nVIDIA november assault

[Cookie Monster]

8800GT has 112SP clocked at 1.5ghz

Several sources are suggesting that the upcoming GeForce 8800 GT has 112 stream processors like the new GeForce 8800 GTS 640MB SKU. We also heard that its shader clock is running at 1.5GHz. CJ did some nice analysis on the Nvidia's strategy here on why is there a new 8800 GTS SKU. He revealed that there is a shortage of GTX cores but plenty of GTS cores going around that didn't quite make it as a GTX. So if you launch a new product that is faster than a 8800GTS, no one will buy the more expensive GTS unless you up its stream processor count for higher performance. Also with a 320-bit memory interface, it certainly perform better at high resolutions and with lots of AA/AF/HDR. This way Nvidia can still get rid of the inventory.

8800GT has 112SP

8800GT scores 11760 in 3Dmark06

We have got the initial scores of GeForce 8800GT runnning on a ASUS P5K-D board with Core 2 Quad Q6600 processor. It scored 11760 in 3dmark06. The Core runs at 600MHz while memory is at 1800MHz.

8800GTS revised

It seems that Nvidia is planning another GeForce 8800 GTS SKU with more stream processors. This is to widen the performance gap between the upcoming 8800 GT and 8800 GTS as 8800 GT is coming too close and it might hurt the sales of 8800 GTS. As such, the number of stream processors is increased from 96 to 112 while the rest of the specs like 500/800 core/mem clock, 320-bit memory interface, 640MB GDDR3 memories remain unchanged. There is also an interesting side note that NVIDIA will NOT be doing any formal launch activities, press announcements, website updates, or any other marketing around this new sku. Partners can conduct their own marketing activities around the sku as they see fit.

8800GTS 640mb gets a speed boost in terms of ALU power, and this would also lead me to believe that the TMU count has been upped from 24 to 28. If this is priced the same as current GTS, looks to be an interesting buy.

8800GT pictured

More details have popped up and we learned that GeForce 8800 GT (G92) will be clocked at 600MHz core and 1.8GHz memory. However Nvidia might still further increase the core clock depending on how Radeon 2950PRO performs. The reference score provided by Nvidia to the partners is 10.8K in 3DMark06 or to be exact 10769. The launch date is said to be brought forward to October 29th. Mobile101 has a photo of the 8800 GT card even and it is a single slot solution as we have been telling you.

G92 is officially known as 8800GT. The shader count/TMU/ROP count is still abit vague but its defitnately a 65nm GPU, 256bit memory interface and comes with 256/512mb of ram.

Current guess is 64SPs that could be dual precision (or dual MADD), 16 TMUs and 16 ROPs? This is tricky because the ALU:TMU:ROP ratio of G84 (8600GTS) is different to G80.

8800GT OCes well

We heard the overclocking headroom for GeForce 8800 GT is pretty impressive mainly due to the finer 65nm process technology. However, Nvidia has purposely limits the core clock at 600MHz so its performance won't get too close to the GeForce 8800 GTS thereby hurt its sales. However, users can still choose to overclock the core and we should be seeing more than 700MHz easily. However, if Radeon HD 2950PRO performance comes too close for comfort, Nvidia might increase the core clock further.

With rumoured prices of $249 for the 512mb model, and the OCability that could make it perform on par with an 8800GTS sounds pretty nice.

8800GT has 3 heat pipes

CHW has a slide on the the GeForce 8800 GT which confirmed the specs we told you earlier. Also we heard that the single slot cooler on the 8800 GT card has 3 heat-pipes and Nvidia told its partners that it won't be hot.

More pic!

It seems like more GeForce 8800 GT card photos have surfaced online as seen in our forums and apparently another website in China got the inside scoop. This time round the PCB is black in color, which looks better than the reference green PCB. The GPU heat-spreader looks the any other G80 series and the rumored die size of G92 is 289mm2 as opposed to ~480mm2 of the G80.

Discuss.

note - Things will be updated.

 

[Keysplayr]

It's nice that the new 8800GTS will have 112 stream processors (maybe), but I don't see how this 8800GT, with 2/3rds the stream processors, 2/3rds memory bus width, and 4/5ths the TMU's and ROP's could possibly encroach on current 8800GTS variants. I would think that the core speed on the 8800GT would have to be significantly higher than 600 to have the same computational power of another 32 stream processors of the 8800GTS (which is commonly available at 513MHz core or being factory o/c'd).

Think of a current 8800GTS and hack off 32 stream processors, and do nothing else to it to diminish it. How much slower would it be than a 96 shader G80? My guess is 33%. The core would have to be somewhere in the range of 680MHz. I know my math is probably off, but you get the idea (I hope 🙂 )

Compare a 6800nu to a 6600GT

6800nu 325MHz
12 pipes
256bit 128MB GDDR3 at 700MHz
16ROP's

6600GT 500MHz
8 pipes
128bit 128MB at 1000MHz
8 ROP's

The 6800nu was always on par, or faster than a 6600GT. But look where the clocks are between the two. Major differences in core/memory clocks.

Granted I know we are talking vastly different architectures here, but it's something to think about.

 

[BFG10K]

Single slot cooling? Nice. And I don't even see external power connnectors.

I see those die shrinks are helping both vendors. :thumbsup:

 

[Synomenon]

Finally... Another good single slot card.

 

[coldpower27]

My guess is that the new card has 80 stream processors and 20 TMU, at 600MHZ this means 48K Shader Cycles = to the old 8800 GTS. Hence the new 8800 GTS 640 at 112 Stream processors for 57.6K Shader Cycles and 28 TMU. This would put some distance between them. Remember if 112 is possible which is 7/8 there is no reason why 5/8 is impossible.

I am going to assume the 10.7K 3D Mark 2006 Score without any other available information was derived using a Core 2 Extreme QX6850. Anyone have a gather at what a reference 8800 GTS 640 gets with that processor?

 

[nonameo]

I wonder what AMD plans to do about this?

 

[CP5670]

That 8800GTS replacement might be interesting if it comes at the same $350-ish price point as the existing GTS.

There is also an interesting side note that NVIDIA will NOT be doing any formal launch activities, press announcements, website updates, or any other marketing around this new sku.

I think this is further confirmation that there won't be any high end 8800GTX/U type card coming out. They mainly advertise the top end card of a generation at those launches, which would make no sense to do if their own card from a year ago is faster.

 

[Munky]

What matters most to me from this news is it might be a good idea to hold off buying a 8800gts, assuming the prices aren't jacked up as soon as the 112 shader part is released. The gts could never match a stock gtx at high resolutions no matter how much it's oc'd, but the extra shaders would give it a much better chance. I need gtx performance for the price of a gts!

 

[Aharami]

I wonder how soon will these new GTS cards hit the shelves. And whether EVGA will be making these cards or not. I just got a EVGA 8800 GTS 640 mb card and would be nice if I could step-up to this new GTS using EVGA's step-up program

 

[electronicmaji]

will the new gts get a price cut?

Any idea when in november this will hit?

 

[chizow]

Originally posted by: coldpower27
My guess is that the new card has 80 stream processors and 20 TMU, at 600MHZ this means 48K Shader Cycles = to the old 8800 GTS. Hence the new 8800 GTS 640 at 112 Stream processors for 57.6K Shader Cycles and 28 TMU. This would put some distance between them. Remember if 112 is possible which is 7/8 there is no reason why 5/8 is impossible.

I am going to assume the 10.7K 3D Mark 2006 Score without any other available information was derived using a Core 2 Extreme QX6850. Anyone have a gather at what a reference 8800 GTS 640 gets with that processor?

10k 3DMark06 is about the same as a 640MB GTS with a C2D, so ya, the 8800GT is very close in performance.

I also think the number of shaders on the GT would have to be closer to the 96 on the GTS based on clock speeds in order to be competitive, so 80 is definitely a possibility, HOWEVER, I don't think they're simply disabling shader quads, or more accurately, octets on the GT like they did with the GTS. GT is on a 65nm process and geared for the mainstream. Disabling shaders wouldn't make sense as there's no higher end part on this process yet and would eat much of the cost-saving benefits of a move to a smaller, cooler part. This will definitely be a part to keep an eye on for OC'ers if it has closer to 96 shaders on a smaller process, as much higher clock speeds should be possible with cooler temps and less power draw. I wouldn't worry too much about the 256-bit memory interface either, as I think the 320 and 384 on the GTS/GTX is still overkill with current clock speeds.

The improved GTS will also be an interesting part depending on its price point. Will close the gap between the GTX/Ultra provided you OC your GTS to GTX speeds. Still, its going to be on a 90nm process and in this case, will be binned or trimmed GTX cores, so don't expect too much more than what you're seeing with current GTS on the OC'ing side. If the 112 shader GTS it comes in at old GTS prices and the GT performs similarly as predicted, the older 96 shader parts have no where to go but down in price. Win-win situation for those looking for cheaper alternatives with good performance or those who are on the fence about a GTX/Ultra.

 

[GZDynastar]

Anyone have a good guess at the cost of the 8800GT? I think it looks like a good buy. (finally)

 

[lopri]

Sad. In the past we would see this part as 9900 GT and there would be a 9900 GTX/Ultra that'll perform more or less at 150% of 8800 GTX. How long will 8800 GTX stay at the top? 2 years?

Don't get me wrong. This part is indeed like a hypothetical 9900 GT and I do like the return of single-slot card. If this card has the 8600-like (or even better) video processing capability for HD conents, it's a near-ideal solution for mid to mid-high market. I just lament the lack (or delay) of ultra high-end updates.

 

[nitromullet]

Originally posted by: BFG10K
Single slot cooling? Nice. And I don't even see external power connnectors.

I see those die shrinks are helping both vendors. :thumbsup:

You can not see the 3x2 power connector but it's there

http://www.fudzilla.com/index....=view&id=3430&Itemid=1

 

[alcoholbob]

Looks like no matter what, we won't see a card that can play Crysis (or any number of the new games including other U3-engine based shooters) with any framerate headroom until well into next year.

 

[Arkaign]

Originally posted by: Astrallite
Looks like no matter what, we won't see a card that can play Crysis (or any number of the new games including other U3-engine based shooters) with any framerate headroom until well into next year.

SLI 8800GTX + Q6850 @ 4Ghz + 8GB DDR3?

 

[ConstipatedVigilante]

Originally posted by: lopri
Sad. In the past we would see this part as 9900 GT and there would be a 9900 GTX/Ultra that'll perform more or less at 150% of 8800 GTX. How long will 8800 GTX stay at the top? 2 years?

Don't get me wrong. This part is indeed like a hypothetical 9900 GT and I do like the return of single-slot card. If this card has the 8600-like (or even better) video processing capability for HD conents, it's a near-ideal solution for mid to mid-high market. I just lament the lack (or delay) of ultra high-end updates.

The high-end parts don't matter. Competition matters. Unless you're gaming at 1920x1200, the 8800 GTX should be enough for any game. I'm pretty keen on this 8800 GT - if AMD can put out a decent rival, then prices could get as low as $200.

 

[AzN]

Originally posted by: keysplayr2003
It's nice that the new 8800GTS will have 112 stream processors (maybe), but I don't see how this 8800GT, with 2/3rds the stream processors, 2/3rds memory bus width, and 4/5ths the TMU's and ROP's could possibly encroach on current 8800GTS variants. I would think that the core speed on the 8800GT would have to be significantly higher than 600 to have the same computational power of another 32 stream processors of the 8800GTS (which is commonly available at 513MHz core or being factory o/c'd).

Think of a current 8800GTS and hack off 32 stream processors, and do nothing else to it to diminish it. How much slower would it be than a 96 shader G80? My guess is 33%. The core would have to be somewhere in the range of 680MHz. I know my math is probably off, but you get the idea (I hope 🙂 )

Compare a 6800nu to a 6600GT

6800nu 325MHz
12 pipes
256bit 128MB GDDR3 at 700MHz
16ROP's

6600GT 500MHz
8 pipes
128bit 128MB at 1000MHz
8 ROP's

The 6800nu was always on par, or faster than a 6600GT. But look where the clocks are between the two. Major differences in core/memory clocks.

Granted I know we are talking vastly different architectures here, but it's something to think about.

Faster SP clocks might have something to do with it.

 

[manowar821]

If they perform that well, do you guys think that a pair of them in SLI would do better than an 8800GTX, or even an 8800Ultra?

Also, what are the chances that EVGA would allow me to trade in my 8800GTX for a pair of these 8800GT's?

I'm actually getting a little intrigued by this... Only a little, though. I'm still wishing and hoping for a high end card of some sort by EVGA by December. Wishful thinking? Maybe.

 

[imported_Shaq]

Will this new GTS be 65nm and single slot also? I'm hoping it will be 65nm and dual slot, which should OC well enough to pass a GTX and maybe an ultra. Current 90nm GPU's go 620 regularly so I wouldn't be surprised if it went well over 700 at 65nm. That would definitely be worth an upgrade.

 

[Munky]

The new gts is still the same 90nm g80 core, only this time with 112 shaders enabled. But even so, at 650mhz the new gts would be a real competitor to a stock gtx, unless the extra memory and bandwidth on the gtx make a significant difference.

 

[imported_Shaq]

Originally posted by: munky
The new gts is still the same 90nm g80 core, only this time with 112 shaders enabled. But even so, at 650mhz the new gts would be a real competitor to a stock gtx, unless the extra memory and bandwidth on the gtx make a significant difference.

That's unfortunate. That wouldn't be worth upgrading for current 640 owners. Why can't Nvidia take the same 8800GT GPU, make it dual slot and up the voltage, put 1GB of DDR4 memory and sell it for $499? It seems it would be no problem for them. It would be faster than a GTX so they could discontinue them and would save them money due to the smaller process. They would sell extremely well since it is Xmas time and Crysis, UT3, Gears and several other AAA games are coming out.

Why do they want to drop the ball and give DAAMIT time to catch up? The last 2-3 years they have been good at staying one step ahead of ATI.

According to this link off firing squad the current 640 is about 10% slower overall when OC'd compared to a stock GTX. So it should be almost the same with the extra shader processors. http://www.firingsquad.com/har..._gtx_gts_overclocking/

 

[Keysplayr]

Originally posted by: Azn

Originally posted by: keysplayr2003
It's nice that the new 8800GTS will have 112 stream processors (maybe), but I don't see how this 8800GT, with 2/3rds the stream processors, 2/3rds memory bus width, and 4/5ths the TMU's and ROP's could possibly encroach on current 8800GTS variants. I would think that the core speed on the 8800GT would have to be significantly higher than 600 to have the same computational power of another 32 stream processors of the 8800GTS (which is commonly available at 513MHz core or being factory o/c'd).

Think of a current 8800GTS and hack off 32 stream processors, and do nothing else to it to diminish it. How much slower would it be than a 96 shader G80? My guess is 33%. The core would have to be somewhere in the range of 680MHz. I know my math is probably off, but you get the idea (I hope 🙂 )

Compare a 6800nu to a 6600GT

6800nu 325MHz
12 pipes
256bit 128MB GDDR3 at 700MHz
16ROP's

6600GT 500MHz
8 pipes
128bit 128MB at 1000MHz
8 ROP's

The 6800nu was always on par, or faster than a 6600GT. But look where the clocks are between the two. Major differences in core/memory clocks.

Granted I know we are talking vastly different architectures here, but it's something to think about.

Faster SP clocks might have something to do with it.

It is still 32 shaders down. If it is indeed 64 shaders. With a 10,000+ score in 3DMark06, I'm inclined to believe it's more than 64 shaders.

 

[Skott]

Originally posted by: manowar821
If they perform that well, do you guys think that a pair of them in SLI would do better than an 8800GTX, or even an 8800Ultra?

Also, what are the chances that EVGA would allow me to trade in my 8800GTX for a pair of these 8800GT's?

I'm actually getting a little intrigued by this... Only a little, though. I'm still wishing and hoping for a high end card of some sort by EVGA by December. Wishful thinking? Maybe.

My guess is that eVGA wont go for that deal simply because their step up program deals with a card for card swap. Not 1 for 2 or 2 for 1. Its 1 for 1.

 

[AzN]

Originally posted by: keysplayr2003

Originally posted by: Azn

Originally posted by: keysplayr2003
It's nice that the new 8800GTS will have 112 stream processors (maybe), but I don't see how this 8800GT, with 2/3rds the stream processors, 2/3rds memory bus width, and 4/5ths the TMU's and ROP's could possibly encroach on current 8800GTS variants. I would think that the core speed on the 8800GT would have to be significantly higher than 600 to have the same computational power of another 32 stream processors of the 8800GTS (which is commonly available at 513MHz core or being factory o/c'd).

Think of a current 8800GTS and hack off 32 stream processors, and do nothing else to it to diminish it. How much slower would it be than a 96 shader G80? My guess is 33%. The core would have to be somewhere in the range of 680MHz. I know my math is probably off, but you get the idea (I hope 🙂 )

Compare a 6800nu to a 6600GT

6800nu 325MHz
12 pipes
256bit 128MB GDDR3 at 700MHz
16ROP's

6600GT 500MHz
8 pipes
128bit 128MB at 1000MHz
8 ROP's

The 6800nu was always on par, or faster than a 6600GT. But look where the clocks are between the two. Major differences in core/memory clocks.

Granted I know we are talking vastly different architectures here, but it's something to think about.

Faster SP clocks might have something to do with it.

It is still 32 shaders down. If it is indeed 64 shaders. With a 10,000+ score in 3DMark06, I'm inclined to believe it's more than 64 shaders.

A 8800gts has a SP clock of 1200mhz. Very low SP clock even compared to 8600gts. These new 8800gt or whatever is going to have faster SP clock to compensate. It all equates to shader operations per second.

8800gts has 115200 Shader operations/sec

Let say these new 8800gt has a SP clock of 1600mhz which equates to 102400.

Not much difference when you all add it up.