Sweclockers has posted a leaked spec sheet from Palit that probably has the final specifications of nVIDIA’s next release, the GTX 570. The new card of course will based on the GF110 GPU and will come with one shader module disabled, meaning 480 CUDA cores and two memory controllers less giving a 320bit memory interface with 1280MB of GDDR5 memory onboard. It will feature clocks of 732/1464/3800MHz for the GPU, the CUDA cores, and for the memory respectively. We can expect similar performance levels as the GTX480, but with lower power consumption so the GTX 570 will need “only” two 6-pin PCIe power connectors with its TDP being 225W. On connectivity it will feature two dual-link DVI outputs and one mini-HDMI connector meeting the 1.4a standards. Embargo on the cards will be lifted on the 7th of December, so cards can be expected to arrive in e-tail the following day, putting the launch on the 8th of December.
nVIDIA seems to be preparing a refresh for their mainstream line-up early next year, with cards that utilize fully enabled chips or so do the guys at Fudzilla and Expreview. The GT 530 will come with somewhat raised clocks and GDDR5 memory to avoid the the bandwidth limitations of the GDDR3 used on the GT430. The GTS 550 will probably come with bit higher clocks and all 5 shader modules enabled totaling 240 Cuda cores and with two extra memory controllers in addition to enable a 192-bit memory interface. It’s all speculation so take it with a grain of salt, but here are my bets:
16ROPs/40TMUs/240CCs @ ~830MHz
192bit- 1.5GB GDDR5 @ ~3800MHz
8ROPs/16TMUs/96CCs @ ~780MHz
128bit – 1GB GDDR5 @ ~3400MHz
nVIDIA released the first WHQL driver for the GTX 460 SE and GTX 580 cards. Apart from support for the new cards it’s pretty much the same package as the previous version, if you want the latest and greatest you can grab them bellow.
Download – WinVista / Win7 32bit | WinVista / Win7 64 bit | WinXP 32bit | WinXP 64bit
A couple of HD 6900 Series presentation slides made it out. The HD 6970 will come with 32ROPs, 96TMUs, 1920SPs @ 890MHz and 2GB of GDDR5 videomemory on a 256-bit memory interface clocked at 5GHz effective. Most importantly AMD seems to doubled the geometry throughput in the Cayman chips accompanied with some of the tessellation engine enhancements introduced in Barts would give a considerable boost in heavy tessellation scenarios. The HD 6990 (Antilles) comes with two Cayman chips onboard totaling in 64ROPs, 192TMUs, 3840SPs running at a 775MHz and a whopping 4GB of GDDR5 videomemory clocked at 4.8GHz. Idle board power is a measly 30W however the maximum TDP is 300W just fitting in the PCI-SIG high performance power envelope.
Update: Added a couple of more slides. New slides reveal active power-management to enforce the desired TDP and a new MSAA method called EQAA (Enhanced-Quality Anti-Aliasing) which is quite similar to nVIDIAs CSAA (Coverage Sample Anti-Aliasing) in terms that it uses extra coverage samples too for better AA quality while memory foot-print remains about the same.
nVIDIA seems to the have a dual-GPU board in the works to counter AMDs HD 6990. The board has two GF110 GPUs onboard and nVIDIA is probably using their NF200 chip to bridge the two GPUs. Each GPU has 12 memory chips giving a 384-bit memory interface and 1.5GB videomemory each. The card has soldering points for two 8-pin PCIe power connectors meaning that the cards TDP probably won’t fit in the PCI-SIG 300W power envelope. On the connectivity front there is place for three DVI connectors meaning it’ll be 3D Vision Surround ready.