From what is posted, they will employ the same strategy as the HD 3800 series, where a dual GPU card will be the flagship.
Take this only as a rumor though.
Since the RV670 did about 528Gflops at 825MHz, this would mean that the new architecture isn't particularly tweaked, since these numbers account only for theoretical throughput. For real world performance, these figures don't allow for a precise estimation of performance.
Now, take 528 and multiply it by the 50% increase in available shader units, and you get 792GFlops. Multiply that by 1050/825 - the increase in clockspeed - and you get 1008Gflops, which looks awfully familiar, and so does the 256bit wide memory bus.
If this is indeed the form of the R700 architecture, it seems nothing more than a tweaked RV670, which was a tweaked R600. Also, the "V" in RV770, has stood for value - performance-mainstream - cards for some time now, until the RV670 showed up; not a groundbreaking card, as we know.
This would be something as readily achievable as they would need to come early with a new product to the market.
Of course that if Nvidia pulls another G80 "revolution" out of it's hat, ATI won't be able to compete for who knows how long. Taken that the HD 3870X2 scales relatively well in some cases; a single GPU card from Nvidia with close to the same 2TFLOPS will wipe out the floor with the RV770X2 .
I'm a fan of recycling stuff but for the mainstream, not the high-end. High-end needs new architectures every two years to 18 months, which bring huge amounts of performance to the table. I'm all in for double the performance from the HD3870 for $249 though.
Still, unless they have some kind of architectural improvements, that don't reflect on theoretical throughput, these are either fake specs or a not so good R700. Time will tell.
Keywords: RV770 R700 specifications details
2 comments:
The Radeon 3870 is already better than the 8800 GT for Stream Processor power.
Ask a F@H user about it.
What the RV670 lacks is Z-Pixel ( http://tinyurl.com/2co9hr ) and multi-texturing capacity ( 12400 MTex/s vs 33600 MTex/s for the 8800 GT G92 ).
That's not exactly the problem.
Although most games do need some to work very fast with Z-Buffer operations, the problem with the R600 architecture is that for the 320 stream processors to work at full efficiency, the mini-compiler in the driver must have five instruction to issue to every of the 64 sp block.
It works like some vliw processors were each vliw instruction must carry 5 instructions, for optimal performance.
Obviously, it will not always be possible to have full efficiency all the time.
Nvidia also manages to run 128 threads at the same time, while ATI's architecture, AFAIK for the RV670 also, it only manages 64, although in an optimal scenario those 64 threads can result in more work.
That coupled with the fact that nvidia clocks the shader processors higher than the core, helps mitigate the performance advantage ATI could've have.
For a more through explanation:
http://www.anandtech.com/video/showdoc.aspx?i=2988&p=6
and the next page also.
Post a Comment