D9P/G94: 9600 GT on 19th of February

AnarchX

Veteran
EDIT: This GPU has now been delayed to February 19th according to Fudzilla (most recent) and February 21th according to Expreview. However, it is worth pointing out that's a board issue and it's still an A11.
Expreview has some interesting informations about NVs upcoming performance-part of the ninth Geforce series:


Source also tells some specs of the nex-gen GPU D9P. The core clocks at 500MHz while memory at 2000MHz, GPU will have [strike]96SP [/strike] 64. It is a main-stream product and only have one SLI connector so it can not support 3-Way SLI.

http://en.expreview.com/?p=132

Die-size seems to be ~200mm², so it may be base of GX2-type D9E with a bit higher-clocks than on 9600 GT and allows a high margin SKU, like 79x0GX2.


update:

3DMark2006 results?:
http://64.233.179.104/translate_c?h...conline.com.cn/topic.jsp?tid=8000957&pageNo=1



http://www.pcpop.com/doc/0/258/258133.shtml

Package with new lettering and different* look than the one from Expreview.

*other resistor layout and bigger die
 
Last edited by a moderator:
The fact it might be pin-compatible is quite a surprise to me - even though it might make some sense for a number of reasons. The process is also confusing: Digitimes claimed June for a 55nm D9P, and now it's February for a 65nm one? Errr? Yay @ intentional FUD galore!

Anyway, my guess right now is that speculating on specs based on performance or vice-versa would likely be lost time as if any architectural tweak could change things quite a bit.
 
Maybe first 65nm and then in mid-2008 an optical-shrink to 55nm, like G73?

But one thing I recognized now: Why use this slide the old NV lettering? :???:
 
Expreview has some interesting informations about NVs upcoming performance-part of the ninth Geforce series:




http://en.expreview.com/?p=132

Die-size seems to be ~200mm², so it may be base of GX2-type D9E with a bit higher-clocks than on 9600 GT and allows a high margin SKU, like 79x0GX2.

Months before NV new card launch there is already exact rumors about launch date, in the last few times all come out as true, or card coming before the rumors suggest.

This card is born to kill the hd3850 and than NV has almost all segment (50-100$ segment can be still for ATi) in they own hand.
(btw. core clock speed looks very low)

What can be ATi answer to this card? because 2xrv670 X2 3xxx card there is no rumor about anything.
 
Maybe first 65nm and then in mid-2008 an optical-shrink to 55nm, like G73?

But one thing I recognized now: Why use this slide the old NV lettering? :???:

Until now expreview not was fudzilla style site, hard to belive they start now to made the site fudzilla style :smile:
 
Anyway, my guess right now is that speculating on specs based on performance or vice-versa would likely be lost time as if any architectural tweak could change things quite a bit.

G86-like shader-core?
2 FLOPs MADD + 1 FLOP MUL (partially GS) + 1 FLOP SFUs per clock.

9600GT : with 96SPs @ 1.25GHz (todays ratio) = 480 GFLOPs

D9E-GX2: 2x96SPs @ ~1.9GHz = ~1450 GFLOPs => x2 = ~2.9TFLOPs (the rumored "near 3 TFLOPs" setup... :eek:)
 
Last edited by a moderator:
So this card will wash away the stains left by the 8600GTS?

:D The 6600GT legacy lives on!

Wonder what the rest of the specs are seeing as its rumoured to have a 256bit bus, core clock of 500MHz and 96 SPs.
 
So this card will wash away the stains left by the 8600GTS?

:D The 6600GT legacy lives on!

Wonder what the rest of the specs are seeing as its rumoured to have a 256bit bus, core clock of 500MHz and 96 SPs.

I'm betting 1.2GHz on the shader clock and around 8800 GTS (the old G80-based 320MB/640MB ones, not the new G92-based 512MB version) general level of performance. Not bad at all for (likely) sub-200 dollar cards.
 
I'm betting 1.2GHz on the shader clock and around 8800 GTS (the old G80-based 320MB/640MB ones, not the new G92-based 512MB version) general level of performance. Not bad at all for (likely) sub-200 dollar cards.

Yep, specs look identical to the GTS. Assuming some architectural enhancements and possibly a bump in shader clock then it should give slightely higher performance. As you say, pretty good for a mainstream part!

This also bodes well for the highend part which at an absolute minimum should double the mainstream parts performance.
 
This part looks a bit strange to me in some ways - assuming that those D9E chips are still very similar to G92. First, the die size has magically gone down - 96 instead of 128 SP (and thus likely 48 tmus instead of 64) do not account for a die size reduction of ~290mm^2 to ~200mm^2. [edit: that should be ~320mm^2 to 200mm^2, actually]
Second, the core clock seems very low. Either it's clocked down to be low-power, or the chip itself has been redesigned to reach lower clocks (with fewer transistors). Either way, it should have a very favorable full load power consumption (it could certainly compete in idle power too with hd 3850/3870 if nvidia decides to enable a bit more of the mobile parts power-saving measures). If it's just lower clocked to have a low TDP, it should overclock very very well (though might need a voltmod).
Also, the reported memory clock of 2000Mhz seems very high. The 8800GT only has 1800Mhz, but on this lower performance part nvidia uses the most expensive gddr3 memory they could find so it has more memory bandwidth than the 8800GT?
Assuming a shader clock of 1.25Ghz, I agree it could indeed have similar performance than an old 8800GTS - similar shader power, less ROP throughput, same memory bandwidth. Thus performance closer to 3870 rather than 3850 probably.
 
Last edited by a moderator:
It`s all strange to me. If D9P is based on another 9x architecture core so why NVIDIA call it GF9xxx series?? D9P (if it has the same architecture as GF8800GT aka D8P) is supposed to be slower than GF8800GT so it doesn`t make any sense - D9P slower than D8P?? I doubt it.

The second thing is that Expreview said D9P supports DX10.1 and we know that D8P doesn`t support it then maybe it means that in D9P (and other 9-series GPUs) has more architectural improvements and changes than G92 has to G80.

The third question is if D9P is the same architecture like G92 (with only a few SPs, TMUs and ROPs less) why they don`t disable 16 SPs in G92 to get D9P (it is supposed to have 96SPs). NVIDIA have made this before with G71 (G71 - GF7900GTX/GT and GF7900GS ; G80 GF8800Ultra/GTX and GF8800GTS).

The last question is what about D9E-20 and D9E-40 (dual D9E-20 card with dual PCB) GPUs which were rumoured on PCInlife Forum and they will be probably launched in Feb 08 too?? Any more info about it??
 
Last edited by a moderator:
info update,so sorry for the mistake

Update: SP number changed and PCB/Memory design confirm, sorry for the mistake.

Source also tells some specs of the nex-gen GPU D9P. The core clocks at 500MHz while memory at 2000MHz, GPU will have 64SP (8800GS will be 96SP 192bit 384MB). It is a main-stream product and only have one SLI connector so it can not support 3-Way SLI.

We just confirm that GeForce 9600 GT use P545 PCB design (8800GT use P393 PCB design), and the memory will be 256bit 512MB, still don’t know support DX10.1 or not .
 
Darn (beaten to the punch), I was just about to post that this can't have 96SP and be D3D10.1 - the transistor budgets just don't work out.

Jawed
 
Hmmm. Wonder if this means the next MacBook Pro will use RV670 or RV635. (Yes, Shirley, I am waiting to buy an MBP.)
 
Update: SP number changed and PCB/Memory design confirm, sorry for the mistake.

Source also tells some specs of the nex-gen GPU D9P. The core clocks at 500MHz while memory at 2000MHz, GPU will have 64SP (8800GS will be 96SP 192bit 384MB). It is a main-stream product and only have one SLI connector so it can not support 3-Way SLI.

We just confirm that GeForce 9600 GT use P545 PCB design (8800GT use P393 PCB design), and the memory will be 256bit 512MB, still don’t know support DX10.1 or not .

Errm, that's even more strange. Now, 64SP sounds believable, and performance should still be quite good (if the shader clock is at least 1.5Ghz), and the die size would probably about fit now, but the 256bit 2000Mhz memory doesn't make a whole lot of sense with such a part.
It would make more sense if this part would have a 192bit memory interface rather than the 8800GS - then at least the 2000Mhz memory wouldn't be there just to increase costs.
 
but the 256bit 2000Mhz memory doesn't make a whole lot of sense with such a part.
Agreed, this is a $150 part we're talking about, with 8800GS at $200. And I can't imagine those prices holding up for long.

Jawed
 
Agreed, this is a $150 part we're talking about, with 8800GS at $200. And I can't imagine those prices holding up for long.

Jawed

The old 7600 GT "2.0" (G73-B1, 80nm) had 1.8GHz GDDR3 chips, so 2.0GHz for a mainstream part a year later is not that surprising, especially if they're sticking with those cheap Qimonda ones already found in most 8800 GT's and 8800 GTS 512MB.
 
expreview: Hello, welcome to the forum! :) Were you explicitly told 64 SPs, or did you (or your source) infer that from another number? For example, a 32 TA/TF design with a higher tex ratio (->96SPs) seems very plausible and likely more balanced.

Jawed: 'm not sure how we can know what the density is unless we are certain NVIDIA reused the same synthesis they did for G92/G98. Given the clock speed, I'd kinda expect that not to be the case. Also, we don't know for sure whether it's 65nm or 55nm, and what foundry it is produced at.

Oh, just to make sure an outdated rumour is not considered as fact in this thread: G92 is ~320mm2, not ~290mm2.
 
The old 7600 GT "2.0" (G73-B1, 80nm) had 1.8GHz GDDR3 chips, so 2.0GHz for a mainstream part a year later is not that surprising, especially if they're sticking with those cheap Qimonda ones already found in most 8800 GT's and 8800 GTS 512MB.
Where the memory runs at 1.8 and 1.94GHz effective, respectively. So, NVidia is going to spend extra to get the right memory for this cheaper part?

Jawed
 
Back
Top