Tuesday, February 3rd 2009
RV790 Could Hold Higher Core Clock Speeds
Conceived to be a successor to the RV770 graphics processor, the RV790 emerged out of various media-leaks as the next big thing for ATI, in its never-ending pursuit for market leadership. Information on this mysterious new GPU has been trickling in at leisure, and it has been shaping up to be a current-generation GPU built using newer manufacturing technologies that could make it cheaper to produce. What we do know about RV790XT based engineering samples so far, is that they carry faster memory chips that are rated to operate a higher frequencies, 1 GHz to be precise, but specified by AMD to run at 975 MHz.
Fresh details sourced by Hardware-Infos from internal sources point out to the possibility of AMD stepping up the core clock speeds by as much as 100 MHz, taking it to 850 MHz. The ramp in clock speeds however, come at the expense of increasing core voltage to 1.3 Volts. There have been several reports speculating the core voltage so far, on of which points at 1.23V for 750 MHz. Another grey-area for now is the RV790's shader domain, with a certain level of ambiguity surrounding stream processor count. All of this leads us to believe that AMD hasn't finalized the specifications of RV790 and could be racking its brains on this one. AMD is expected to make critical announcements on its new GPUs in March and might choose the CeBIT event as the ideal platform.
Source:
Hardware-Infos
Fresh details sourced by Hardware-Infos from internal sources point out to the possibility of AMD stepping up the core clock speeds by as much as 100 MHz, taking it to 850 MHz. The ramp in clock speeds however, come at the expense of increasing core voltage to 1.3 Volts. There have been several reports speculating the core voltage so far, on of which points at 1.23V for 750 MHz. Another grey-area for now is the RV790's shader domain, with a certain level of ambiguity surrounding stream processor count. All of this leads us to believe that AMD hasn't finalized the specifications of RV790 and could be racking its brains on this one. AMD is expected to make critical announcements on its new GPUs in March and might choose the CeBIT event as the ideal platform.
13 Comments on RV790 Could Hold Higher Core Clock Speeds
Source:
www.fudzilla.com/index.php?option=com_content&task=view&id=11821&Itemid=1
EX: GTX260: 28 ROPs x 576mhz = 16.1 gp/s
64 TMUs x 576mhz = 36.8 gt/s
4970: 16x850mhz = 13.6 gp/s
40-something TMUs x 850: 34 gt/s (40 tmus) or 37.4 (44 tmus) or 40.8 (48 tmus)
While I agree, an increase in ROPs sounds nice, I would argue overall their method is more efficient, and tying the shader domains surely saves some transistors. I believe it's tied to the memory interface as well for both companies, as 4 ROPs are used proportional to each 64-bit part of the memory controller, and 2 ROPs per 32-bit controller on Nvidia's products. I believe this will double in the coming generation, ie rv870 may have 32 ROPs and be 256-bit, just as RV740 is 128-bit and 16 ROPs. While slightly overdue, the leap in computational power surely justifies such a change. ATi was late to the party with jumping up the tmu number, and I choose to believe ROPs are next on the platter.AMD's partners will all be at CeBit. Why spend money (you REALLY REALLY don't have) advertising at a booth, when companies that depend on yours will do it for you?Dimes to donuts they replace the old price bracket for the 4800 series: $199 and $299.
New crackpot projection: Rv790 is 880sp/44 tmus; 176x5; 11 arrays. 850x1760 = ~1.5 Gflops, a nice 25% gain over 4870. Additionally, 700x1760 is ~1.25Glops, a nice 25% over 4850.
4970:
880sp
16 ROPs (13.6 gp/s)
44 TMUs (37.4 gt/s)
256-bit
1GB
850/3900 (1.5TF, 125 Gbp/s)
GTX 260 (216):
216sp
28 ROPs (16.1 gp/s)
72 TMUs (41 gt/s)
448-bit
896MB
576/2000 (805 Gflops, 112 Gbp/s)
I think it will perform in-between the 216 and the gtx280, perhaps squarely so...and yeah, you can blame it on the ROPs. :mad:
At $300, the MIR 280's would still be the way to go. I think the only way ATi pulls an upset, and totally fooks Nvidia is if the thing turns out to be 960sp/48 tmus. If so, it at worst would go tit-for-tat with the 280, all the while being a cheaper chip to make, and a similarly priced chip at market...ie turning GT200 into a massive FAIL.