9900GTX render (authed 2X!) pic!

MrWizard6600

Supreme [H]ardness
Joined
Jan 15, 2006
Messages
5,791
All images taken from VR-zone and Xtrems-systems forums.

http://www.vr-zone.com/articles/GeForce_9900_GTX_Card_Design_Exposed/5763.html

p651frontsideze0.png


GT200. Can't tell you how happy a single die makes me, I was afraid the GT200 was going to be some kind of frankin-dual-G92b's.

:):):):)

edit: man VR zone is so much better than everyone else for speculative rumors. When they post something, its usually pretty concrete, unlike some people.

edit2: what I know about specs:

- 10 SP banks, each with 24SPs, for a total of 240.
- 8 ROP sectors, each containing 4 ROPs (32 units total),
- Each sector connected to a single 64bit memory controller, 512bit bus width total :).
- Clocked at a reasonable speed for .8ns memory that leaves you with over 125GB/s total memory throughput without breaking a sweat.
- 65nm, but NV is threatening to move it to 55nm so long as the G92b (first 55nm from NV) isn't bleeding too bad (which we'll find out when the 9800GT is released or delayed).
- 1024MB on referance design, spread across 16 GDDR3 :)() 64MB chips (my bad, I knew it was double sided but just sorta forced the numbers to fit).
- 240W TDP (!!!) per die. If my information holds true, Quad SLI will require (and I'm the guy who thinks I can SLI two 8800s on my 550W) PSUs above 1KW.
- Stock will be dual slotted cooling, and will require something of a no-nonsense design (perhaps like ATIs stock HD3870X2 cooler).
- 2nd core from NV to use an IHS.
- $500?. If you've ever taken a look at Fud's front page its always pro ATI anti NV, so I take him saying $500 to mean "at most, $500"
 
wow! a NEW GPU and not just another 8800 variant!?!?! It's about time!
 
All images taken from VR-zone, kudos to them for the skoop!

http://www.vr-zone.com/articles/GeForce_9900_GTX_Card_Design_Exposed/5763.html


GT200. Can't tell you how happy a single die makes me, I was afraid the GT200 was going to be some kind of frankin-dual-G92b's.

:):):):)

edit: man VR zone is so much better than everyone else for speculative rumors. When they post something, its usually pretty concrete, unlike some people.

edit2: what I know about specs:

- 10 SP banks, each with 24SPs, for a total of 240.
- 8 ROP sectors, each containing 4 ROPs (32 units total),
- Each sector connected to a single 64bit memory controller, 512bit bus width total :).
- Clocked at a reasonable speed for .8ns memory that leaves you with over 125GB/s total memory throughput without breaking a sweat.
- 65nm, but NV is threatening to move it to 55nm so long as the G92b (first 55nm from NV) isn't bleeding too bad (which we'll find out when the 9800GT is released or delayed).
- 1024MB on referance design, spread across 8 GDDR3 :)() 128MB chips. There will be no 2048MB variant for some time as a 256MB GDDR3 chips are far from mass production
- 240W TDP (!!!) per die. If my information holds true, Quad SLI will require (and I'm the guy who thinks I can SLI two 8800s on my 550W) PSUs above 1KW.
- Stock will be dual slotted cooling, and will require something of a no-nonsense design (perhaps like ATIs stock HD3870X2 cooler).
- 2nd core from NV to use an IHS.

Welcome to 3 days ago, and vr-zone didnt scoop anything. The original source is from xtremesystems :eek:

http://www.hardforum.com/showthread.php?t=1304792
 
Welcome to 3 days ago, and vr-zone didnt scoop anything. The original source is from xtremesystems :eek:

http://www.hardforum.com/showthread.php?t=1304792

yeah but they it was just some guy on a forum. We've got someone whos got something to lose claiming its real now.

But your right, its not much of a scoop. I looked around for anyone else who beat me to it, sorry I didn't find that thread.
 
that allen wrench looking thing may supposedly be the "turbo pwnage" switch ive heard so much about ;) ive heard it can push 300 fps in crysis 0_0
 
1) 125GB/s <--> 975MHz for memory..well pretty low clocks for memory, but they wouldn't have to use the most expensive GDDR3 at least.

2) QUAD SLI?
This is weird because that card would have two SLI connectors --> You could use three of these cards together --> TRI SLI or SEXTO SLI

3) It's less than 9800 GX2 on single core, but GT200 would have 45W higher TDP than 9800 GX2 --> GT200 has MUCH larger core and shader clocks than 9800 GX2

4) Why couldn't there be 2GB version of this card when there is MSI Geforce 9600 GT which has 2GB vram?

http://www.tomshardware.com/news/cebit-9600gt-2gb,4932.html
 
Please GOD, no lights visible on the card like the GX2's.
Please GOD, no lights visible on the card like the GX2's.
Please GOD, no lights visible on the card like the GX2's.
Please GOD, no lights visible on the card like the GX2's.
Please GOD, no lights visible on the card like the GX2's.
 
Quad SLI, no. Tri-SLI yes.

I'm going to be so tempted to pick up three of these to show the real power of my 780i.....

Would give me a reason to pick up the new Corsair PSU too.
 
ok so what psu would be the min for this, 600+? I'm interested in a single slot config.
 
* Each sector connected to a single 64bit memory controller, 512bit bus width total.
* 1024MB on reference design, spread across 8 GDDR3 () 128MB chips. There will be no 2048MB variant for some time as a 256MB GDDR3 chips are far from mass production.

These two "features" above are impossible, because they cannot co-exist on the same board.

Why? DDR3 modules are 32 bits wide. Eight modules means a maximum of 256-bit bus (8 * 32 = 256). Witness the fact that the GeForce 8800 GTX has 12 DDR3 chips to feed it's 384-bit (12 * 32) bus, and the Radeon HD 2900 XT has 16 DDR3 chips to feed the massive 512-bit (16 * 32) bus:

http://www.techpowerup.com/reviews/ATI/HD_2900_XT/4.html

The fact is, if you checked your own link, the card pictured has 16 chips! Here's the back view:

f_9900GTXbackm_074e18c.jpg


You can overlap multiple modules, but you must provide at least a minimum of the bus width addressed independently. And adding modules increases the cost of the card; combine that with the increase of building a complex PCB for that 512-bit bus, and Nvidia will hardly make any money on this card if it's got a 512-bit bus.

But I guess I can understand their flip-flopping back and forth (384-bit to 256-bit to 512-bit). The industry failed them with GDDR4, and GDDR5 is nowhere near available yet, so they have to do SOMETHING.

Everything else listed is quite realistic: double the pipes for roughly double the power of current chips, and stick to the 65nm process for the first revision.
 
Please GOD, no lights visible on the card like the GX2's.
Please GOD, no lights visible on the card like the GX2's.
Please GOD, no lights visible on the card like the GX2's.
Please GOD, no lights visible on the card like the GX2's.
Please GOD, no lights visible on the card like the GX2's.

Aww.. I think it looks hot. LEDs are sexy :p
 
So what will increasing the bus width to 512 exactly do for real world performance ?
 
So what will increasing the bus width to 512 exactly do for real world performance ?

It will likely help with very high resolutions + AA. Not sure what sort of real world gain there would be, however...
 
So what will increasing the bus width to 512 exactly do for real world performance ?

If you double the number of execution units, you will need to double the bandwidth to feed them properly with data. This is simple math.

With the 256-bit bus, current cards are stretched to the limit. the 384-bit bus on the 8800 GTX actually means it outperforms faster-clocked cards at higher resolutions. You can imagine what a boon more memory bandwidth will be if you increase the number of shader, ROP and texture units.

I really believed that we wouldn't see another 512-bit bus for a long time; I thought GDDR5 would be released this year, and solve our bandwidth needs through 2010 with the cheaper 256-bit bus. But it seems I was wrong; DDR3 is all the industry has, so they'll be brute-forcing things again.
 
These two "features" above are impossible, because they cannot co-exist on the same board.

Why? DDR3 modules are 32 bits wide. Eight modules means a maximum of 256-bit bus (8 * 32 = 256). Witness the fact that the GeForce 8800 GTX has 12 DDR3 chips to feed it's 384-bit (12 * 32) bus, and the Radeon HD 2900 XT has 16 DDR3 chips to feed the massive 512-bit (16 * 32) bus:

Your right, I forced the numbers to fit, rather then reading the words on the page, lol. Correct, its 16 chips, not 8. 64MB each, not 128.

The release date on this card is really up in the air. I've heard as late as september and as early as June.

The transistor count for this thing is apparently almost a billion, makes sense: 1 billion transistors, 65nm, not hard to imagine such a core producing 240W TDP.

ATI's HD4870 is rumored to have ~4GHz GDDR5 from samsung on a 256bit bus yielding near identical bandwidth to the specs I posted about the GT200 (double the frequency, half the width per cycle). I think both the GT200 and the RV770 are going to be real superb cards. We've already seen the questionable HD4870 vs 9800GX2 numbers, but I think those numbers might not be far from the truth.

The really awesome thing about all of this is ATI has apparently put the HD4870s MSRP at $299, and the HD3870X2 (TBA) at $499. Thats going to bring prices of the 8 and 9 series toppling down and should put the GT200 in even the reasonably cheap cheapskates ball park. If NVs expecting me to put $600 for one of the GT200s, then I won't even consider it.
 
While I would LOVE to see this card at $299 I HIGHLY doubt it. I expect it to have an MSRP of 549-599 with retailers adding an additional $50-100 in the early months. If ATI has a competitive product and undercuts nVidia, an MSRP of $499-549 wouldn't surprise me.

If my some miracle it is priced at $299, I'll buy two of them, atleast!
 
While I would LOVE to see this card at $299 I HIGHLY doubt it. I expect it to have an MSRP of 549-599 with retailers adding an additional $50-100 in the early months. If ATI has a competitive product and undercuts nVidia, an MSRP of $499-549 wouldn't surprise me.

If my some miracle it is priced at $299, I'll buy two of them, atleast!

Looks like they are going to be $499. Check the link in this thread...
 
I could careless about cost 400-600ish as long as it has 2x the performance of a 8800GTX I'll be happy. Games like Vanguard can then play in SLI at a sustained FPS of 45-60fps with all eye candy on and games like Crysis might reach their true potential.



 
Does anyone know based on the given images if existing waterblock models will fit on this card? I hate the noise so much on newer video cards I only use them if water cooled.
 
So what will increasing the bus width to 512 exactly do for real world performance ?

If your not SLI then not much. The current 256 bit bus is hitting a wall when three 9800GTXs or 2 9800GX2s (sometimes even 2 9800GTXs) are used at very high res / high AA. Let me see if I can pull some links up....



From the CoD4 page

9800 GTX vs. SLI vs. 3-way SLI. vs. Quad SLI @ [H]
GeForce 9800 GTX 3-way SLI


When we moved up to GeForce 9800 GTX 3-way SLI we experienced further performance improvements. We were able to play at 2560x1600 with 2X TRMS AA with very high framerates. However, when we tried to enable 4X AA at this resolution we ran into a severe bottleneck, the game went into the single digits in certain outdoor areas.

GeForce 9800 GX2 Quad SLI


The GeForce 9800 GX2 Quad SLI also ran into this same problem, not being able to run at 4X AA at 2560x1600 due to a severe performance drop.

For high AA and high res you will need the higher memory bandwidth. This really is a move that is needed only (for now) for the SLI user. Using the 256bit bus on the 9800GTX, 9600GT, 8800GT, 8800GTS is one of the reasons they could put out those cards at very very nice price points.
 

Beg to differ all you want. Companies have a thing about "always tell the optimal". Why? If you can generate interest in the item...you might be able to get money in advance to cover the amount of money you are bleeding into R&D. The 787 was supposed to be in revenue service already...that was the "ideal plan"; the engineers knew different. ;)

Right now the graphics world is very competitive...especially in the sub $200 range. Therefore even a $10 price reduction will yield signficant results to the companies bottom line. If nVidia is smart (and I think they are), I'm willing to bet the memory controllers can handle both GDDR3 and GDDR5. They will move over to the 5 when the $/performance ratio makes sense.
 
Beg to differ all you want. Companies have a thing about "always tell the optimal". Why? If you can generate interest in the item...you might be able to get money in advance to cover the amount of money you are bleeding into R&D. The 787 was supposed to be in revenue service already...that was the "ideal plan"; the engineers knew different. ;)

Right now the graphics world is very competitive...especially in the sub $200 range. Therefore even a $10 price reduction will yield signficant results to the companies bottom line. If nVidia is smart (and I think they are), I'm willing to bet the memory controllers can handle both GDDR3 and GDDR5. They will move over to the 5 when the $/performance ratio makes sense.

Nope... last part is wrong.... for their flagship cards they will move when they can still make a profit on a $500-600 card. I'm guessing that will be either this next card or the one following. Flagship cards don't care about $10 price reductions. This card is to demonstrate the raw performance Nvidia can offer, the cards that follow over the next 2 yrs after that flagship are the one's that may revert back to GDDR3 in order to save that $10.



 
While I would LOVE to see this card at $299 I HIGHLY doubt it. I expect it to have an MSRP of 549-599 with retailers adding an additional $50-100 in the early months. If ATI has a competitive product and undercuts nVidia, an MSRP of $499-549 wouldn't surprise me.

If my some miracle it is priced at $299, I'll buy two of them, atleast!

I'm expecting MSRP's to be around $499 USD... That's what some of my suppliers have told me to expect, and they've usually been right on the past with prices...

However, at a 299-399 price point, I would upgrade to a SLI system... and grab a 30" monitor.
 
Right now the graphics world is very competitive...especially in the sub $200 range. Therefore even a $10 price reduction will yield signficant results to the companies bottom line. If nVidia is smart (and I think they are), I'm willing to bet the memory controllers can handle both GDDR3 and GDDR5. They will move over to the 5 when the $/performance ratio makes sense.

I think that's a given. G80 also supported GDDR4, besides GDDR3, but GDDR4 never really offered many benefits over GDDR3, to justify the sometimes insane price difference from one model to another (i.e. HD 2900 XT 512 GDDR3 vs HD 2900 XT 1 GB GDDR4 which cost $100 more).
GT200 will no doubt support GDDR3 and GDDR5. GDDR5 may be reserved for an Ultra model.
 
I am VERY excited about this new card coming. It should be everything I was hoping the 9800GTX would be. I will be going SLi on this go round for sure. I will also be upgrading to Intel's brand new Bloomfield once it is out. Glad I waited to upgrade. You see it DOES actually pay to wait it out a gen or so. So many people say not to wait that waiting is dumb...they are wrong.
 
Back
Top