|
|
 |
|
|
|
 A Dutch e-trader is currently offering Jetway brand Radeon HD 2600s and 2400s at quite interesting prices::
Radeon HD 2600PRO 256 MB DDR2 PCI-e DVI TVout 69 €
Radeon HD 2600PRO 512 MB DDR2 PCI-e DVI TVout 89 €
Radeon HD 2600XT 256 MB DDR3 PCI-e DVI TVout 99 €
Radeon HD 2600XT 512 MB DDR3 PCI-e DVI TVout 149 €
Radeon HD 2600XT 512 MB DDR4 PCI-e DVI TVout 179 €
Radeon HD 2400PRO 256MB DDR2 PCI-e DVI TVout 44,95 €
Radeon HD 2400PRO 512MB DDR2 PCI-e DVI TVout 54,95 €
Radeon HD 2400XT 256MB DDR3 PCI-e DVI TVout 59,95 €
For comparison, the GeForce 8600 GT is found for around 120€, the aging Radeon X1950 Pro 512MB for 140€, while the 8600 GTS starts at 170€. Going down a level, the 8500 GT is around 75€ and Nvidia wants 55€ for the 8400 GS.
Of course, you will have to wait to know the performances of the cards before making any sure decisions. However, even if they aren’t bad, the fact that AMD’s graphics division is offering the Radeon HD 2600XT 256 MB below the symbolic 100€ mark will bring some fresh air on the entry level, which needs it most considering the performances in this price range. |
| |
|
 According to our colleagues at VR-Zone Creative has confirmed its desired to commercialize X-Fi chips through partnerships similar to the one already well underway with Auzentech. They will concentrate more on audio processors than complete solutions. Discussions are currently in progress with other potentially interested parties. In fact, the creator of the Sound Blaster supplies its clients with a reference design kit and helps them with the conception and improvement of their product.
In terms of development, Creative says it will continue to work on improving EAX support via ALchemy as well as audio compression, even if the latter always implies losses. We also note that the company wants to extend its activities into new markets, notably car stereos. Finally, while the X-Fi Xtreme Audio PCI-E was recently spotted in Akibahara, their chips contain a bridge that handles the conversion of PCI signals into PCI-E resulting in higher latencies. An X-Fi chip natively supporting PCI-E is expected out at the end of 2007. |
| |
|
 OCZ has announced its DDR3 PC3-10666 Platinum Dual Channel series certified to support latencies of 7-7-7-20 in DDR3-1333 with a voltage of 1.8 V. The two kits that make up the line differ in size as the OCZ3P13331GK will have two 512 MB memory modules while the OCZ3P13332GK will have two of 1GB. Their prices are not yet known.
Note that Kingston already offers a similar kit, the KHX11000D3LLK2/2G, which is certified for DDR3-1375 at 1.7V in 7-7-7-20. |
| |
|
 The transition to the 1333 MHz FSB , notably supported by the recent Bearlake chipsets, is on its way and at the end of the year, Intel will stop producing at least five of its Core 2 Duo processors using the 1066 MHz FSB, which are the E6300, E6320, E6420, E6600, and E6700. The delivery of the last supplies of these processors will be expected in the first quarter of 2008.
You may remember that they will partly be replaced by other Core 2 Duo processors that will be officially introduced with July 22 price decreases. For those who want to hold onto their motherboard limited to 1066MHz FSB support, the Core 2 Q6600 will be the only possible upgrade. |
| |
|
 For several days now the price of DDR2 has climbed. Since our last report on June 12, the 64 MB DDR2-667 chip has increased from $1.79 to $2.36, or a 32% gain. On the contrary, the price of DDR-400 is decreasing, which has meant that the two are getting closer because it is now at $2.66, or a 2.9% drop. |
| |
|
 It was today that Nvidia decided to unveil a new line of products, the Tesla. After the GeForce destined for the general public and gamers, the Quadro line intended for image professionals, the Tesla attacks the market in calculation power.
Nvidia thus has concretized its initiative launched with the GeForce 8800 and CUDA, the component that allows using the GPU as a mathematical coprocessor through C language extensions. AMD, via ATI, was the first to announce such an interface with the CTM, using assembler programming language, and was also the first to announce a specific product for this market with the Stream Processor. Now that we heard the announcements and seen AMD’s demos, it would be nice to see something more concrete. Everything has been reserved for a handful of developers, CTM isn’t public and the innovations announced with the Radeon HD 2000 are still in development
On its side, Nvidia has published a beta version of CUDA (0.8) in February and supplied us with a more advanced version (the 0.9) before the release of the 1.0, which is due out next week. We can attest that the software component has strongly evolved. Otherwise, we were able to see a number of real GPU uses, which were demonstrated by different companies during the Tesla presentation a few weeks ago. This is proof that CUDA is now really usable.
For now, Nvidia has announced 3 products. The first, the Tesla C870, is like a GeForce 8800 GTX without the video output and therefore meant to only serve as an accelerator. Other than that, the card has a video memory of 1.5 GB instead of 768 MB. Its price is fixed at $1299, which is reasonable, because a Quadro FX 5600 also with 1.5 GB of memory costs $2999. The TDP is 170W.
 The second in the Tesla line is the D870, which uses the concept of the Quadro Plex. Two Tesla C870 cards are found in an external casing, which is connected to the PC via a special PCI express card with an adapted cable. The TDP is now 350W and the price jumps to $7500, which in spite of everything is still a « good deal » compared to the equivalent Quadro Plex offered at $17500. Two of these casings can fit into a bay and together occupy 3U.  Finally, the third product in the line is a 1U rack, the Tesla S870, equipped with no less than four Tesla C870s or four G80s and 6 GB of video memory in total. The rack is connected to a main system also in PCI express and already ready for PCI Express 2.0. Transfers between the CPU and GPUs will therefore be boosted. The TDP is 800W, although Nvidia announces a consumption that in practice is generally around 550W. This 1U rack is commercially available for a price of 12000$.  Nvidia is now ready to try and attack the calculation power market and solidify its advance on AMD in the software area. As for the long term strategy for CUDA, Nvidia assures us that it will be available on all product lines from the Quadro to GeForce and won’t just be reserved to the Tesla. CUDA should also soon be an integral part of general public drivers. However, in the future, some CUDA functions or upcoming GPUs could be reserved for the Tesla. This is notably the case (and maybe some of the high end Quadros) for the 64 bit calculation precision on floating numbers, which will be introduced with the G92. More details on this in a small article published in the near future… |
| |
|
|
Copyright © 1997- Hardware.fr SARL. All rights reserved.
Read our privacy guidelines.
|
|
