If the future ATI/AMD GPU RV 770 will be unveil officially unveil in 10 days, the company decided to communicate about on of its new card, the Firestream 9250. It does not offer any video output, as this card is a stream processor specifically designed to accelerate algorithms in high-performance computing (HPC), mainstream and consumer applications. It will directly compete with NVidia's GeForce GTX 2x0.
The FireStream 9250 breaks the one teraflop barrier for single precision performance computing. Interestingly, AMD also decided to communicate about the performance/watt efficiency ratio, and the card can deliver up to eight gigaflops per watt.
The FireStream 9250 gets installed on a single PCI slot and sports 1 GB of DDR3. The card is expected to ship for Q3 2008 and should cost around 999 USD.
Interestingly for Mac users, AMD indicated:
In keeping with its open systems philosophy, AMD has also joined the Khronos Compute Working Group. This working group’s goals include developing industry standards for data parallel programming and working with proposed specifications like OpenCL. The OpenCL specification can help provide developers with an easy path to development across multiple platforms.
Orange announced to have sold out all iPhone rev1. They should within the next days start communicating about their forthcoming iPhone 3G-dedicated offers. Orange could also release an iPhone upgrade program to its iPhone customers to allow them to migrate from the Rev1 to the rev2 version. Indeed, bringing 3G support will also open the door to financially interesting service for the French carrier.
Whereas the manufacturers of Flash NAND memory hoped that the market of their product would again rise, it has just undergone a new fall, reaching a new low price. A chip of 1 GB of memory costs no more than $2,35 while that of 4 GB is only $4.
Once again, the manufacturers seem to be misled by their estimates. They supposed that the market of SSD disks, the large-scale consumer potential for these chips, was going to take off while that is still not the case.
Even though the chipmakers are morose, the consumers profit from products that are less expensive. But the first to suffer are still the integrators of these chips, manufacturers of USB keys or other devices of low technicality. They have not only the bad luck to work in this saturated market, but they have also to compete with the manufacturers of these chips who start to sell end products in order to recover a bit of the margin taken by their former customers.
NVidia raised the veil on its new Geforce series GTX 2x0. Even if one knows almost nothing about this series, the figures remain impressive. The chip has 1,4 billion transistors where 80% are dedicated to 3D calculations, it functions with GDR on a 512 bits bus with a bandwidth that can reach 142 GB/s. ;
Two models are already announced:
Geforce GTX 260:
- 192 calculating units
- Frequencies of 576/1242 MHz
- 896 MB of DDR3 with 1 GHz on a 448 bits memory bus
Geforce GTX 280:
- 240 calculating units
- Frequencies of 602/1296 MHz
- 1 GB of DDR3 with 1,1 GHz on a 512 bit memory bus
Most interesting is certainly the sales strategy of NVidia. The company does not seek to center its sales pitch on the power of the graphics card, but instead on the supercomputer that it represents. They speak primarily of acceleration using CUDA for video encoding and other calculations.
And obviously, they do not lie since
PCiNpact has tested the card. With a customer Folding Home optimized to benefit from the card, they succeeded in making 5000 points per day where a PS3 reaches only 900 and one old Pentium E two cores 600 points.
Difficult to speak of a revolution, in the past we already knew aboutt coprocessors, however this record breaking entry of NVidia into this field, a thing that was inconceivable only one year ago, is likely to to cause a big change in our data-processing environment.
PS:
Hardware.fr has put online a complete test of the card. It confirms that it is as powerful as another card equipped with two GeForce 9800 chips. It remain to be seen how ATI/AMD will answer this with its next Radeon 48x0.
.
Roughlydrafted has published a very interesting interview with Bertrand Serlet. An aspect particularly drew our attention, Open CL. This system will oversee the computations and route some calculations to be carried out by the graphics chips and thus accelerating many processes.
According to this interview, this collection of APIs will be independent of the architecture of the video cards and will compile on-the-fly the instructions according to the chip that is fitted inside the machine.
Beyond the acceleration of video encoding and other mathematical calculations, one could imagine that Apple could also use this process for Open GL and can thus develop universal drivers for all the video cards.
Thus, they would have finally become masters of the situation and have graphic drivers able to compete with those of NVidia and AMD, who take time to polish the drivers that run under Windows.
So in the absolute sense, this business could be fragile and potentially a big waste of time. However it is necessary to keep in mind that the trend in architecture is to have more and more cores that are often under exploited and which would have time to spare for these video encoding. It appears that the goal of Snow Leopard is to maximize the use of all the cores.
