Nvidia's Hybrid SLI

[Arun]

Maybe I didn't make myself clear... When I said 'a full G98', I meant 4ROPs/8TMUs/16SPs. Given both G98 and MCP78's die sizes (at least relative to G92), you'd be completely nuts to think the chips themselves had less than that.

My guess is that current benchmarks are on the cut-down version of the chips. I'm even nearly willing to bet MCP78U on a HT3 motherboard with a HT3 CPU (or MCP7A on any Intel CPU) could score ~3000 in 3DMark05. If it can't then, well, there's a serious problem here.

 

[AnarchX]

Yes, I agree to you.
MCP78 has to be much faster than current results show, to be competitive against RS780, which seems to be able to reach ~1200+ points in 3DM06 HDR/SM3.0:
http://enthusiast.hardocp.com/article.html?art=MTQzMiwxLCxoZW50aHVzaWFzdA==
MCP78 @PCo only ~400

 

[DavidC]

Arun, then tell me why is there a 60% difference between the G86 and G98 at the same clock??: http://en.expreview.com/?p=77&page=13

Granted, GPU-Z shows the amount of shaders as same for both, which is 16. But then G86 gets 1614 and G98 gets 1080 in 3dmark06 when adjusted for same clock. That tells me there's something else that's different between the two.

The site claims:
"G98 only have 8 Stream Processors, 4 Texture Filtering Units, 4 Texture Addressing Units and 4 ROPs. The core frequency is running at 567MHz, memory frequency goes at 500MHz and shader frequency at 1400MHz."

AnanchX: Yes, I agree to you.
MCP78 has to be much faster than current results show, to be competitive against RS780, which seems to be able to reach ~1200+ points in 3DM06 HDR/SM3.0:
http://enthusiast.hardocp.com/articl...50aHVzaWFzdA==
MCP78 @PCo only ~400

What's Anandtech's claims about 3-4x in 3dmark06 compared to 690G mean then?? 690G gets low-300 in 3dmark06. 3-4x would put the score at 1000-1200.


And about Intel: Intel claims 3x the G33/GMA 3100 in 3dmark06 for the GMA X4500. The problem is that with 3dmark03/05/06, the driver runs them on software T&L/VS. The more relevant comparison is the comparison against the G35, which they say it'll be 1.7x. And from my experience with the G965, software results in 3dmark is only comparable with games running in software mode. How do we explain then?? Maybe the X4500 has 4 ROPs and 8 TMUs?? It also has 10 EUs, which is 20 ALUs. From that I don't think the X4500 will be weak at all hardware-wise.

 

[Arun]

First of all, the overall 3DMark06 score is useless. It's based on both the CPU and the GPU scores and the formula isn't even linear; so 4x the 3DMark06 score would not imply 4x the GPU score. This is why I used 3DMark05 for my target score...

Secondly, I believe those specifications (and benchmarks) are for a cut-down G98, which will likely be called GeForce 8200. I'm seriously not sure why anyone should believe the contrary, even more so when those leakers who claim that's the full chip are the same morons who take GPU-Z seriously.

 

[DavidC]

http://play.tm/wire/cluster/1655280

What, and the Geforce 8200 name rumor for the IGP is just a coincidence?? After saying that you don't believe there will be such a pitiful version of G98 with only 8 shaders, now you are saying there's two versions. Geforce 4 MX IGP was between the Geforce 4 MX 420 and the Geforce 4 MX 440. I am thinking that the MCP78's fastest version will be between the two lowest IGPs based on Geforce 8.

We will see.

 

[Arun]

Sigh, you're really slow, aren't you? I never ever said I don't believe there will be a 8 shaders version of G98. Heck, there was one of G86: the 8300GS (but that still had all the TMUs). What I'm saying is the *full chip* isn't that pitiful, when nothing is disabled.

http://www.ixbt.com/short/images/edj8zgs6d5cped7.jpg
'Best' is 8400, 'Better' is 8200. Current rumours and benchmarks have been on the 8200. No G98s or MCP78s based on the 8400 have leaked out in the wild yet. This is what I've been telling you in a bunch of replies already - I'm not sure what your problem is...

 

[DavidC]

Sigh, you're really slow, aren't you? I never ever said I don't believe there will be a 8 shaders version of G98. Heck, there was one of G86: the 8300GS (but that still had all the TMUs). What I'm saying is the *full chip* isn't that pitiful, when nothing is disabled.

http://www.ixbt.com/short/images/edj8zgs6d5cped7.jpg
'Best' is 8400, 'Better' is 8200. Current rumours and benchmarks have been on the 8200. No G98s or MCP78s based on the 8400 have leaked out in the wild yet. This is what I've been telling you in a bunch of replies already - I'm not sure what your problem is...

Sorry, maybe its my fault that I didn't read all of the 90 or so posts in the 4 pages. But you said that its still not clear on the MCP78 specifications.

If that is the case and it's 85mm², then G98 is literally the most pitiful GPU ever. I'm not sure I believe that; it's not impossible, but just like I'll need some proof to believe something is insanely good, I'll also need some proof to believe somrthing could possibly be THAT bad... My understanding is that this slide (on MCP79, but same thing) indicates what's happening here: you've got a SKU with the full GPU, and another SKU with GPU redundancy. Maybe the latter will even be called "GeForce 8400", who knows.

I wouldn't be surprised if, just like what they did with MCP72, NVIDIA redesigned the TMUs to work as half-quads; that way, they can not only disable half the ALUs, but also half the TMUs. Yes, I know TMUs are currently 8-wide, but I still suspect *part* of them is double-pumped (presumably the integer ALUs for ADDs/MULs at least, which take part in both addressing and filtering).

From the above it sounds like you are speculating. And your speculation is that the current one is based off a 8200 or whatever part and the "full" version is a 8400 based part. Then how can you even presume that the "full" IGP isn't out yet?

Do you already her something about G98-8400GS, which appears in newest FW:
Quote:
NVIDIA_G98.DEV_06E2.1 = "NVIDIA GeForce 8400"
?

It should be weaker than G98-8400GS?

So MCP78 should not be fast as I earlier thought, because MPC78 provide a 100% boost-up combined with "GF 8400".

I woud guess: 2 ROPs/4 TMUs/8SP physically for MCP78/7A.

More speculations, and one that says that the MCP78 is weak. Explain the 3dmark06 difference between OCWorkBench's MCP78 and PCOnline's MCP78. And What's this picture say?: http://www.pconline.com.cn/diy/main/reviews/0712/1189121.html

I see MCP78U-A1, which obviously means the Ultra...

 

[Arun]

No matter, you'll see on January 28th...

 

[INKster]

So, MCP78S is the "Geforce 8100", while MCP78U is the "Geforce 8300", according to this small remark.
Any significant difference in terms of ALU/ROP/shader unit configuration, or is it just a matter of higher clocks on the beefier part, Arun ?

 

[Arun]

Ahhh, so the nomenclature is:
- G98: 8400 (8 TMUs/16 SPs/4 ROPs)
- Half G98: 8200 (4 TMUs/8 SPs/? ROPs)
- MCP78: 8300 (8 TMUs/16 SPs/4 ROPs)
- Half MCP78: 8100 (4 TMUs/8 SPs/? ROPs)

Makes sense, I guess. As for clocks or the number of ROPs for the cut-down versions, I don't know about that, sorry - and I won't try to speculate since I see David doesn't take me very seriously when I do that! (perhaps not wrongfully though)

 

[AnarchX]

Ahhh, so the nomenclature is:
- G98: 8400 (8 TMUs/16 SPs/4 ROPs)

But why is G98 8400GS@ 567/1400/500MHz slower than G86 8400GS:
http://en.expreview.com/?p=77&page=13

I have doubts that it is a G86@64Bit.
The question is how much die-space costs VP3 or is D3D10.1 allready on-board, to sell it with a bit higher clocks as GF9300 or something.

 

[Arun]

And we should believe the final name for that SKU is '8400GS'... why? Because GPU-Z or a random driver string which was probably put in there months ago says so? I'm tired of this, so I'll just repeat the same thing again: January 28th, or whenever MCP78 will officially be announced.

 

[Speccy]

And we should believe the final name for that SKU is '8400GS'... why? Because GPU-Z or a random driver string which was probably put in there months ago says so?

Do we have any evidence to point to anything contrary at the moment? Have you downloaded the latest driver and showing that the G98 8400 GS is showing a different driver device id?

[Edit] Here are the Device ID's for G98 from the latest 169.21 drivers:

NVIDIA_G98.DEV_06E2.1 = "NVIDIA GeForce 8400"
NVIDIA_G98.DEV_06E4.1 = "NVIDIA GeForce 8400 GS"

 

[AnarchX]

And we should believe the final name for that SKU is '8400GS'... why?

Because of this?

Sure G98 with its 85mm²@65nm should be 4ROPs/64Bit+8TMUs+16SPs, but there are still some factors: VP3, D3D10.1?, manifacturing at UMC...

But sure, lets wait what will happen on 19th...

 

[INKster]

Because of this?

Sure G98 with its 85mm²@65nm should be 4ROPs/64Bit+8TMUs+16SPs, but there are still some factors: VP3, D3D10.1?, manifacturing at UMC...

But sure, lets wait what will happen on 19th...

Be advised that certain strings on some early forceware drivers called the G98 by the formal name of "Geforce 9300", and that even the old 6200 TC was later renamed to Geforce 7100 GS, so a relatively old internal slide may not mean much.
The two "G98's", Geforce 8400 GS and Geforce 9300, might even coexist in the market with disabled parts in either to further differentiate them in the eyes of end-users.
It won't be long now.

 

[AnarchX]

http://hkepc.com/?id=551

GeForce 9200? D3D10.1-support?

But no support for the interesting smart power feature?

 

[INKster]

Although several motherboards carrying it have surfaced already, this is, i believe, the first photo of MCP78's die.

Source: http://my.ocworkbench.com/bbs/showthread.php?p=424798#post424798

For a single chip solution with built-in Geforce 8 class DX10 IGP core and brand new VP3 video processor it doesn't seem to be that big to me..., 65nm perhaps ?

 

[DavidC]

Oh, I just saw why PCOnline's and OCWorkbench's 3dmark06 for MCP78 was different. One runs at 800x600 and the other runs at 1024x768.

EDIT: NVM, lol.

 

[Blazkowicz]

Although several motherboards carrying it have surfaced already, this is, i believe, the first photo of MCP78's die.

Source: http://my.ocworkbench.com/bbs/showthread.php?p=424798#post424798

For a single chip solution with built-in Geforce 8 class DX10 IGP core and brand new VP3 video processor it doesn't seem to be that big to me..., 65nm perhaps ?

damn that would be the perfect mobo but I want a RS232 port . even it that means no VGA connector and the use of a dongle.
the number of outputs has me wondering, does it support dual screen? (not that I really need it)

 

[Speccy]

http://www.digitimes.com/mobos/a20071231PD208.html

Seems like MCP78 is having a tough time keeping up with RS780. It may even be that we'll not be seeing it on the 28'th afterall.