The First Single-PCB GTX295 From Inno3D Unearthed

[nitromullet]

Linkage:

http://en.expreview.com/2009/0...earthed.html#more-3470 (this link is dead)

http://en.expreview.com/2009/0...-gtx295-unearthed.html (new link with Inno3D branding removed from the cards.)

This was originally posted on page 9 of this thread, but it seemed like interesting enough news to deserve its own thread.

Lots of good pics of the card. Although, honestly, I find it hard to justify a GTX 295 for $550-600 when dual GTX 260s can easily be found for $350-400 without rebates.

 

[Idontcare]

NF200 chip can support PCI-E 2.0, and has 48 built-in PCI-E channels to connect one group of 16X channel with external PCI-E 16X interface, and connect the two GPUs with 16X channels. As soon as the data enters NF200, the driver of graphics card will distribute it to GPU in charge according to preset algorithms.

An onboard NF200? Doesn't that thing actually slow down the fps versus a non-NF200 SLI rig setup? Or do I have have the wrong understanding of what the NF200 is/does?

 

[nitromullet]

The NF200 is NVIDIA's PCIe 2.0 bridge chip. It was added alongside nForce 680i to make the 780i, and was actually integrated onto the die itself for 790i. The dual gpu cards need the bridge chip to work. The 4870/50 X2 cards have one as well, only it's a third party chip.

 

[Cookie Monster]

Originally posted by: Idontcare

NF200 chip can support PCI-E 2.0, and has 48 built-in PCI-E channels to connect one group of 16X channel with external PCI-E 16X interface, and connect the two GPUs with 16X channels. As soon as the data enters NF200, the driver of graphics card will distribute it to GPU in charge according to preset algorithms.

An onboard NF200? Doesn't that thing actually slow down the fps versus a non-NF200 SLI rig setup? Or do I have have the wrong understanding of what the NF200 is/does?

By maybe 1~2 fps. The real downside to using these chips is that they are literally space heaters.

 

[OCGuy]

Originally posted by: Cookie Monster

Originally posted by: Idontcare

NF200 chip can support PCI-E 2.0, and has 48 built-in PCI-E channels to connect one group of 16X channel with external PCI-E 16X interface, and connect the two GPUs with 16X channels. As soon as the data enters NF200, the driver of graphics card will distribute it to GPU in charge according to preset algorithms.

An onboard NF200? Doesn't that thing actually slow down the fps versus a non-NF200 SLI rig setup? Or do I have have the wrong understanding of what the NF200 is/does?

By maybe 1~2 fps. The real downside to using these chips is that they are literally space heaters.

You should sit in my room for a while :laugh:

 

[Idontcare]

Originally posted by: Cookie Monster

Originally posted by: Idontcare

NF200 chip can support PCI-E 2.0, and has 48 built-in PCI-E channels to connect one group of 16X channel with external PCI-E 16X interface, and connect the two GPUs with 16X channels. As soon as the data enters NF200, the driver of graphics card will distribute it to GPU in charge according to preset algorithms.

An onboard NF200? Doesn't that thing actually slow down the fps versus a non-NF200 SLI rig setup? Or do I have have the wrong understanding of what the NF200 is/does?

By maybe 1~2 fps. The real downside to using these chips is that they are literally space heaters.

So why do they use them? If it doesn't improve performance, it just increases cost and power-consumption?

I must have missed the boat entirely, I thought the NF200 requirement for mobos is gone (replaced with a SLI licensing fee structure) but why incorporate a seemingly needless IC onto a single-PCB dual-GPU system?

What possible purpose could it serve? I'm just scratching my head here.

 

[Zap]

Originally posted by: Idontcare
What possible purpose could it serve?

Uh, it allows the two GPUs to talk to each other?

Dunno why y'all are getting all excited about some pics based off partially functional engineering samples. We all know these things are coming.

Reminds me of other companies pre-announcing products that never seem to show up for sale. Of course sometimes the truth is that there are products and there is vaporware.

 

[Idontcare]

Originally posted by: Zap

Originally posted by: Idontcare
What possible purpose could it serve?

Uh, it allows the two GPUs to talk to each other?

If I have a SLI certified mobo w/o NF200, and I SLI two NV cards together, do I need to insert a NF200 chip somewhere in there to allow the two SLI'ed GPU's to talk to each other?

 

[nitromullet]

Originally posted by: Idontcare

Originally posted by: Zap

Originally posted by: Idontcare
What possible purpose could it serve?

Uh, it allows the two GPUs to talk to each other?

If I have a SLI certified mobo w/o NF200, and I SLI two NV cards together, do I need to insert a NF200 chip somewhere in there to allow the two SLI'ed GPU's to talk to each other?

No, but this is "SLI" on a separate PCB. The NF200 supplies the gpus the PCIe 2.0 lanes that allows them to communicate with each other and the single PCIe 2.0 slot they share. Think of the GTX 295 or 4870 X2 as a "daughter card" that adds extra PCIe 2.0 lanes for the gpus via a bridge chip.

edit: the 4870 X2 uses a PLX PCIe chip to do the same thing. http://www.plxtech.com/products/expresslane/bridges.asp

 

[Tempered81]

idc, i thought x58's had the nf200 chip on them?

is that not true?

 

[nitromullet]

Originally posted by: jaredpace
idc, i thought x58's had the nf200 chip on them?

is that not true?

nope. That was NVIDIA's plan, but no one would add the unnecessary chip to the X58 boards. NV eventually agreed to "SLI certification" for some X58 boards. That is the reason why some X58 boards are SLI compatible, and some are not. There is no actual technical reason for any motherboard with dual PCIe 16x slots not to support SLI and Crossfire.

 

[Tempered81]

so you could bios update an dual pci-e x58 board to make use of sli video cards, even if it's not "certified" by nvidia? Or do all dual and triple slot Pci-E X58 mobos support sli?

 

[nitromullet]

Originally posted by: jaredpace
so you could bios update an dual pci-e x58 board to make use of sli video cards, even if it's not "certified" by nvidia? Or do all dual and triple slot Pci-E X58 mobos support sli?

I'm not sure exactly how it works, but basically SLI certification just means that the NV driver recognizes the motherboard as being certified and allows SLI to be enabled. I'm not sure if you can flash the BIOS to perform a "DYI SLI certification".

 

[imported_Shaq]

Originally posted by: jaredpace
so you could bios update an dual pci-e x58 board to make use of sli video cards, even if it's not "certified" by nvidia? Or do all dual and triple slot Pci-E X58 mobos support sli?

It's doable. My Gigabyte UD3R didn't originally support SLI until they added a BIOS update on 3/31. But I think it is the same board as the UD3R-SLI just without an SLI bridge. How many X58 boards don't support SLI? I imagine it wouldn't be many.

 

[lopri]

I hear that the SLI licensing fee is $5 per board. Each design should be validated by NV for compatibility then NV will give some sort of 'key' for the board's BIOS, upon which the OS/drivers recognize the board as SLI-ready. (sounds almost like a DRM scheme. Heh)

 

[james1701]

When they first started talking about doing it without the 200 chip they said there would be a small performance loss. I have not seen any testing to confirm this. I have been waiting on some cracked bios's for my P45 board. If it can be done with the X58, then it can be done with any board.

 

[Idontcare]

Originally posted by: james1701
When they first started talking about doing it without the 200 chip they said there would be a small performance loss. I have not seen any testing to confirm this. I have been waiting on some cracked bios's for my P45 board. If it can be done with the X58, then it can be done with any board.

The testing I've seen proves the opposite, the NF200 just gets in the way of an already fully capable SLI topology and needlessly adds latency to it.

I think it was HardOCP that did a review with two otherwise identical SLI rigs (same mobo except one had NF200 while the other was sans NF200) and showed this. At least that's how I remember it.

I really don't understand why two GPU's can talk to each other just fine on a non-NF200 mobo but when you stick those same GPU's on the same PCB then suddenly they can't communicate with one another unless you shove a NF200 chip in there.

I guess it is just the perceived (by me) illogic of it all that bothers me so much, I really could care less where and how NF200 is used provided it really is adding value in some way. I just haven't read a review that concluded as much. Hence my question in the first post I made in this thread.

 

[james1701]

One remote possibility is the boards bios chip does not have enough ram for the sli option in addition to the regular programming, and it would cost more to add a different bios chip than add a NF200 chip to the board.

Would a video bios differ that much from a mobo bios to the point it could not be programmed the same way to communicate to both processors?

 

[nitromullet]

Originally posted by: Idontcare
I guess it is just the perceived (by me) illogic of it all that bothers me so much, I really could care less where and how NF200 is used provided it really is adding value in some way. I just haven't read a review that concluded as much. Hence my question in the first post I made in this thread.

As I understand it, because the dual gpu card essentially is two complete video cards in one, it needs a bus for data to travel on. In a standard dual video card + motherboard situation, the bus for the data to travel on between the two cards is provided by the chipset.

I can't find a diagram for the GTX 295, but here's how it works for the 4870 X2:

http://www.anandtech.com/video/showdoc.aspx?i=3372&p=3

They are actually showing the Sideport (or lack thereof), but they show the role of the PCIe bridge role in as well.

 

[Tempered81]

Here is Saaya from XS's "confidential" photo of a single PCB GT300 GX2 Prototye (It's not real ;p). He is poking fun at how absurd the PCB layout is on the Nvidia cards compared to the organized streamlined 4870x2 and 4890 reference PCB's:

http://img219.imageshack.us/img219/6229/dualgt300.jpg

mounting holes all over the board
a mounting hole on the Pcie slot, and one on the sli goldfinger
a smileyface of mounting holes to the right of the 2nd gpu
GDDR3 ram all over the place
DVI on the top of the card, and rear
PCI-E power connectors on the bottom of the card
GPUs mis-matched and not aligned ....

heheh

 

[eklock2000]

These are better...photos of the GTX390...

http://www.extrahardware.cz/ex...orce-gtx-390-v-redakci

LOL

 

[MrK6]

Ouch, I'd hate to see what kind of temperatures this thing is going to run at. Hopefully, if this design is going to be cheaper to produce there will be a decent supply available at a decent price. The fact that GTX 295's are currently $50 more expensive than when I bought mine over five months ago is absurd.

 

[Keysplayr]

Originally posted by: Idontcare

Originally posted by: james1701
When they first started talking about doing it without the 200 chip they said there would be a small performance loss. I have not seen any testing to confirm this. I have been waiting on some cracked bios's for my P45 board. If it can be done with the X58, then it can be done with any board.

The testing I've seen proves the opposite, the NF200 just gets in the way of an already fully capable SLI topology and needlessly adds latency to it.

I think it was HardOCP that did a review with two otherwise identical SLI rigs (same mobo except one had NF200 while the other was sans NF200) and showed this. At least that's how I remember it.

I really don't understand why two GPU's can talk to each other just fine on a non-NF200 mobo but when you stick those same GPU's on the same PCB then suddenly they can't communicate with one another unless you shove a NF200 chip in there.

I guess it is just the perceived (by me) illogic of it all that bothers me so much, I really could care less where and how NF200 is used provided it really is adding value in some way. I just haven't read a review that concluded as much. Hence my question in the first post I made in this thread.

I could be mistaken, but I think the NF200 chips were only required for boards expected to support "Tri-SLI". The boards for standard SLI, did not. Any arguments? I'm not 100% on this. It's been a while.

 

[nitromullet]

Originally posted by: Keysplayr

I could be mistaken, but I think the NF200 chips were only required for boards expected to support "Tri-SLI". The boards for standard SLI, did not. Any arguments? I'm not 100% on this. It's been a while.

You are correct, although NVIDIA actually supports three-way SLI (16x/8x/8x) without the NF200 as well:

http://www.anandtech.com/cpuch...el/showdoc.aspx?i=3395

 

[nitromullet]

Originally posted by: jaredpace
Here is Saaya from XS's "confidential" photo of a single PCB GT300 GX2 Prototye (It's not real ;p). He is poking fun at how absurd the PCB layout is on the Nvidia cards compared to the organized streamlined 4870x2 and 4890 reference PCB's:

http://img219.imageshack.us/img219/6229/dualgt300.jpg

mounting holes all over the board
a mounting hole on the Pcie slot, and one on the sli goldfinger
a smileyface of mounting holes to the right of the 2nd gpu
GDDR3 ram all over the place
DVI on the top of the card, and rear
PCI-E power connectors on the bottom of the card
GPUs mis-matched and not aligned ....

heheh

I fail to see how http://www.techpowerup.com/rev...70_X2/images/front.jpg is more "clean and streamlined" than http://en.expreview.com/img/20...5/12/GTX295_card_3.jpg

Sorry, but this sounds like like stupid troll BS to me.