The Official IGP Rumours & Speculation Thread

[Arun]

[THIS POST IS EARLY WORK IN PROGRESS]
Please feel free to already begin posting in this thread, though.

This thread aims to focus on the upcoming IGP products that are rumoured to be released within the next 12 months. It will be divided or replaced if needed, based on the volume of contributions and/or a massive focus on one specific chip and/or SKU.

Rumoured Upcoming IGPs
- SiS Mirage 3+ (Q107, DX9, Intel)
- NVIDIA MCP68 (Q207, DX9, AMD)
- NVIDIA MCP73 (Q207, DX9, Intel)
- Intel G33 (Q207, DX9, Intel)
- Intel G31 (Q307, DX9, Intel)
- Intel G35 (Q307, DX10, Intel)
- AMD RS740 (Q307, DX9, AMD)
- AMD RS790? (Q307 or Q406, DX10, AMD)
- AMD RS780? (H108, DX10, AMD)
- SiS Mirage 4 (Q307, DX10, Intel)
- VIA IGP (?) (Q307 or Q407, DX10, Intel)
- VIA IGP (?) (Q307 or Q407, DX10, AMD)
- SiS Mirage 4 (Q407, DX10, AMD)
- NVIDIA MCP79 (Q108, DX10, Intel)

Rumoured Data Points
AMD
- Certain reports claimed RS740 to be single-chip, ala MCP61.
- New reports are unclear, it clearly has SB600, but it might be integrated.
- Initial rumours mentioned RS790 as the first AMD DX10 IGP.
- New rumours claim RS780 is a 55nm RV610 with PCI-E Gen2 and HT3.
- These two chips are probably the same; was the info wrong, or was it renamed?
AMD Chipset RS780 [ChileHardware]

NVIDIA
- MCP73 is a ~135mm2 single-chip IGP for Intel platforms.
- MCP68 is presumably the AMD equivalent of that; die size is unknown.
- MCP79 is a single-chip DX10 IGP slated for Q1 2008, apparently.
- The codename of the AMD equivalent is currently unknown.
CeBIT 2007 - More MCP73 from MSI [Beyond3D]
CeBIT 2007 - ECS do new NVIDIA IGP too [Beyond3D]
CeBIT 2007 - Foxconn MCP73 mainboard [Beyond3D]
NVIDIA’s push on Intel IGP market – 55nm trial scheduled next year [HKEPC]

Intel
- The Bearlake family of chipsets is supposedly manufactured on 65nm.
- G35 might or might not be an evolution of G965. Supports DX10.
- G33 is a minor evolution of G965. Might or might not support DX10.
- G31 is a minor evolution of G965. Might or might not support DX10.
Intel's never ended change has annoyed manufacturers [HKEPC]

VIA
- DX10 IGP based on a derivative of an upcoming S3 DX10 GPU in H2 2007.
- Little else is known at this point...
S3 roadmap reveals DX10 and 10.1 chips [The Inquirer]

SiS
- Mirage 3+ is an overclocked Mirage3, to achieve Vista Premium compliance.
- Mirage 4 is currently being finalised, they're in post-sim work on FPGAs.
- Goal is to have it drain as little as 8-10W under load. Bullish on market acceptance.
CeBIT 2007 - SiS talk up D3D10 plans with Mirage 4 [Beyond3D]
SiS DX10 IGPs in Q307 [Beyond3D]

Extra Tidbits & Fun Facts
- MCP73 is likely the largest IGP chip, ever. Although it's also more integrated...
- G35 only supports DDR2, while the lower-end G33 also supports DDR3. Err?
- How much of a limitation is memory bandwidth, really?

 

[Kaotik]

Isn't Intel supposed to be coming out with DX10 variant of G965?

 

[Arun]

Isn't Intel supposed to be coming out with DX10 variant of G965?

Bah, yes! G35 is the DX10 part. Small typo there while copy-pasting around to make the original post, sorry. And yes, I'm actually hearing some very... interesting things about G35. If they are correct, I suspect it'll need his own dedicated thread since it might not be so much of a boring part after all. We'll see.

 

[Techno+]

I've read somewhere that the G35 will destroy the 'no playable game' reputation of IGPs

 

[Ailuros]

I've read somewhere that the G35 will destroy the 'no playable game' reputation of IGPs

*mumbles to himself* I shall not get excited until I see some substantial announcement

 

[Blazkowicz]

no stand alone S3 GPU on the list ?

 

[Techno+]

no stand alone S3 GPU on the list ?

this is an IGP thread, for more info on future S3 AIB GPUs read this

http://www.beyond3d.com/forum/showthread.php?t=36519

 

[DavidC]

http://www.tgdaily.com/2007/03/08/gdc2007_intel_g965_integratedgraphics/

"Nick Knupffer, PR Manager for Intel, showed off a beta hardware vertex driver that should speed up many first-person shooters. Time demos of Call of Duty 2 and Half-Life 2 Episode One were played and we saw that the G965 chip averaged around 30 to 60 frames per second."

Good to see we'll see proper drivers soon. I have a G965 motherboard but the combination of a slow CPU I have and the poor 3D performance makes me not want to replace the aging 945G(yet).

 

[Geo]

http://www.tgdaily.com/2007/03/08/gdc2007_intel_g965_integratedgraphics/

"Nick Knupffer, PR Manager for Intel, showed off a beta hardware vertex driver that should speed up many first-person shooters.

Ah, so *that's* what he does when he's not ignoring my emails!

 

[pelly]

Wouldn't it be interesting to see a "high-end" IGP platform where there were multiple PCI-E slots for flagship discrete GPU's and the IGP could be dedicated to aid in rendering, physics, or some other task? (Or, have a pseudo-SLI scenario for mainstream systems where the IGP would assist a single discrete GPU)

 

[Jawed]

Wouldn't it be interesting to see a "high-end" IGP platform where there were multiple PCI-E slots for flagship discrete GPU's and the IGP could be dedicated to aid in rendering, physics, or some other task? (Or, have a pseudo-SLI scenario for mainstream systems where the IGP would assist a single discrete GPU)

Methods and apparatus for processing graphics data using multiple processing circuits

Now if you'd posted that idea 2 years ago...

Jawed

 

[Geo]

Wouldn't it be interesting to see a "high-end" IGP platform where there were multiple PCI-E slots for flagship discrete GPU's and the IGP could be dedicated to aid in rendering, physics, or some other task? (Or, have a pseudo-SLI scenario for mainstream systems where the IGP would assist a single discrete GPU)

Well, I must admit I've wondered if that's where Intel is going. The facts and rumours seem to indicate both integrated and discrete for them.

 

[Techno+]

Wouldn't it be interesting to see a "high-end" IGP platform where there were multiple PCI-E slots for flagship discrete GPU's and the IGP could be dedicated to aid in rendering, physics, or some other task? (Or, have a pseudo-SLI scenario for mainstream systems where the IGP would assist a single discrete GPU)

Nice idea, for a while, i have been thinking of a better way of IGP implementation. We could have a mini-socket in our system where we can place the IGP, and then change it when we need to, comments?

 

[pelly]

Now if you'd posted that idea 2 years ago...

Funny you should mention that, as I remember sharing that idea a few years back to "the powers that be" over drinks one CES and getting some interesting looks (possibly a few grins).

 

[Techno+]

does anybody have an idea of the peak GFLOPS of Intel's G965?

 

[Blazkowicz]

Nice idea, for a while, i have been thinking of a better way of IGP implementation. We could have a mini-socket in our system where we can place the IGP, and then change it when we need to, comments?

we would call that socket a PCIe 16x slot, and have motherboards with multiple such slots even though most people will only use one.

 

[Techno+]

we would call that socket a PCIe 16x slot, and have motherboards with multiple such slots even though most people will only use one.

Sure we have that, but for the low-end market, it would be better to put the IGP in a mini-socket (thus the IGP will be very small and cheap) instead if the chipset. It would be much cheaper than a whole gfx card, even a low-end one.

As for using that IGP+discrete for gfx processing, has there been\planned any practical implementations of this?

 

[BRiT]

Sure we have that, but for the low-end market, it would be better to put the IGP in a mini-socket (thus the IGP will be very small and cheap) instead if the chipset. It would be much cheaper than a whole gfx card, even a low-end one.

No it wouldn't. The cost of the motherboards supporting such a socket and seperate memory modules would increase. They would also have high costs of research and development that will have to be recouped trying to perfect such a situation. That increase offsets the costs of going with a PCI Express 16 slot + Discrete Graphics Card from the low-end market.

 

[Techno+]

No it wouldn't. The cost of the motherboards supporting such a socket and seperate memory modules would increase. They would also have high costs of research and development that will have to be recouped trying to perfect such a situation. That increase offsets the costs of going with a PCI Express 16 slot + Discrete Graphics Card from the low-end market.

the socket is mini, i dont think that a mini-socket would cost much, and it can be UMA, however, i do agree with u that it would need extra research, however, it all depends on whether how many people will use it.

 

[Pete]

The point is that ppl would also have to buy more expensive motherboards and populate them with more RAM modules to accommodate the necessary bandwidth. Though I think using an IGP for IO and eliminating the discrete card's backplane would be interesting, though maybe not feasible b/c of latency.

I'm a little skeptical of how much current IGPs can help GPUs even with physics considering their relatively pitiful performance. I'm guessing SM2 won't help much, so that rules out RS690's 4 pipes, and I'm not sure how much NV's 2 SM3 pipes'll contribute, or Intel's whatever (considering it's slower even than NV's IGP).

As for newly improved vertex processing, that sounds great, but that all takes place on the CPU, no? So I'm not sure how that makes the IGP look any better.

That said, I was hoping Micron/Rendition's IGP socket would work out, so I'm not against the idea, just extremely skeptical it'll work with current business models.