Cell accelerator boards, NVIDIA GPU servers and HPC

This is pretty interesting considering your basic CPU does something like 30GFLOPS (something around 16 GFLOPs per POWER 6 cores, 10GFLOPS for a Itanium cores). A cell board like this does 180GFLOPs.
(Don't take this is a benchmark or anything. This is just some RAW data).

Some NVDIA something like 500 (technically the G80 has 128 fp32 ALUs @ 1350MHz with MADD - about 350 GFLOPs), a R600 is supposed to have like 500 and a Realizm 800 (Dual Wildcat VPUs) about 700 GFLOPS :-). So yeah, with 16 or so of these cards used right, you could score yourself a place on TOP500 SuperComputers. "Hey, my 4 graphic stations can beat your 1000-node Xeon cluster!".

And this is no joke, since GF8 series and the whole NVIDIA CUDA thing, NVIDIA has also started making... erm.. servers.

NVIDIA Tesla S870 GPU computing system peaks something like 2TFLOPS.

While one of those "low powered MIPS 64 CPU's" in the SiCortex, about 1GFLOP :-). But they have clusters of up to 5832.

PCI-E Cell accelerator board:

• Cell BE processor at 2.8 GHz
• More than 180 GFLOPS in PCI Express accelerator card
• PCI Express x16 interface with raw data rate of 4 GB/s in each direction
• Gigabit Ethernet interface
• 1-GB XDR DRAM, 2 channels each, 512 MB
• 4 GB DDR2, 2 channels each, 2 GB
• Optional MultiCore Plus™ SDK software

A WildCat 800:



There's an awesome potential HPC market here... GPUs, Playstation 3s with Cells, Cell PCI-E cards... exploited properly, it can make some pretty fast clusters. See Folding@Home for example where where GPUs count for 58.3 and PS3's count for 18.1 average computations per client.

6000 CPU Linux Cluster in a single 20KW machine

"SiCortex 5832 is a 5-teraflop single-unit supercomputer."
Bet that got your attention.

"It uses low-power, custom 64-bit MIPS-processor packages, which are basically entire computers on a single chip. 5832 processor cores and 8TB of RAM in one chassis, which draws less than 20 kilowatts of power."



"The SiCortex systems are completely open source, even down to the microcode."

They even say it runs a modified version of *cough* Gentoo Linux and the (now Sun) Lustre Filesystem.

SiCortex also offers a 72 CPU desktop machine that's as big.. well, as a desktop machine :-).

Say hell yes to Desktop HPC:



Read more about their Kautz digraph based fabric and implementation here:
http://www.sicortex.com/products/white_papers

GPU accelerated password cracking

Modern versions of Windows like Vista no longer use LM (which could be cracked in minutes using rainbow tables and tools like Ophcrack, especially since it grouped passwords in 7 character pieces). NTLM tends to be more difficult to crack.

Now that GPU HPC clustering has been around for a while, it was only a matter of time until someone implemented the concept for usage in password cracking. I've seen specialized FPGA based machines like Copacobana accelerate DES cracking for example. Or custom chip machins like Deep Crack. But there are cheaper and more effective resources available, right on your home desktop!

A simple GeForce 8 card can have 128 stream processing units and they are very suitable for fixed point arithmetics.

Elcomsoft has released a product that uses such a technology: Elcomsoft Distributed Password Recovery 2. And it will soon incorporate the technology into all their products.

What does this mean? It means that cracking a typical 8 character NTLM has can take as little as 4-5 days instead of months. Using a GPU means the process can be 25 times faster than normal! Not to mention you can just use 4 GPU's on your machine to really speed things up :-).

Guess it's time to enforce a more strict password policy. 12 or more characters should be a minimum now.

Want to use GPU clustering for other kind of HPC applications? Grab the NVIDIA CUDA toolkit and an MPI manual, and start coding. Or maybe you're an ATI person? Then take a look at Folding@Home on ATI GPU's page.