Australian e-tailer Austin Computers has already begun
listing two future NVIDIA stock keeping units (SKUs) - which obviously - are yet to arrive. The first listing is that of a GeForce GTX 280+, which preliminary specifications show could be a 55nm variant of the same GeForce GTX 280. Looking at how NVIDIA dealt with the 9800 GTX+, it could be assumed that the new GTX 280+ could feature higher clock speeds in order to make it more competitive.
Second to be
listed, which looks rather surprising, is that of a GeForce GTX 350, based on the GT300 graphics processor. Again, whatever little specifications listed, show that the card is based on the 55nm silicon fab. process and holds 2 GB of GDDR5 memory across a 512-bit wide memory bus. It is mentioned that the product could be available any time in Q4 2008. Is NVIDIA gearing up for X'mas?
33 Comments on GT200(b), GT300 SKUs Make for Early Sighting
I cant see them making a new card with 512 bus 2gigs of GDDR5, it has to be a duel card like the 9800GX2.
But seriously, I think it must be a misprint, as the GT300 is probably the GT2xx or something.
The only problem I have with this info is that its fud at best (GTX 350). Why would NV suddenly break the mold and use a ton of super expensive memory on a single card. Whatever, I just hopes it costs a reasonable amount. They don't need to go through anymore GTX200 series fiascos.
I doudt that it is.
Logic says it's possible it could be a dual chip varient on the 55nm scale, with a 256bit (or close) bus as apposed to the 512bit bus. This would make sense for the GDDR5 use and its amount.
If that is the case, then the GTX280+ could have a shrink in the die with a changed memory bus in order to allow for the core to be scaled down (and save more money just like the 9800 GTX vs 8800 GTX).
But then again I could easily be wrong.
NVIDIA Tesla GPU Computing Processor, a dedicated computing board that scales to multiple Tesla GPUs inside a single PC or workstation. The Tesla GPU features 128 parallel processors, and delivers up to 518 gigaflops of parallel computation. The GPU Computing processor can be used in existing systems partnered with high-performance CPUs.
I mean really what more would you ever need?
For this Tesla BS - you mean theoretical peak performance : "to sustain half a teraflop and just sending out one 32-bit single precision FP number per flop to the memory - not even counting the reads for the operands - would take two terabytes/s sustained memory speed. GPUs don't have nearly as much cache or other memory on chip as typical CPUs. Tesla offers DP throughput at 1/8 of the SP peak".
In real world applications Tesla will be compatible with two Intel quad core CPUs. And how many real world CUDA coded application are available at this moment?
Sorry, not in a very speculative mood today :o
The inherent inefficiency is another thing, but seing how GPUs increase their raw power much much faster than CPUs, while staying far superior in peak performance per watt, I bet it won't be a problem. FASTRA (the $4000 supercomputer made with 4x9800GX2) was a very promising solution that has had good acceptance on some circles. I mean there is a place for CUDA and TESLA and the like solutions, even though they might not be as flexible as CPU based supercomputers, cheap specialised solutions is the future for most areas. Why would you want a supercomputer to be able to perform 10 different complex tasks at the same time and cost you millions of dollars+extreme cooling+maintenace, if you can have 10 cheap machines that perform better and for a fraction of the money and will be far more accesible to the researchers?
BTW TESLA on many real world aplications (like tomography, F@H) can be MUCH (10x, 20x, 100x?) faster than two Quads. The aforementioned FASTRA "supercomputer" is indeed faster than a 256 node 3.5 million euro blade supercomputer on certain tasks. And that was based on a pure GPU architecture with no "compromises" in favor of GPU computing.
www.dvhardware.net/article27538.html
I really hope Nvidia does the die shrink and the DDR5 memory. That would make for some very nice performance increases I would think. They already OC well with the current die and often have good memory OCing as well.
I will be watching this for an upgrade path through EVGA ; )
I hope ATI will have something to compete. AMD needs everything it can muster to stay afloat right now : (
[AMD and Nvidia fan, no ati for me.]
This would effectively mean
GT300>X2
280+>4870 1gb
etc.