SERVING THE QUANTITATIVE FINANCE COMMUNITY

 
User avatar
Cuchulainn
Posts: 62394
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

sample cuda problems in finance

April 19th, 2015, 12:36 pm

QuoteOriginally posted by: AlexEroQuoteOriginally posted by: CuchulainnQuoteAs for AMP I can not accept for serious work anything, created within Microsoft. (Visual Studio is actually a GUI shell to Lattice C compiler, originally created by another company). Loud and clear, you have shown your true colours. FYI I have use Microsoft C/C++ since 1989. So, essentially you are calling people like me idiots.I think you have just lost 90% of the Wilmott community with this crass remark. You don't have a clue. Anyhoo, in fairness I see where you are coming from. What was your name again?QuoteRecently we have released new version 1.70 of our "Graalino" trend following Expert Advisor (EA) for MT4 platform - with full support of CUDA acceleration.C'mon. Do You beleive MS spokeperson?"The Lattice C Compiler was released in June 1982 by Lifeboat Associates and was the first C compiler for the IBM Personal Computer.[1] The compiler sold for $500 and would run on PC DOS or MS-DOS. ... The compiler was subsequently repackaged by Microsoft under a distribution agreement as Microsoft C version 2.0.[3] .....A vendor spokesman said Version 3.0 is the first C compiler developed internally by Microsoft."Lattice C compiler
 
User avatar
Cuchulainn
Posts: 62394
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

sample cuda problems in finance

April 20th, 2015, 6:48 am

I copied some AMP code and compiled. Can anyone say whether this is so terribly bad and why?Of course, CUDA is closer to the hardware do it is more efficient bla bla bla :D We know all that.
 
User avatar
Cuchulainn
Posts: 62394
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

sample cuda problems in finance

April 20th, 2015, 7:01 am

Sure, raw CUDA is faster than both OpenCL and C++ AMP (about 30-40%..) Now serious is that? Performance
 
User avatar
Cuchulainn
Posts: 62394
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

sample cuda problems in finance

April 20th, 2015, 7:04 am

QuoteOriginally posted by: outrunCan you time the performance compared to simple non-parallel code?Suppose the parallel version is slower, would that answer your question?That's a good test. But the arrays would need to be large enough (start/stop threads).Practical neuebie question: where do I buy CUDA boards and can you plug them into a desktop?
 
User avatar
Cuchulainn
Posts: 62394
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

sample cuda problems in finance

April 20th, 2015, 7:07 am

QuoteOriginally posted by: outrunNo, I'm talking about your parallel code, vs your sequential code. The reason you use AMP is because your boss told you to speed up your code, no? Now report to your boss how good you are doing your job!;-)A lot of code is sequential. So, boss she say get it faster and quick,. Help!1. AMP 100 times faster, learning curve OK. 2. CUDA 200 times faster, we need AlexHero, yes? 3. Buy one of those fancy Intel boxes?
Last edited by Cuchulainn on April 19th, 2015, 10:00 pm, edited 1 time in total.
 
User avatar
spursfan
Posts: 893
Joined: October 7th, 2001, 3:43 pm

sample cuda problems in finance

April 20th, 2015, 7:16 am

The answer to your question is typically yes - but you might have to an adapter for £5 or so to link the power connectorsHere's what Mike Giles currently recommendshttp://www.oerc.ox.ac.uk/projects/cuda-centre- ... sDepending on your budget, you might stretch to high-end NVIDIA consumer/gaming cards such as the GTX 980
 
User avatar
Cuchulainn
Posts: 62394
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

sample cuda problems in finance

April 20th, 2015, 8:07 am

QuoteI strongly suspect that your code rewrite actually *worsen* the performance. if so then it clearly demonstrates that you need hardware knowledge before you can be an effective parallel coder. This sounds like the early days of computing in that you need to know the hardware. And the boss might that 200 times faster is good enough even though 400 times faster with more effort. A GPU h/w box is much less expensive than a developer.Is C++ AMP optimized for hardware?
 
User avatar
Cuchulainn
Posts: 62394
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

sample cuda problems in finance

April 20th, 2015, 10:12 am

Is C++ AMP optimized for hardware? In a few years this GPU will become a commodity product and Microsoft and others will hide all the hardware aspects under a wrapper.
Last edited by Cuchulainn on April 19th, 2015, 10:00 pm, edited 1 time in total.
ABOUT WILMOTT

PW by JB

Wilmott.com has been "Serving the Quantitative Finance Community" since 2001. Continued...


Twitter LinkedIn Instagram

JOBS BOARD

JOBS BOARD

Looking for a quant job, risk, algo trading,...? Browse jobs here...


GZIP: On