My next motherboard

Finally got a 60GB Sony PS3. Took forever to get them in stock. Will be installing Debian Linux on it:

http://news.softpedia.com/news/First-PlayStation-3-LiveCD-44527.shtml

Using the PS3's multicore IBM Cell Processor to develop high performance software will be interesting. I am going to contrast the PS3 performanace against the test machine that I built above with the ATI X1950XTX GPU.

Supposedly, the performance in GFLOPs should be comparable...

nitro
 
Just began research into storing ticks into a SQL database. I decided to start with SQL Server 2005, and was soon hit with this:

"(NOTE: Although SQL Server allows for 3 digits of fractional precision, Microsoft's documentation for its datetime (i.e., timestamp) data type says:
"Date and time data from January 1, 1753, through December 31, 9999, with an accuracy of three-hundredths of a second, or 3.33 milliseconds.")"

I can't believe it :eek: 3.33 milliseconds resolution :confused: That makes it worthless for storing financial tick-data, which can have resolution of less than 1 millisecond. Unless I am missing something, I am moving to Oracle 10g, which has nanosecond resolution:

http://www.dbazine.com/db2/db2-disarticles/pelzer2

Bummer :(

nitro
 
Some people through PMs have asked me to update my use of GPUs. The reason I have not done anything with them is that it turns out that they use single precision arithmetic.

Fourth 1/4 of this year nVidia is coming out with double precision cards. Then it becomes a serious effort to get things working on them. Until then, this is on hold.

nitro
Quote from nitro:

I have found a new direction in my search for supercomputer like performance from my systems. It is basically described in this thread (although slightly scattered between posts):

http://www.elitetrader.com/vb/showthread.php?s=&threadid=82244

As I gain more experience with using GPUs for general computation, I will try to post it to this thread.

One thing about these GPUs is that the state of the art GPU cards require PCIe X16 slots. The bad part is that the Quad Opteron that started this thread has _no_ PICe X16 slots :( :(

So over the weekend, I went out and built a single processor computer from parts using a DFI motherboard that has two PICe X16 slots, and I put an ATI Radeon X1950 XTX. This is a research machine for me, but while I am not using it, it is running folding@home:

http://folding.stanford.edu/FAQ-ATI.html

My next step is to put in another ATI X1950 card in, but that is a way off until I gain experience programming them.

nitro
More...
 
The Quad Core AMD opterons are out. For me at this point I am on the outside looking in, as the Tyan described in this thread uses an older socket that is incompatible with the new opterons.

Why anyone would use an INTC server chip at this point is beyond me. The Opteron design is superior in multicore scenarios, and equally important, they generate less heat. AMD is green!!

Soon I will likely have to get a rack of machines as I am going to be trading an ATS using OPMs. I will have a small but not insignificant technology budget. Unless INTC does something drastic, I know which chips will drive my machines:

http://www.blade.org/productdetail.cfm?RecordID=8

with AMD quad core opterons.

You know, for years I never understood why AAPL didn't take over the world. It never made sense. I guess you just had to wait for the right time. I believe this even more strongly about IBM. I have never understood why this company doesn't dominate its space. It seems to do so much right, especially in R&D, and yet it's stock price rarely does anything spectacular. Where are the products that are the result of all that great R&D??? I don't understand why IBM can't sell it's products. Perhaps it too will have its day...

nitro
 
Ok, nVidia high end GPU card is out:

http://www.nvidia.com/object/quadro_fx_4500_x2.html

This is now the card to beat, but I believe it does not support double precision still. I need to call nVidia to make sure...

I think the 128-bit spec means 4-wide 32-bit operations, not the same thing as double precision.

nitro
Quote from nitro:

Some people through PMs have asked me to update my use of GPUs. The reason I have not done anything with them is that it turns out that they use single precision arithmetic.

Fourth 1/4 of this year nVidia is coming out with double precision cards. Then it becomes a serious effort to get things working on them. Until then, this is on hold.

nitro
More...
 
Now looking for a high density system, I am at a crossroads.

I am considering these systems:

http://www.dell.com/content/products/productdetails.aspx/pedge_1955_new?c=us&cs=04&l=en&s=bsd

http://www-306.ibm.com/software/data/integration/info_server/blade/

http://h18004.www1.hp.com/products/blades/components/enclosures/c-class/c3000/index.html

http://www.supermicro.com/products/superblade/

The Dell is the most cost effective, but doesn't have integrated storage blades and is Intel Xeon based only.

The IBM is very nice, but I feel like I am getting an avarage product for more $$ than other comparable systems. Support for opterons is big.

The HP is perfect in every sense. It supports opterons, allows built in storage over iSCSI, integrated tape backup etc. Problem is it is outrageously expensive. For example, a 1TB iSCSI storage is $9000!!! A PCIe expansion bay is $800!!!

The supermicro is comparable with the DELL, and is certainly a better buy. Problem is, they claim to support opterons but the opteron blade server is nowhere to be seen.

If anyone has any other suggestions, I would be glad to hear them.

nitro
 
nitro - forgive my ignorance, but what are your plans with these systems? I assume it's for trading and as much as I love my toys, I'm not sure where these systems would be used. If it's not for trading, sorry for intruding.
 
You must log in or register to reply here.
Back
Top