Nvidias Next GPGPU to Have 3 Teraflop Performance

Previously we had an article about cloud computing, distributed computing and the importance of GPGPU acceleration.

HPCwire reports that Nvidia’s next-generation GPU design, the G300, may turn out to be the biggest architectural leap the graphics chip maker has ever attempted. If the early rumors are true, NVIDIA has decided move the architecture a step closer to the CPU and make GPU computing even more compelling for HPC (High power computers/Supercomputers).

According to Valich’s sources, the GT300 will offer up to 512 cores, up from 240 cores in NVIDIA’s current high-end GPU. Since the new chips will be on the 40nm process node, NVIDIA could also crank up the clock. The current Tesla GPUs are running at 1.3-1.4 GHz and deliver about 1 teraflop, single precision, and less than 100 gigaflops, double precision. Valich speculates that a 2 GHz clock could up that to 3 teraflops of single precision performance, and, because of other architectural changes, double precision performance would get an even larger boost.

In a later post Valich writes that the upcoming GPU will sport a 512-bit interface connected to GDDR5 memory. If true he says, “we are looking at memory bandwidth of 256GB/s per single GPU.”

More importantly though, NVIDIA is said to be moving from the traditional SIMD (single instruction, multiple data) GPU computing model to MIMD (multiple instruction, multiple data) or at least MIMD-like. As the name suggests, MIMD means you can run different instruction streams on different processing units in parallel. It offers a much more flexible way of doing all sorts of vector computing, and is a standard way to do technical programming on SMP machines and clusters. Presumably CUDA will incorporate MIMD extensions to support the new hardware.

There was some speculation that the GT300 would hit the streets this year, but reports of trouble with TSMC’s 40nm manufacturing technology may have slowed NVIDIA’s plans.

Nvidia also will soon have the ION. It is a system/motherboard platform that includes NVIDIA’s GeForce 9400M (MCP79) GPU and Intel’s Atom on a Pico-ITXe motherboard designed for netbook and nettop devices. In February 2009, Microsoft certified the upcoming ION-based PCs as Vista-capable. The small form factor ION-based PCs are expected to be released in the summer of 2009, starting at $299.99. The ion will have about 50 Gigaflops of performance at a netbook price.

Latest Solid State Drives and Hard Drives

Transcend SSD18M holds 128GB of flash for about $340 [at time of writing]

Corsair CMFSSD-256GB flash drive for $699

Western digital has a 4 terabyte hard drive for about $650.

Solid State Drives and Hard Drive Enterprise Combo for Database I/O

A memory system setup to handle 53,850 simultaneously online game users.

The IT group now reserves its DRAM SSD-based RamSan-400 for the part of the database that’s accessed the most and uses its newer “cached flash” RamSan-500 — which has 64 GB of DRAM cache and 2 TB of RAID-protected single-level cell (SLC) NAND flash — for the bulk of the database calls. It credits solid-state drives with helping the system to handle 53,850 simultaneous users in mid-May.

The DRAM solid-state drive-based RamSan-400 system claims to be able to attain 400,000 IOPS, for both reads and writes. Although CCP hasn’t verified that number through testing, it appeared fairly accurate based on database usage and percentage usage of the RamSan device, according to Mayes. The RamSan-500, which has both DRAM and NAND flash technology, claims read performance of 100,000 IOPS and write performance of 25,000 IOPS, according to Texas Memory Systems.

CCP is also testing a flash-only RamSan-20, which has 450 GB of SLC NAND flash storage attached via PCI Express. The RamSan-20 claims to handle 120,000 read IOPS vs. 50,000 write IOPS, illustrating the difference in performance for reads/write that users might expect to see in a dedicated SLC-based flash system.

For systems produced by Texas Memory Systems, the list price of DRAM SSD is $300 per gigabyte, while SLC-based flash SSD is $40 to $70 per gigabyte, according to Woody Hutsell, the company’s president.

The 2 TB Flash-based RamSan-500, which has 32 GB of DRAM cache, lists at $150,000, whereas pricing for the all-Flash 5 TB RamSan-620 is $220,000, Hutsell said. The latest DRAM SSD-based 512 GB RamSan-440 lists at $180,000.

Today’s list price for the RamSan-400 is $61,000, while the RamSan-20 is $18,000.

Server Farms

Apple is prepping a new US data center that may cost as much as $1bn, which would nearly double what Google typically spends on the mega data centers backing its web-based applications and services.

Google has about 35 data centers

Google data center FAQ