NVIDIA GF100 & Friends speculation

[jimbo75]

/HUGE_sigh...

No Display Port makes me a very sad panda. Means it's unlikely for ultra thin Direct Drive monitors to appear anytime soon.

Regards,
SB

I think that was expected.

ATI have done a lot more towards displayport over the past couple of years while nVidia basically ignored it

nVidia also said their eyefinity would be done in software but requiring two cards in sli, which was another way of them saying there will be no displayport with fermi.

 

[Sontin]

I think that was expected.

ATI have done a lot more towards displayport over the past couple of years while nVidia basically ignored it

nVidia also said they're eyefinity would be done in software but needing two cards in sli, which was another way of them saying there will be no displayport with fermi.

Eyfinity is software, too.
It's not possible to use more than two active display with a nvidia card at the same time. Two cards give you more than these two active displays.

 

[FrameBuffer]

Eyfinity is software, too.
It's not possible to use more than two active display with a nvidia card at the same time. Two cards give you more than these two active displays.

I think what he is saying is that NV's Surround Gaming is a pure software solution (think SoftTH) where as ATI has distinct hardware changes that allow it to run >2 displays per card.. One $150 card (5770) will easily drive 3 1024/1050 displays where as NV would require at least $400-500 to accomplish the same performance.

Edit: it's not just that NV can't drive more than 2 displays at once per chard (thus needing a 2nd card) their solution is software only based and needs SLI to deliver the same level of performance.

 

[Florin]

Something Awful at it again.

If you take a look at the picture above, you’ll notice the following text “with a minimum 12V current rating of 42A”. This means that those that have a PSU with split 12V rails won’t be able to run a Fermi card, no matter if it’s a 600W or a 1.21GW (ok, there are no such power supplies, but hey). There’s of course a solution to the problem, buy a new PSU with a single rail and there are several PSUs on the market that has a single 12V rail that will deliver 42A or more.

I believe this may be a misrepresentation. If the 2 PCIe connectors plugged into the card are on separate rails that are able to provide 42A combined it should also work. The max rating for each of the 3 +12V pins on 6 and 8-pin PCIe connectors is 9A afaik, which would mean that something like 1 rail with 24A and 1 with 18A should be sufficient.

 

[Jawed]

That looks like a nonsense article to me - one that's been called out in the site's forum, too.

Changing the subject slightly, the "SLI Certified for Fermi" cases don't seem to have made themselves known at the show as far as I can tell.

Jawed

 

[L233]

Something Awful at it again.

I believe this may be a misrepresentation. If the 2 PCIe connectors plugged into the card are on separate rails that are able to provide 42A combined it should also work. The max rating for each of the 3 +12V pins on 6 and 8-pin PCIe connectors is 9A afaik, which would mean that something like 1 rail with 24A and 1 with 18A should be sufficient.

8-pin = 150W / 12V = 12,5 A
6-pin = 75W / 12V = 6,25 A
PCI-E Slot = 75W / 12V = 6,25 A

Am I missing something?

 

[Sontin]

I think what he is saying is that NV's Surround Gaming is a pure software solution (think SoftTH) where as ATI has distinct hardware changes that allow it to run >2 displays per card.. One $150 card (5770) will easily drive 3 1024/1050 displays where as NV would require at least $400-500 to accomplish the same performance.

"Eyefinity" is marketingname. You need more than two active displays for it. Even for SoftTH you need two GPUs in your system (i.e. a discret card and a igp). I don't know but maybe it is possible to split a signal into two...

Edit: it's not just that NV can't drive more than 2 displays at once per chard (thus needing a 2nd card) their solution is software only based and needs SLI to deliver the same level of performance.

No, that's wrong. It has nothing to do with the performance.

 

[jimbo75]

I think what he is saying is that NV's Surround Gaming is a pure software solution (think SoftTH) where as ATI has distinct hardware changes that allow it to run >2 displays per card.. One $150 card (5770) will easily drive 3 1024/1050 displays where as NV would require at least $400-500 to accomplish the same performance.

Edit: it's not just that NV can't drive more than 2 displays at once per chard (thus needing a 2nd card) their solution is software only based and needs SLI to deliver the same level of performance.

Yes that is pretty much what I was alluding to.

If Fermi had a displayport output, chances are it wouldn't require SLI for their surround gaming to work.

I do feel ATI's groundwork with displayport really helped to drive them towards eyefinity sooner than the industry expected to see such a feature. Nvidia's method does look to be more of a panic measure, and I would expect them to go with 3 active outputs per card in the next round.

 

[Sontin]

Yes that is pretty much what I was alluding to.

If Fermi had a displayport output, chances are it wouldn't require SLI for their surround gaming to work.

That's wrong, too. Look at their quadro cards: They have a DP but you can only use 2 of 3 digital displays at the same time: http://www.nvidia.com/object/product_quadro_fx_5800_us.html
The quadro nvs has two gpus: http://www.nvidia.com/object/product_quadro_nvs_450_us.html

 

[FrameBuffer]

"Eyefinity" is marketingname. You need more than two active displays for it. Even for SoftTH you need two GPUs in your system (i.e. a discret card and a igp). I don't know but maybe it is possible to split a signal into two...



No, that's wrong. It has nothing to do with the performance.

IIRC, Eyefinity can do 2x1 displays and NV's surround vision (outside of Fermi/GF100) requirement of SLI has a LOT to do with Performance more so to do with the lack of display outputs but performance is a very limiting factor.. unless you seem to think that 2 9600GTs are going to drive 3x1680x1050 with 2XAA and 16AF @ 60FPS or 5760x1200 at a solid 35+ fps. Nv's solution iirc is akin to SoftTH with a substantial penalty applied across multiple display/cards. A GTX275 in SLI should drive a setup with satisfactory results.. I imagine the GTX260 would be the bare minium for near HD resolutions and 88000/9800/250 in Sli (not sire if tri-sli is to be enabled) could deliver 3x1280x1024 with minimal performance

 

[Silent_Buddha]

IIRC, Eyefinity can do 2x1 displays and NV's surround vision (outside of Fermi/GF100) requirement of SLI has a LOT to do with Performance more so to do with the lack of display outputs but performance is a very limiting factor.. unless you seem to think that 2 9600GTs are going to drive 3x1680x1050 with 2XAA and 16AF @ 60FPS or 5760x1200 at a solid 35+ fps. Nv's solution iirc is akin to SoftTH with a substantial penalty applied across multiple display/cards. A GTX275 in SLI should drive a setup with satisfactory results.. I imagine the GTX260 would be the bare minium for near HD resolutions and 88000/9800/250 in Sli (not sire if tri-sli is to be enabled) could deliver 3x1280x1024 with minimal performance

I believe their software based ultra widescreen is limited to GT200 and up (http://blogs.nvidia.com/ntersect/2010/01/surround-yourself-with-nvidia-3d-vision-technology.html), so performance is already taken into consideration. Nvidia just aren't going to let you run it on less powerful hardware.

And as has been noted by ATI multiple times. Eyefinity required both hardware specific changes and software in order to enable it to perform at speed.

Regards,
SB

 

[Sontin]

IIRC, Eyefinity can do 2x1 displays and NV's surround vision (outside of Fermi/GF100) requirement of SLI has a LOT to do with Performance more so to do with the lack of display outputs but performance is a very limiting factor..
unless you seem to think that 2 9600GTs are going to drive 3x1680x1050 with 2XAA and 16AF @ 60FPS or 5760x1200 at a solid 35+ fps. Nv's solution iirc is akin to SoftTH with a substantial penalty applied across multiple display/cards. A GTX275 in SLI should drive a setup with satisfactory results.. I imagine the GTX260 would be the bare minium for near HD resolutions and 88000/9800/250 in Sli (not sire if tri-sli is to be enabled) could deliver 3x1280x1024 with minimal performance

What's the point?
They need two gpus for more than two active displays. So, they will use SLI for this. SoftTH needs two gpus, too. The difference is that only one GPU is used for the rendering.

 

[FrameBuffer]

I believe their software based ultra widescreen is limited to GT200 and up (http://blogs.nvidia.com/ntersect/2010/01/surround-yourself-with-nvidia-3d-vision-technology.html), so performance is already taken into consideration. Nvidia just aren't going to let you run it on less powerful hardware.

And as has been noted by ATI multiple times. Eyefinity required both hardware specific changes and software in order to enable it to perform at speed.

Regards,
SB

ahh I thought NV was going to allow pre-GF200 (8800+) to do Surround Vision.. and yes I know that Eyefinity has hardware and software changes that allow it to run multiple displays at a reduced performance penalty (in relation to a pure software/SoftTH comparison). I was simply trying to point out that the perf penalty (along with the pocket book/ka-ching penalty) for Eyefinity is much less than competitor's products and that sunspot entails much more than "software" and should allow 2x1 output.. though it would bring up targeting issues etc..

sorry for the rambling.. feeling under the weather and the nyquil hasn't fully kicked in yet

What's the point?
They need two gpus for more than two active displays. So, they will use SLI for this. SoftTH needs two gpus, too. The difference is that only one GPU is used for the rendering.

are we sure that NV's Surround Vision only uses 1 GPU for rendering and others simply act as dummies, used only as display outputs rather than harvesting the potential speed increase ?

 

[Silent_Buddha]

are we sure that NV's Surround Vision only uses 1 GPU for rendering and others simply act as dummies, used only as display outputs rather than harvesting the potential speed increase ?

Sort of, from what I've read for the Nvidia setup, one GPU will render the middle screen and the second GPU will render both of the surround screens. There's a lot of questions on how this unbalanced rendering load will be handled.

Oh and the bit about hardware specific to enabling and facilitating Eyefinity was a general response to the thread not you. I have a bad habit of starting a response to a specific person and then throwing in general responses at the end.

Regards,
SB

 

[Florin]

8-pin = 150W / 12V = 12,5 A
6-pin = 75W / 12V = 6,25 A
PCI-E Slot = 75W / 12V = 6,25 A

Am I missing something?

No, the card shouldn't use more than what the 3 of those connectors can safely provide (although I guess this has been debatable before, at least with the 4870x2 and GTX295). What the recommendation seems to say that if there is a single +12v rail then it ought to be able to deliver 42A. Because obviously the motherboard, CPU connector, SATA and molex connectors will also be drawing current on that rail. With PSUs that do have split rails, the rails with PCIe connectors must not have other components on them which leave less than the numbers you list.

Which is basically why this story is so oversimplified that it isn't useful, except to further the site's agenda.

 

[Dave Baumann]

If Fermi had a displayport output, chances are it wouldn't require SLI for their surround gaming to work.

Not necessarily the case. Irrespective of DP support or not what make Eyefinity possible (along with DP needing only one clock soure) its really increase in display pipelines from 2 to 6. Had the display pipes not increased, irrespective of DP we'd still be limited to two displays.

Technically RV7xx can drive 6 DP outputs, just only two of them could be independant.

 

[aaronspink]

And as has been noted by ATI multiple times. Eyefinity required both hardware specific changes and software in order to enable it to perform at speed.

Hardware should be pretty much additional display scan engines on chip to pull the framebuffer data and output it to an output. There have been a lot of card with >2 display support, but only 2 independent displays.

The software part is probably just programming existing functionality in the display scan engines for windowing the frame buffer, plus driver/CCC support for configuring the displays. Which means that the software side really is minimal, as I understand it the graphics device presents a single XxY render target/frame to the application. A lot of the reduced slow down probably comes from the this as all the geometry work only needs to be done once.

 

[Rys]

So it's about the same size as GT200b ?

I can never remember how big GT200b is. I'm pretty convinced GF100's < 500mm/sq at any rate.

 

[LordEC911]

I can never remember how big GT200b is. I'm pretty convinced it's < 500mm/sq at any rate.

~470-480mm2.
So, three weeks away from launching and we still have two very different numbers on die size from two very good sources. Crazyness.

 

[Ailuros]

Something Awful at it again.
I believe this may be a misrepresentation. If the 2 PCIe connectors plugged into the card are on separate rails that are able to provide 42A combined it should also work. The max rating for each of the 3 +12V pins on 6 and 8-pin PCIe connectors is 9A afaik, which would mean that something like 1 rail with 24A and 1 with 18A should be sufficient.

The article is still nonsense either way you turn it. My 700W PSU at home has 4*+12V (18A) each and a combined power of 48A. A GTX295 has a minumum requirement of a 650W PSU and 46A.