:) But I'm glad that such an easy to use environmental reverb exists. Although I don't know if I can buy that after its getting out. But I will buy it immediately if I can.

I'm now using a dual quad Xeon 5410 with 8GB RAM. I have to say that core number is only one question. Applications have to be fully optimized for multi-thread or multi-core in order to make full use of CPU. There are some problem with DAW on windows. I'm using Ableton Live now but I found that Live will drop out even my CPU has only 50% usage. I just hope that MIR is fully opitimized for multicore. cause in my view, multi-IR convolution is kind of parallel work so It is great for multicore.

Also, Core i7's integrated on-chip memory controller and L3 cache design can greatly improve memory performance. So I think that's why VI will run faster on i7. But i still believe memory bandwidth is only one factor on processing power. Processor's raw power will always be the most important factor.

and BTW, there always been supercomputer somewhere. I've seen 256 Xeon CPU server when it's 2002 in my city's electricity control center. Although these machines are for industrial control or more mission critical environment like finacial data center or something like that. My point is that there always are ways to build SMP(Symmetrical Multi-Processors) computers with 64, 128 or more CPU or cores using Server CPU like Xeon. But this always is my fantasy to have such a computer for music production :)

And, I don't think my 8 core Xeon 5410 with 8GB RAM will be enough for MIR and VI....may be

Also, BTW, it's good to be back.

I have the same question about inputting midi while using Mir.  All the examples that have been given seem to use really high (1024) latency settings.  From a composition/workflow standpoint, how does one input lines at lower latency?

@ozoufonoun_29353 said:

I have the same question about inputting midi while using Mir.  All the examples that have been given seem to use really high (1024) latency settings.  From a composition/workflow standpoint, how does one input lines at lower latency?

May be we could only use faster computers for MIR. Lower latency means higher computing intensity. So logically only solution should be using faster computers.

I can bare with 256-512 samples ASIO buffer size for composition. But no one is the same I think.

@YWT said:

I don't think my 8 core Xeon 5410 with 8GB RAM will be enough for MIR and VI

it will not be possible to investigate in detail on the performance of every computer/processor combination, but i'd estimate such a machine can run about 16 instruments placed on a MIR stage .... in any case it is powerfull enough for a lot of VIs ...

welcome back btw, christian

Thanks, but I can only afford the 920 at the moment. By getting a lower powered i7 am I only restricting how many instruments I can handle, and what would you take a guess at that number being?

 An unofficial result of a recent stress test undertaken by one of the MIR Beta testers:

@Another User said:

Primary System:  I really have not been able to stress it.

128 latency:  roughly 50-55 instruments @ 65-70% cpu

512 latency: 80  @ roughly 65-70%

1024: over 100 instruments @ 75%

Like Dietz, I used Perf Legato patches for each instrument.  Same line with adjusted ranges.  Buffer @ 1.  Bringing buffer up allowed me even more cpu room to work with.

... this was on a new built-to-the-task system.

HTH,

@Dietz said:

... this was on a new built-to-the-task system.

christian

Is Mir dependant on processor speed? I am planning on overclocking my 920 to compensate.

of course much depends on processor speed ... overclocking is a science by its own ... read through various overclcocking sites and you will find suddenly you need to care about memory types and voltages, heat dissipation, BIOS settings, bus frequencies, ect.

in a nutshell: overclocking is nothing supported by VSL as hardware in general is not our business. we gladly provide information about working configurations and some insight where possible though.

in other words: to transport a ton of bricks we recommend to use a truck. maybe you could do that with your cadillac too, but do this at your own risk.

christian

 Hi Christian

Do you happen to know what motherboard the dual Xeon system detailed above, which delivers such good results, is running on?

I'm ordering a system tomorrow as I've hit a capacity ceiling on our current VI machine and need to move all our Vienna stuff onto a separate system.  It would be great if I could spec something that will run Mir easily.

Many thanks

Jules

 INTEL S5520, Supermicro X8D, Gigabyte GA-7T - all available in several versions depending on your other needs.

again: strictly stick to the compatibility list of the mainboard when ordering memory!

be aware only 1 memory module per channel (= 3 per processor) allows 1333 - if you plan to insert more (6 per processor) you are fine with 1066 anyway.

christian

Many thanks Christian.

I suspect I'll have to go the 2GB module route (6 per processor) in order to minimize cost, and may even have to start with 1 CPU and add the 2nd when prices fall a little.  Better that way than have a system with no room for 'growth'.

I'm finding that the 24GB requirement really limits spec options a little.  I had an X58 system all ready to go, but I just couldn't find reasonably priced 4GB unbuffered RAM modules, and I didn't fancy maxing it out with 6x2GB modules, and then discovering it wasn't enough.  4GB ECC registered modules are a fraction of the price, meaning you can spec a Xeon system for pretty much the same cost.

I'm hoping 3 1TB drives with the VI's spread across them and a 9500GT should be ok on the HD and graphics front.

Thanks again for your help.

Jules

Hi,

Just ofr your information, I have very bad low latency performances with the Intel S5520.

Standard 4GB sticks are very expansive, right, but you can put a Xeon 35xx on your X58 motherboard and use 4GB Ecc sticks.

the X58 motherboards are not compatible with ECC memory modules (see above) and i can't confirm the latency issue for the 5520

christian

That's the reason I asked about the motherboard in the system which is reporting such good performance.

That said, I would have thought that if one X5500/5520 chipset motherboard has latency issues, then others will have too, but maybe not.

Anyhow, the spec for our rig has evolved, the system has been ordered and is currently being built - luckily I only needed one extra mortgage on my house to pay for it!

Spec is Supermicro X8DT3, 2xIntel Xeon W5580 (3.2GHz), 24GB Ram in 4GB modules (this way the memory bus can still run at 1333MHz because there's only 1 DIMM in each channel - this mobo has 12 memory slots), Vista 64bit Ultimate, mixture of 1TB SAS and SATA drives, 512MB 9500GT Graphics etc.

It should be as fast as current components allow.  I'll post as soon as I have some results.   If it doesn't work I'm in deep trouble!!

Jules

http://www.trailermen.com

@cm said:

the X58 motherboards are not compatible with ECC memory modules (see above) and i can't confirm the latency issue for the 5520

christian

Yes I agree, X58 isn't compatible with ECC, but it doesn't have to. The memory controler isn't in the chipset but, in the Cpu. Ecc support is one of the few difference between the Xeon 35xx/55xx and the Core i7. So you can use Ecc memory on a standard X58 motherboard if you use a Xeon.

But.......... but the thing I didn't notice is that you can't use ECC registered sticks on a X58 motherboard (I don't know why), and non-registered ECC sticks are more expensive than non ECC.

@Trailerman said:

Spec is Supermicro X8DT3, 2xIntel Xeon W5580 (3.2GHz), 24GB Ram in 4GB modules (this way the memory bus can still run at 1333MHz because there's only 1 DIMM in each channel - this mobo has 12 memory slots), Vista 64bit Ultimate, mixture of 1TB SAS and SATA drives, 512MB 9500GT Graphics etc.

I have very bad results with the S5520, but everything is ok on a X8DAi, so I think you shoudn't have any problem with the X8DT3.

 Thanks Stephane - I feel better about the motherboard choice.

You're right that the memory situation is a little confusing, and the wild variations in price between buffered, unbuffered, ECC, non-ECC, registered etc. doesn't help.

Jules

IMO the most important specification for an intel 5520 based motherboard is that it offers a PCIe-x16 PEG slot for grafic card ...

for the memory i can only repeatedly recommend to strictly stick to the compatibility list of the motherboard and actually have a close look at the possible combinations of memory modules (dual rank, quad rank, buffered, unbuffered, ECC)

christian