Saw this, NVIDA 9 series launch date....

ebeattie

Limp Gawd
Joined
Jan 29, 2007
Messages
441
http://www.digitimes.com/mobos/a20080222PD207.html

apparently, it appears the ENTIRE 9 series is a stop gap until the true next gen (duh). Whats interesting is that according to the article, the 9800GX2 IS the top line model.

If this is true, then its safe to say that these cards look like a quick way to "appear" competitive, while at the same time giving them time to finish next gen.

So, when R700 is released by ATI Nvidia will release the 9 series and lose? Think they might be playing possum by giving up the crown or will it really take them until late this year to get the next gen out? Will ATI sit on r700 to release near or after the next Nvidia chip?

It really looks like the graphics wars may be heating up again! R700 v. Nvidia's next gen. Looks like this coming Chrismas will be a good one!
 
Nvidia's slow down of new gpu releases (imo) is in direct response to ATI's lack of competition. If ATI were nipping at Nvidia's heals then the 8800GTX/Ultra's would of been EOL long since the end of 2007. With this "stop gap" gives Nvidia the time they need to further mature their true next gen gpu.

I'de be disappointed with performance numbers not equalling two 8800GTX's for their flagship GeForce 10 gpu's. But I'm really really satisfied with the longevity I'm getting with my 8800GTX's (I got one at launch and a 2nd a year later).
 
"In June, the company will launch the value-level GeForce 9500 GT based on the G96 graphics core with a reduced 128-bit memory interface. To complete the series, the GeForce 9500 GS will launch in July."

Well this sucks. I wanted the 9500GT but since I have to wait so long, I will just get the 8600GT instead.
 
i might get a GTS or the 3820x2 now and then wait till next year to spend over £300 on the next gen card that totaly blows away a ultra.

will the next gen cards still be pcie? or will pcie be dead by then?

plus how long till a quad core q9450 becomes a bottleneck in games?
 
i might get a GTS or the 3820x2 now and then wait till next year to spend over £300 on the next gen card that totaly blows away a ultra.

will the next gen cards still be pcie? or will pcie be dead by then?

plus how long till a quad core q9450 becomes a bottleneck in games?

PCI E is going to be around quite a while, I would imagine. It is very scalable and flexible. PCI E 1.1 isn't really even saturated yet and 2.0 has just come out, 3.0 specs are already being finalized.

I would give PCI Express the lifespan of PCI or better. So uh, another 10-15 years, more?
 
The rumored specs looks even worse than expected:

According to sources, 9800GTX is a product based on G92, coded as G92-420. Core/SP/Memory frequency rated at 675/1688/1100MHz. Memory interface is 256bit, equipped with 512MB 0.8ns DDR3. NV use a new designed layout, call P392. The product supports 3-Way SLI.

For your information, 9800GX2’s GPU is G92-450

The card will hit the market in the end of the market, according to Digitimes. Price still remains unknown.

http://en.expreview.com/?p=274

Hope it´s not true, but it probably is. The Gainward 1gb 8800GTS looks more interesting (less uninteresting) then this :(

http://www.fudzilla.com/index.php?option=com_content&task=view&id=5640&Itemid=34
 
Hmm...if the 9800GX2 will be the top end of the 9000 line of GPUs, it is expected the 9800GX2 will be 30% faster than the 8800Ultra. I wonder how much faster the 9800GTX will be over the 8800GTX and how much faster the 9800GT will be over 8800GTS or 8800GTs.
 
The rumored specs looks even worse than expected:

According to sources, 9800GTX is a product based on G92, coded as G92-420. Core/SP/Memory frequency rated at 675/1688/1100MHz. Memory interface is 256bit, equipped with 512MB 0.8ns DDR3. NV use a new designed layout, call P392. The product supports 3-Way SLI.

They sell faster (GPU) clocked 8800GTs and 8800GTSes (overclocked) it would pretty much be suicide if they did that, it still doesn't beat the memory bandwidth of the old GTX so it still will prob get beaten with high AA senerios, and its 15 months old. Over the past few months I have been hearing rumors of the 9800GTX being a "new" chip called D9P, but I didn't hear anything above that, only that it was going to be out with a g92 based GX2 (sounds odd)

8800GT = G92-270 112sp
8800GTS = G92-400 128sp
9800GX2 =? G92-420 ???
9800GTX =? G92-450 ???

Three things. The 9600GT that just came out seems to be an actual and competent midrange chip with 64 sps, but it performs like it has 100-110 SPs out of a G92 so that chip got an increase in IPC. I was under the asumption that this card was called D9P, the 9800GTX was called D9E, their would be some lowend part called D9M, it made sense, but apparently now that was all fud, the 9600GT is referred to as G94.

Secondly, G92-420 and G92-450 are codenames for what, it was assumed that their are only 128 pipelines in the G92, what does 420 and 450 describe? the G92-GT to GTS goes from 270 to 400 (weird) and their is only 12 sp's difference.

Third, if all of the above is true (which I don't concede) Why wouldn't nvidia use the 8800GTX/ultra pcb, the G92 is believed to be pin compatible with G80, it was good for 384bit memory bus, so despite these new cards with higher clocks, they are still getting beat by a 15 month old card because it has more memory bandwidth, so unless G92-450 is something new which is unforeseen, its still going to get edged by an 8800 ultra STILL

So the question is, why after 15 months, would nvidia use this new updated chip in a configuration which essentially restricts it when they already have which is asumed to be the same chip, in the same configuration as discribed above, slightly faster memory, but on a pcb with a slower memory bus than the previous high end. One step forward and two steps back? I don't think so, someones lying to us.
 
Taking this information to be mostly correct, it does seem that nVidia is going for cheap, decent performance as a stop-gap before the 10 series (or whatever it will be called). Correct me if I'm wrong, but I think G92 only supports 4 64-bit lanes based on its design, so 256-bit is as high as it goes. However, this is a cheaper design, and it seems nVidia attempts to make up for the lack of bandwidth by using faster RAM (RAM is cheap now, so this compensation does save cash on each board). This will put decently fast and much cheaper video cards on the market and will actually net nVidia a lot more cash. While the G80 was a monster, I think nVidia saw that the $450 and $600 MSRP of the products at launch alienated a lot of its customers. In the end, nVidia is a business, not a research facility.

What I don't understand about this trend is its inclusion with nVidia's 3-way SLi platform. Originally, the G80 series worked very well with SLi because its memory bandwidth was much higher than the capabilities of the GPU. However, because SLi stores the same graphics information on each card (if I'm wrong here, please correct me), SLI GPUs could use this bandwidth and two GPUs to a much higher potential. However, with this current set-up, one would already be taking a memory bandwidth limited card and attempting to run even more information on a limited bus, which would give much more diminished returns, or so one would think. So what happened? A new development? Better drivers? It doesn't add up, that's for sure (unless nVidia thinks the public is stupid and will just buy whatever new product/platform they put out).
 
Taking this information to be mostly correct, it does seem that nVidia is going for cheap, decent performance as a stop-gap before the 10 series (or whatever it will be called). Correct me if I'm wrong, but I think G92 only supports 4 64-bit lanes based on its design, so 256-bit is as high as it goes. However, this is a cheaper design, and it seems nVidia attempts to make up for the lack of bandwidth by using faster RAM (RAM is cheap now, so this compensation does save cash on each board). This will put decently fast and much cheaper video cards on the market and will actually net nVidia a lot more cash. While the G80 was a monster, I think nVidia saw that the $450 and $600 MSRP of the products at launch alienated a lot of its customers. In the end, nVidia is a business, not a research facility.

What I don't understand about this trend is its inclusion with nVidia's 3-way SLi platform. Originally, the G80 series worked very well with SLi because its memory bandwidth was much higher than the capabilities of the GPU. However, because SLi stores the same graphics information on each card (if I'm wrong here, please correct me), SLI GPUs could use this bandwidth and two GPUs to a much higher potential. However, with this current set-up, one would already be taking a memory bandwidth limited card and attempting to run even more information on a limited bus, which would give much more diminished returns, or so one would think. So what happened? A new development? Better drivers? It doesn't add up, that's for sure (unless nVidia thinks the public is stupid and will just buy whatever new product/platform they put out).

I am inclined to believe that the memory controller is identical to the one found on G80 so that means ATLEAST 384bit memory bus, I have yet to see any definitive evidence saying that G92s memory controller is any different than G80 other than the packaging on which G92 has been used. But the key determing factor is not just the GPUs memory controller, but the PCB it sits on, 384 bit memory bus requires 12 peices of memory, which requires more PCB layers, which requires deeper stage power circuitry which requires more power which all in turn costs more. The same logic and the same factors apply to R600/RV670.

SLi shares the same texture info on all cards so you are right. /

Nvidia cards are much bandwidth hungry cards than ATis are. Look at 2900XT, it has a 512bit external (1024 internal) memory bus with ether 512mb DDR3 @ 825mhz (x2) for a total of 105.6GB/s bandwidth or 1gb of DDR4 @ 1ghz (x2) (was varied between manufacturers) for around 120GB/s memory bandwidth, and you got the RV670 on half the memory bus and you don't notice the difference possibly because R600 was so poorly designed. Where as the single reason why the 8800GTX/ultra is still considered a better card than the new 8800GTS is because of the memory bandwidth, despite the 65-100mhz GPU clock difference, and it shows on 19x12 and 25x16 resolutions with AA. Is because the 24GB/s memory bandwidth difference between the GTX and the G92 GTS (62.08 v 86.4).

The 384bit/768MB memory bus / size was a really good stop gap for nvidia for those extremely high resolutions where 256bit/1GB would otherwise be ineffective due to bandwidth, I don't know why they are regressing.

I believe there is some other aspect to G92-420 and G92-450 that hasn't been realized, ether in IPC gain like which has been realized with G94 or extra pipes of some kind which would mean this is a NEW piece of silicon with the same name. Because if the 9800GTX is a 8800GTS thats 675/2200, than they would of been farther along in performance if they kept making newer higher clocking G80 revisions on G80 PCBs with G80 power usage, seriously, most 8800 ultras had enough headroom in them to hit 675-700 mhz, and these are chips manufactured in early 2007.
 
They sell faster (GPU) clocked 8800GTs and 8800GTSes (overclocked) it would pretty much be suicide if they did that, it still doesn't beat the memory bandwidth of the old GTX so it still will prob get beaten with high AA senerios, and its 15 months old. Over the past few months I have been hearing rumors of the 9800GTX being a "new" chip called D9P, but I didn't hear anything above that, only that it was going to be out with a g92 based GX2 (sounds odd)

8800GT = G92-270 112sp
8800GTS = G92-400 128sp
9800GX2 =? G92-420 ???
9800GTX =? G92-450 ???

Three things. The 9600GT that just came out seems to be an actual and competent midrange chip with 64 sps, but it performs like it has 100-110 SPs out of a G92 so that chip got an increase in IPC. I was under the asumption that this card was called D9P, the 9800GTX was called D9E, their would be some lowend part called D9M, it made sense, but apparently now that was all fud, the 9600GT is referred to as G94.

Secondly, G92-420 and G92-450 are codenames for what, it was assumed that their are only 128 pipelines in the G92, what does 420 and 450 describe? the G92-GT to GTS goes from 270 to 400 (weird) and their is only 12 sp's difference.

Third, if all of the above is true (which I don't concede) Why wouldn't nvidia use the 8800GTX/ultra pcb, the G92 is believed to be pin compatible with G80, it was good for 384bit memory bus, so despite these new cards with higher clocks, they are still getting beat by a 15 month old card because it has more memory bandwidth, so unless G92-450 is something new which is unforeseen, its still going to get edged by an 8800 ultra STILL

So the question is, why after 15 months, would nvidia use this new updated chip in a configuration which essentially restricts it when they already have which is asumed to be the same chip, in the same configuration as discribed above, slightly faster memory, but on a pcb with a slower memory bus than the previous high end. One step forward and two steps back? I don't think so, someones lying to us.

All good points...
Thanks for the post.
I want the 9800GTX to be a runner! :D
 
Back
Top