NVIDIA: Beyond G80...
- Thread starter kyetech
- Start date
Well, at least we'll see basically every site test whether the GTX is bottlenecked by bandwidth, as the Ultra seems to have a 13% higher core clock (and presumably similarly faster shader clock) and a 20% higher memory clock. We just check if the framerate gains fall above or below 13%.
Of course, if we see gains like Hanners saw with his GTS, then nuts to analysis.
Edit: Good catch, INKster. I'm just amazed they're still using GDDR3 at 1GHz. I thought that was already GDDR4 territory.
Of course, if we see gains like Hanners saw with his GTS, then nuts to analysis.
Edit: Good catch, INKster. I'm just amazed they're still using GDDR3 at 1GHz. I thought that was already GDDR4 territory.
Well, at least we'll see basically every site test whether the GTX is bottlenecked by bandwidth, as the Ultra seems to have a 13% higher core clock (and presumably similarly faster shader clock) and a 20% higher memory clock. We just check if the framerate gains fall above or below 13%.
Of course, if we see gains like Hanners saw with his GTS, then nuts to analysis.
Personally, it's an amazing coincidence that the rumored R600 XT bandwidth is nearly the same as the rumored 8800 Ultra:
R600 XT -> (512 x 1600) / 8 = 102.4GB/s
G80 Ultra -> (384 x 2160) / 8 = 103.7GB/s
Perhaps... too much of a coincidence ?
Jawed
Legend
Yep, I'm still waiting to see an explanation for that:Of course, if we see gains like Hanners saw with his GTS, then nuts to analysis.
http://forum.beyond3d.com/showpost.php?p=971673&postcount=47
The extremely high fill-rates and unified architecture might be, jointly, doing nice things...
Jawed
Personally, it's an amazing coincidence that the rumored R600 XT bandwidth is nearly the same as the rumored 8800 Ultra:
R600 XT -> (512 x 1600) / 8 = 102.4GB/s
G80 Ultra -> (384 x 2160) / 8 = 103.7GB/s
Perhaps... too much of a coincidence ?
Last time I checked the R600 XT had 105.6GB/s = (512 x 1650) / 8
icecold1983
Banned
so nvidias refresh is a 25 mhz core speed increase over whats available today. so no real refresh from nvidia, no high end part from ati. YAY
Umm, what?650-575=75. The scalar units should be clocked at least at 1520(assuming the same ratio as in the GTX is employed, and they`re not pushing them higher). If you`re arguing(in a fairly tiring way, as I`ve gathered from most of your posts) about the delta from factory overclocked GTXs, what makes you think there aren`t going to be factory overclocked Ultras?A hunch?Could we leave those out please?
And if you think the current(and near future like the R600) crop of cards are crap, simply don`t buy them. TADDA, a solution to your woes.
And if you think the current(and near future like the R600) crop of cards are crap, simply don`t buy them. TADDA, a solution to your woes.
Unknown Soldier
Veteran
One small factor might be that if G8x latency tolerance isn't ridiculously high, GDDR4's higher latency might actually hurt some workloads slightly.
The solution to that, of course, is to increase the size of the register file relative to the number of ROPs and TMUs. And one way to do that is to simply increase the number of ALUs (and scale the register files accordingly, as ATI did between R520 and R580). That's yet another reason to believe G9x will at least double the ALU ratio, IMO.
If you read Icecolds post he says 25Mhz over what's available today, pointing at the 620Mhz OC models available.
And if you read mine, I suggest that it means jack-squat that OC models are available on the market today, as one could just as easily extrapolate that Ultra models will have their own factory OCed versions with the same delta from nVs suggested clock rates, thus nullyfying his base argument. Wanking can go on endleslly from here, along the lines of:but the Ultra won`t have that much headroom yadda yadda. Fact is we don`t know, and fact is that people expect waaaaay to much from these pieces of silicon.
Keeping costs low I think.
jimmyjames123
Regular
Maybe there will be 8700 GTS after all I think it is less confusing to put a 64sp version as 8700 instead of 8800. One way or another, they need something to fill the huge performance/architectural gap between 8600 GTS 256MB and 8800 GTS 320MB.
I think it is less confusing to put a 64sp version as 8700 instead of 8800. One way or another, they need something to fill the huge performance/architectural gap between 8600 GTS 256MB and 8800 GTS 320MB.I believe it was a mistake, they wouldn't want to spoil the market for 8600GTS by telling something better is coming.
But obviously there is, not in a month or two perhaps but it's coming. Looking at the specs that huge gap between G84 and G80 is waiting only to be filled.