Intel Unveils More Xeon Phi Co-processors

Prunes - June 21, 2013 03:37PM in CPU's

I do not know how much you are into supercomputers, but often they use some quite different processors than what is found in consumer PCs. Some of Intel's supercomputer processors are called the Xeon Phi. At the International Supercomputing Conference in Leipzig, Germany, Intel announced some new additions to the Xeon Phi family. Last November, Intel announced 5100 series, which is meant for dense servers. The 5100 series has now been expanded upon with new one called the 5120D. The 5100 series differs from the 7100 and the 3100 by being produced in a different form factor. The 5100 series is shipped as a PCB with the co-processor installed and with 8 GB of on-board GDDR5 memory. The 60 cores inside the co-processor are clocked at 1.05 GHz, and have 30 MB of L2 cache. The PCB uses a 230-pin edge connector, which fits in a x24 PCIe slot. The PCB does not come with a thermal solution, and it is therefore up to the buyer to provide sufficient cooling for the 245 W TDP. The co-processor has a peak performance of 1,011 GFLOPS.

The 7100 series, Intel's high performance co-processors, will be implemented as PCIe cards that house the co-processor and the on-board GDDR 5 memory. The cards have 61 cores that operates at 1.283 GHz. The clock frequency can be brought up to 1.33 GHz via the Turbo Boost feature. The cards have 30.5 MB of L2 cache and 16 GB of GDDR 5 memory. The co-processors' peak performance comes in at over 1.2 TFLOPS. The difference between the two cards is the cooling solution used to manage the 300 W TDP. The 7120P uses passive cooling, while the 7120X uses active cooling.

Intel’s answer to "value" supercomputer co-processors is the 3100 series. The 3100 series are much like the 7100 series; the differences are in the specifications. The 3100 series has only 57 cores as opposed to the 7100 series' 61 cores, and the clock speed has been reduced to 1.1 GHz along with the Turbo Boost feature being disabled. The amount of L2 cache has also been reduced to 28.5 MB, and the on-board memory has been reduced to 6 GB. The co-processors' peak performance is 1,003 GFLOPS. The difference between the two models is the cooling solution. Like the 7100 series, the 3120P uses passive cooling, while the 3120A uses active cooling.

Model Cores Threads Frequency Turbo frequency L2 cache On-board memory TDP Price
3120A   57 228 1.1 GHz N/A 28.5 MB 6 GB 300 Watt $1695
3120P   57 228 1.1 GHz N/A 28.5 MB 6GB 300 Watt $1695
5120D   60 240 1.05 GHz N/A 30 MB 8 GB 245 Watt $2759
7120P   61 244 1.23 GHz 1.33 GHz 30.5 MB 16 GB 300 Watt $4129
7120X   61 244 1.23 GHz 1.33 GHz 30.5 MB 16 GB 300 Watt $4129
 
Source: CPU World