Login

Questions and facts one needs answered to build nebula3 PC

Tips & tricks, working results, technical support

Questions and facts one needs answered to build nebula3 PC

Postby reort » Wed Jun 22, 2011 3:34 am

Below are questions and facts one needs to answer to purchase a PC tailored for Nebula3 Server.

PC specifcations used as a guidline:
Intel Core i7-960 Bloomfield 3.2G, 12gig tripple channel DDR2 1033. (approx. 20% cpu savings on i7-2600 Sandy Bridge 3.4GHz with AVX under same conditions), 64bit OS (32 bit OS wrapping nebula instances with Jbridge), Gforce GTX 285 2Gig memory. Nebula3 server installed on Intel 160 GB X25M .


1.Above PC setup used as a guidline the first resource nebula seems to saturate are CPU resources.
2.Fast sequencial read flash drive (assuming a fast processor in case of frequency conversion) will shorten program load times (good for comparisons).
3. Does memory bandwidth and speed beyond DDR2 1033 tripple channel make any difference in loading times or CPU resources.
4, How does local server allocate resources:
a) host independent threads per nebula instance?
b) host independent multiprocessor allocation?
c) Huge memory savings (as reported on acusticaudio forum) when multiple of the same programs are loaded running local server. Are there also CPU savings under the same conditions?
d) Giancarlo is coding so that we can run multiple instances of server (cuda, CPU, local, network, network2, etc.) on the same PC. Are these instances going to share resources with each other ( if so then how would netwoking bandwidth effect resource sharing)?

Cuda:
5. "CUDA is for long tailed kernels, such as: kernels greater than 150 ms, huge performance gain if ALL kernels are tuned at the same kernel size 3) possibly kernels should not be dynamic
Use CUDA for: reverbs, static eq, where all kernels are longer than 150 milliseconds and possibly all kernels have the same length."
c) Maybe developers could release cuda optimised versions of a particular library?
6. Giancarlo is working on PURECUDA, a solution for running the whole nebula engine on cuda
a. Would that allow us to use Cuda for all libraries?
b. is that going to reduce cuda introduced latency?
c. currently memory (Gforce GTX 285 2Gig memory) is the first resource to be saturated how will 20 kernel programs effect this?


Secondary questions:
7. does renaming Nebula3 Reverb.dll to Nebula3 Reverb2.dll, etc.serve the same purpose as wrapping with jbridge?
reort
User Level I
User Level I
 
Posts: 12
Joined: Mon Apr 05, 2010 6:13 am

Re: Questions and facts one needs answered to build nebula3 PC

Postby giancarlo » Wed Jun 22, 2011 10:01 am

about purecuda, we don't know anything yet. We are simply porting the whole engine to cuda, I hope it will solve all cuda issues... I don't know.
For all other questions, you can check new version of nebula server 1.3.499a. Maybe it could crash at loading time, but you get an idea about new nebula engine. Buy a fast computer, the faster the better, based on sandybridge. The faster, the better. We are supporting all new intel optimizations because we just bought ipp 7.0, compiled in april 2011. All new processors are supported properly.
User avatar
giancarlo
Founder
Founder
 
Posts: 9143
Joined: Mon Sep 21, 2009 10:40 pm
Location: Italy


Return to Working with Nebula

Who is online

Users browsing this forum: No registered users and 4 guests