G71 to be 7900 (750Mhz/1.8Ghz); ~13k Mark05; release: early March?

[Turok]

http://www.hexus.net/content/item.php?item=4346


NVIDIA GeForce 7900 512MB (700MHz GPU / 1800MHz GDDR-3 (not GDDR-3+) (AMD A64 FX-60 2.8GHz)
~13,000 3DMarks (?)


http://www.hexus.net/content/item.php?item=4343


It would seem that a few key NVIDIA partners have already been sampled with GeForce 7900, and they're currently working towards a hard launch planned for CeBIT 2006 in Hannover, Germany (9th through to 15th March), and if so, our money's would be on watching for announcements from NVIDIA, ASUSTeK, eVGA, XFX and possibly MSI, between the 9th and 11th March...

...The flagship single-card product GeForce 7900 GTX is claimed to be a 32 pixel processor part with a core GPU frequency of 700/750 MHz, and should ship with 1.1ns graphics RAM clocked at 800/900MHz.

EDIT: (again)
x1900xt + FX 60 @ 2.8Ghz = ~11,500 Mark05
7900 + FX 60 @ 2.8 Ghz = ~13,000 Mark05 (prediction, I guess)

Looks like nVidia could win again if this is true :rolleyes:

[n00b 0f l337]

Erm.. I think your forgetting x1900xt/xtx is the 7900's competition..

[Piotrsama]

Uhm, If 7900 will be the top model, you have to compare it with the X1900XTX... no?

[n00b 0f l337]

And x1900** has "48" pipes.

[Turok]

And x1900** has "48" pipes.

oh, oops. I meant 1900xt = 11k Mark05. :p:
Change it to 1900xt = 11k. Sry about that.

EDIT:

http://resources.vr-zone.com/newspics/Jan06/12/3DMark05-625-725-FX60.gif

[onewingedangel]

And by march we'll likely have the X1900 TOP at 700mhz + to counter Nvidia's launch. I can't believe that the newer core will clock lower than the x1800 considering the x1900 currently runs on lower voltage and is cooler. Up the voltage to x1800 levels and clock it up!

A 32 pipelined part is going to be more limited by the 256bit memory bus isn't it? Even at 1.8ghz thats like a 6800 ultra running with 450mhz (900effective)gddr3 (16 vs 32 pipes, half bandwidth). Insane clocks for the nvidia architecture though, I doubt even a higher clocked x1900 would rival it.

[n00b 0f l337]

Well the benchy's with the ati only on an fx57 seem to be well matched. And traditionally ati's clock farther.

[MaxxxRacer]

Well there goes the house again.. 1up for nVidia when they come out with the 7900.. ATI has been on a roll of being 1 step behind nVidia at al times..

[Cybercat]

darn, I was coming in here hoping for some new news.

[n00b 0f l337]

Whao wait.... March for this, x1900 already out?
How does that put ati behind?

[OmegaMerc]

Sorry to ask the stupid question but whats the diffrence between GDDR3 and GDDR3+?:slap:

[onethreehill]

x1900xt + FX 57 @ 2.8Ghz = 11k Mark05
7900 + FX 60 @ 2.8 Ghz = 13k Mark05

Looks like nVidia might win again :rolleyes:


oh, oops. I meant 1900xt = 11k Mark05. :p:
Change it to 1900xt = 11k. Sry about that.

EDIT:

http://resources.vr-zone.com/newspics/Jan06/12/3DMark05-625-725-FX60.gif

You need to edit again :)

The X1900XT was benchmarked with FX 60 @ 2.6 GHz

FX-60 @ 2.80GHz will score ~ 11614 3DMark05

BTW 7900 score is a prediction

[sickness]

The only thing I don't like about these new ATI cards is the 2 slot cooler. If you look at Nvidia, they had a 2 slot cooler on the 6800 Ultra but the newer, better 7800GTX doesn't.

It's just kinda awkward putting watercooling on a dual slot video card. Speaking of watercooling, would the current vga water blocks be compatible with the x1900?

[5-Clicks]

The X1900's have nearly the exact same PCB as the X1800's, so yeah.

BTW sickness, the X1800XT's are better than the 7800GTX's performance wise (oc'ed anyway). The GTX 512mb beats the XT but that uses a dual slot cooler too...so I'm not exactly certain that your statement is entirely true.

[perkam]

Pls do not take the Hexus results at face value...they are estimates and nothing else...poorly done estimates too.

Perakm

[nn_step]

Unless it is an offical Launch.. and there is no good Proof.. I tend to not Automatically assume it as the Truth

[JuanFlaiter]

In the vaporware era January means February and March is April.

[nn_step]

I freaking hate Paper launches...

[JuanFlaiter]

I freaking hate Paper launches...

Well it seems the X1900s will be hard launched. Already in stock in some places...

[Revv23]

The X1900's have nearly the exact same PCB as the X1800's, so yeah.

BTW sickness, the X1800XT's are better than the 7800GTX's performance wise (oc'ed anyway). The GTX 512mb beats the XT but that uses a dual slot cooler too...so I'm not exactly certain that your statement is entirely true.


nm sry

[sickness]

Well it seems the X1900s will be hard launched. Already in stock in some places...

Which also means that the price won't be jacked like the 7800GTX 512. The main reason why the 512 is so expensive in the first place is because it's never in stock. There will be plenty of x1900's available for purchase upon release.

[nn_step]

Well it seems the X1900s will be hard launched. Already in stock in some places...
I know and I already ordered one.. but the lack of the 7900 is pissing me off because they were doing so well with all the hard launches:(

[bakalu]

G71 has 24 TMU or 32 TMU ?
http://www.xbitlabs.com/news/video/d...110232032.html - G71 has 32 TMU
http://www.pcpop.com/doc/0/125/125773.shtml - G71 has 24 TMU

Another question? - GPU has 24 TMU and 24 ROP is stronger or GPU has 32 TMU and 16 ROP is stronger ? Shader pipeline is the same.

Sorry my English.

[Cybercat]

The highend version will probably have 32. There might be some lower versions of it with pipes disabled though.

[turtle]

To my understanding Nvidia's will have 16ROPs, 32TMU's/32ps (ala 32 "real pipes"). It may be 24TMU's though, that is what I thought originally as well.

ATi's has 16 ROPs, 16 TMU's, and 48ps processors.

The pipes arn't similar, so they won't be directly comparible. It seems in some instances the higher shaders will help, but expect the TMU's and nvidia's moving to 90nm (higher freq) to help them out quite a bit.

[Coroner Kyle]

I love the competition, lets just hope that ATi and Nvidia keep at and hopefully the chipset competition will get better too. Unfortunately we can't compare the hardware, but they'll be benched soon enough. I'm sure Nvidia will up the ante now that ATi has gotten their act together again.

[Revv23]

my bet is that its a 80nm 7800, nothing more.

Plus, unless they plan on hard launching in march with crazy amounts of stock, but the looks of it the xtx will be in the mid 500's by that time, with the 7900 closer to $700.


I dont see this as being a huge release, especially because its possible for ATI to release an XTX PE in that amount of time, maybe that is the reason the firm dropped the pe this time around. But yeah i could see XTXPE running stock at like 750mhz, and by then i bet memory options will be better.

Damn graphics market is FLYING right now.

[nn_step]

What's taking it so long...

[Stuperman]

didn't the RD580 chipset get pushed back to about mid march, That would probably help ATI a bit as I imagine it's "tuned" for ATI HW slightly. I think it will be a close race.

[ex2cib]

my bet is that its a 80nm 7800, nothing more.




i really dont see that, what would be the point, thats basically all the GTX 512mb is right now, a clockspeed bump. no one is going to care how much more 90nm can clock with the same amount of pipes as the GTX now IMO. besides, 90nm will make the core smaller, so they can add a few extra pipes to make up for lost size:D

[onewingedangel]

A 24 pipe card at 700mhz (and considering there are gtx512 overclocks at 600mhz on air) or above should compete nicely with the x1900xtx. Why would Nvidia make their card bigger (and more expensive) than needed? Tweaking the core a bit, and higher clocks should be more than enough to match up to the x1900. Upping the pipeline count is not liable to increase performance all that much as surely it would be bandwidth starved, unless they now have a true 512bit memory bus or memory was clockd even higher (which Isn't possible at this time)

[physics_geek]

A 24 pipe card at 700mhz (and considering there are gtx512 overclocks at 600mhz on air) or above should compete nicely with the x1900xtx. Why would Nvidia make their card bigger (and more expensive) than needed? Tweaking the core a bit, and higher clocks should be more than enough to match up to the x1900. Upping the pipeline count is not liable to increase performance all that much as surely it would be bandwidth starved, unless they now have a true 512bit memory bus or memory was clockd even higher (which Isn't possible at this time)
:slapass: :slapass: :slapass: :slapass: :slapass: :nono:

[nn_step]

:rofl: ATi doesn't even have a TRUE 512 bit Memory interface
You'll have to wait Till GDDR4 for that

[[XC] leviathan18]

yeah they have internal ring bus with "512bit" mem interface but between the core and the mem is 256 bit

[Northwood]

i been saying for ages that G71 will be a die shrink of the 7800GTX 512, 90nm will produce 700mhz+ and will compete easily with ATI's top offering, i think this 32 pipes stuff is utter rubbish. we'll see soon anyway.

oh you have my permission to sig this when i'm proved wrong or right :D

[onewingedangel]

:rofl: ATi doesn't even have a TRUE 512 bit Memory interface
You'll have to wait Till GDDR4 for that

Thats why I used the word true in my post, i thought implication was more than enough. A 512bit memory interface would be possible with gddr3 but would require the use of more chips, but that shouldn't be an issue when migrating to 512MB cards and greater.

[nn_step]

Thats why I used the word true in my post, i thought implication was more than enough. A 512bit memory interface would be possible with gddr3 but would require the use of more chips, but that shouldn't be an issue when migrating to 512MB cards and greater.
WRONG it would require MANY MORE traces on the PCB... WHICH at this time isn't cost effective..:fact:
But with the introduction of GDDR4 without increasing the Trace count they can have a 512bit memory interface.. it has to do with the design of the chips:fact:

[onewingedangel]

there would still have to be 512 traces for a 512bit interface, whether this can be done on 4,8 or 16 chips is the only real question. Upping the speed of the memory would benefit without adding to pcb complexity, but that isn't really feasible with gddr3 at this point.

[nn_step]

there would still have to be 512 traces for a 512bit interface, whether this can be done on 4,8 or 16 chips is the only real question. Upping the speed of the memory would benefit without adding to pcb complexity, but that isn't really feasible with gddr3 at this point.
And guess how many traces are on your 256bit Graphics card:rolleyes:

[onewingedangel]

WRONG it would require MANY MORE traces on the PCB... WHICH at this time isn't cost effective..:fact:
But with the introduction of GDDR4 without increasing the Trace count they can have a 512bit memory interface.. it has to do with the design of the chips:fact:

I was highlighting that you said that a 512bit memory interface without increasing trace count - which is impossible. GDDR4 will increase bandwidth on the same memory bus width through speed, just as you would with higher clocked gddr3, but a 512bit bus will have to come at some time unless we move to an even faster serial memory interface. We've had 256bit interfaces for some years now, and clock speed increases of memory have scaled somewhat with the increase in speed and pipelines, but recently (post 7800) this has nt been the case, and more memory bandwidth is required. Short of transitioning to xdr memory, increasing the bus width is the most logical solution, unless gddr4 can scale much faster than I suspect.

[n00b 0f l337]

With the fact that we are now moving away from sheer pipelines to increase other parts estimates like...

Oh yeah it's got more of one part so it'll get 1.5k more is sorta pointless. THere's no way any more to geuss without seeing really benchies. Then we can geuss OC scores.