General Discussion Undecided where to post - do it here. |
Reply to Thread New Thread |
![]() |
#21 |
|
|
![]() |
![]() |
#24 |
|
then afterwards, he confessed. so i gave him a copy of Ocean's Eleven (12, and 13) for his birthday. too bad Greenspan doesn't confess. |
![]() |
![]() |
#25 |
|
|
![]() |
![]() |
#26 |
|
|
![]() |
![]() |
#27 |
|
Super computer or not, it still comes down to "garbage in, garbage out" |
![]() |
![]() |
#28 |
|
|
![]() |
![]() |
#29 |
|
|
![]() |
![]() |
#32 |
|
If you got an extra $1000 for the member fee plus another $20k for an nvidia based CUDA personal supercomputer you can join an elite financial forecasting group. My name is Phil Parker and I'm the Director of Global Marketing for ELEKS Software. It is true that ELEKS does develop custom CUDA-accelerated applications for advance simulations. I'm not sure I would put much merit into this rumor stated above considering that ELEKS does not offer off the shelf CUDA software for this particular simulations. For info on our CUDA efforts, (click here to view ELEKS' CUDA Development page). And I wouldn't take this as very factual considering that the customers we have do not allow us to mention their names and they are keeping quite as to what they do with these CUDA systems. I would also like to add that the writer says that we have developed a cloud array of 10,000 CUDA desktops that have more than the processing power of the U.S. Weather Service. This is partially correct, but I'll expand on this below. ELEKS Software, based Henderson, NV (low state tax rate), has close to 500 very bright developers performing world class software development projects - most of which are based in our development center in Lviv, Ukraine (a 750 year old UNESCO World Heritage site). Most of our work in the CUDA arena has been to move calculation intensive work off of CPU based system over to run on much more efficient low cost CUDA enabled Personal SuperComputers - often called High Performance Computing (HPC) systems. For this system mentioned above, very advanced simulations were once done with data sets that had to be reduce in size to allow for calculations to be performed once per evening. This was as good as it got - until recently. By using the ELEKS developed CUDA accelerated application, larger, more complex data sets can be used with Monte Carlo simulations for improved accuracy. Instead of results being calculated once per night, results are now obtained in minutes. So why don't you not hear more about ELEKS? Because they know they have a silver bullet as part of their arsenal - and you don't. Now - if I may, I'd like to share what makes this possible - if nothing more than to educate each of you about the state of the art in CUDA development. Many of today's most powerful corporate systems have multi-core CPU processors from Intel powering them - lots of them to be exact. They tend to run very hot and require a lot of power and cooling capabilities. They are not very environmentally friendly systems. These CPUs are now equipped with 2 to as many as 8 core processors per CPU. They tend to run multi-tasking programs very well. They are however, not well suited for raw numbers crunching like that required for mathematically intensive 3d graphics. Graphics Processor Units ( GPUs) on the other hand are very good at number crouching. Today's GPU processors have has many as 448 internal core processors per GPU PC cards have multiple GPUs per PC card for true Massively Parallel Processing. A few years ago, Nvidia decided that there was a market that could use this processing power. To expand the market for this processing potential, Nvidia create CUDA - an advanced programming language that would allow developers to harness all of these core processors to do more than just graphics. In this thread, it was noted that ELEKS used 10,000 CUDA desktops. This was overstated. The fact is that it does not take lots of desktops, but rather, the power is in the number of total GPU cores. A typical system could consist of 35 rack mounted blade servers. Each server consists of 2 GPUs, with each GPU containing 448 cores. The rack would include 8 CPU cores to manage where data will be calculated. This is literally a total of 31,640 cores processors. Should additional processing power be needed, additional CUDA-enabled GPU cards or servers equipped with GPU cards can be added to the system for a scalable Personal Supercomputing system. These system drastically reduce the cost for complex computing, while at the same time, slashing the cost for power and cooling as compared to a traditional high end computing environment. ![]() I hope that clarifies what ELEKS involvement was, what this system did and what it didn't do. I assure you that one of these systems did not attempt to predict the total collapse of the financial system on May 15 and fail. I can not however, be more specific with what our customers are doing with these systems. Thank you for your time. Phil Parker ELEKS |
![]() |
Reply to Thread New Thread |
Currently Active Users Viewing This Thread: 1 (0 members and 1 guests) | |
|