As NVIDIA’s World Turns…
It’s been a whirl-wind past few weeks for NVIDIA – bad mobility GPUs, the possibility of seeing NVIDIA inside Apple machines and now the death of the NVIDIA chipset products. Quite a bit for a single company to take all at once, but how much truth is there to the idea of the end of the nForce? We take yet ANOTHER look at NVIDIA…
If anything, NVIDIA is getting a lot of attention these past few weeks. Just the other day I published a story that theorized NVIDIA chipsets would be adopted by Apple in their upcoming MacBook transition and maybe even in their iMac line late this year. Now today, the tables have flipped and rumors of the nForce 790i being cancelled as well a story that NVIDIA will be exiting the chipset business completely have popped up. Quite a swing for three days of news.Obviously when I saw this news I immediately started calling people at the various companies involved to see exactly what was going on.
The first story we saw on this issue came from Charlie over at the Inquirer that noticed that Gigabyte, DFI and Foxxcon have pulled their nForce 790i motherboards from their websites and from retailers. Rightfully so, the story comments that the pulling of already designed motherboard products is a very bad sign for NVIDIA; if an engineering team has already spent the time designing a motherboard, testing it and selling it, it takes a significant event for them to just throw that money down the drain.
Of the three companies listed in the Inq’s story, the most valuable one to NVIDIA is Gigabyte – they are considered one of the “big three” motherboard manufacturers and losing them will definitely hurt the NVIDIA ego. That being said, both ASUS and MSI, as well built-by-NVIDIA partners like EVGA and XFX are still selling products based on the high-end 790i chipsets. And even though we love our smaller motherboard vendors, the truth is that if you have the “big three” on your side, it really won’t matter (much) about the niche motherboards from a financial stand point. This does leave me curious how much the lack of a Gigabyte product based on the 790i will hurt them – it will not be painless, that much is certain.
The good old days?
Today’s story from Digitimes that claims NVIDIA will be exiting the chipset business entirely is an order of magnitude more severe than what the Inq posted. Their sources claim that NVIDIA approached motherboard vendors about interest for future products and was met with none – leading them to conclude that NVIDIA would roll the chipset business into the GPU business and be done with it. While an interesting theory, and not without some merit, there is other news that the story blatantly ignores that goes against their conclusion.
NVIDIA’s Immediate Chipset Strategy
For example, there is at least one desktop chipset that has been developed and finalized that we haven’t seen released yet: the MCP7A product for the Intel platform. This chipset offers up a lot of features, and though it appears to be a little late based on initial availability estimates, I would expect to see it in the fall 2008 time frame as well. To quote Dailytech on information about this product:
The MCP7A line will support a 1333 MHz front-side bus and 20 PCIe 2.0 lanes in a
possible 16×1 + 4×1 setup. This would allow a single GPU expansion card to be
used in conjunction with the GPUs in seven of these chipsets for a Hybrid SLI
setup. The only two SLI boards, the MCP7A-SLI and MCP7A-GL, will feature 2 x8
PCIe 2.0 slots for dual expansion card capabilities.
Each of the chipsets with onboard GPUs will feature 2 analog and 2 digital display heads, with the notable exception of the high-end MCP7A-J which will feature 2 analog and 3 digital heads. Motherboards based on the MCP7A line will have the potential to support DVI/HDMI/Dual Port and RGB formats of video output, while the MCP7A-J chipset will also support LVDS.
Each of the chipsets with onboard GPUs will feature 2 analog and 2 digital display heads, with the notable exception of the high-end MCP7A-J which will feature 2 analog and 3 digital heads. Motherboards based on the MCP7A line will have the potential to support DVI/HDMI/Dual Port and RGB formats of video output, while the MCP7A-J chipset will also support LVDS.
This product and the mobility MCP79 chipset that I discussed in my NVIDIA and Apple story are based off of the same architecture and have very similar features. The fact is that they are both done and are both still to be released.
Moving from Intel to AMD, NVIDIA’s AMD chipsets are still relatively successful as well – according to recent Mercury reports, NVIDIA’s market share has been dominant but stable: Q307: 61%, Q407: 60%, Q108: 59%, Q208: 60%. Would they rather being seeing growth on these numbers? Of course, but AMD’s latest chipsets for their own CPUs (the 780G, etc) have been very good products and their adoption has slowed NVIDIA’s movement upwards. They are still the chipset leader in the AMD market though and the company would simply not give it up at the drop of hat.
According to these numbers from Mercury Research, the gain in AMD chipset market share seems to have come at the expense of VIA’s dying chipset division, not NVIDIA’s.
DMI is the Future, long live DMI!
I think the majority of these concerns about NVIDIA’s chipset business started with the confirmation that NVIDIA was either not granted a CSI license by Intel (also known as Quick Path Interconnect, the bus for upcoming Nehalem/Bloomfield processors) or decided they weren’t going to try to acquire one to make chipsets for the first round of Nehalem processors. Instead, NVIDIA has offered up the nForce 200 bridge chip to potential X58 chipset motherboard vendors to enable and allow SLI technology on Intel’s upcoming platform shift. Many in our space saw this is a white flag from NVIDIA, that they weren’t able to cut into Intel’s chipset business as they’d hoped. I myself expressed concern about NVIDIA’s stepping away from chipset design during this period when I first heard it. I usually believe that the business is like an exercise regiment; if you skip just one day you’ll likely start skipping more and eventually quit altogether.
Nehalem test system from Computex – next generation Nehalem will use DMI chipset interconnects
But NVIDIA has already publicly stated they weren’t going to let that happen – in a conference call earlier this year NVIDIA told reporters and analysts that they were indeed skipping the Bloomfield market by not making a chipset for it but that they had every intention of developing a chipset around Intel’s next bus technology, DMI. For those of you not privy to this information before now, DMI is essentially the replacement for Intel CSI that will be released sometime mid-2009. Essentially a x4 PCI Express bus, DMI should be around for processors AFTER Bloomfield: Lynnfield, Clarksfield, etc.
NVIDIA’s excuse then for skipping Bloomfield chipset support is that the life cycle for CSI was too short for them to make the engineering investment worthwhile. A questionable stance from a corporation that makes graphics cards obsolete after just 6 months in the market for sure, but not necessarily a bad choice for a company struggling to find its new direction. Is it disappointing? Yes. Is it going to cause some pain and suffering for NVIDIA’s business? Likely yes. Does it mean they are exiting chipsets all together? I don’t think so.
Integrated Graphics Chipsets Only
Everything above I have stated is sourced from inside NVIDIA or its partners – I am reasonably sure about its accuracy. I have another theory though that is based just on observations that also might explain the hysteria around NVIDIA’s chipset business: an end to non-integrated graphics chipsets.
When we first heard about the NVIDIA 780a chipset for AMD Phenom processors, and that all versions from the budget to high end SKUs would have integrated graphics on them, my immediate thought was that this choice would definitely be extended to Intel’s chipsets in the future. The nForce 790i was developed well before this decision was likely concrete inside the company and thus the 790i could be the last discrete chipset, ever. The benefits to this design policy include saving engineering dollars by only having to design one chip and easier integration of the chipset to various designs by NVIDIA partners. You can expect all NVIDIA chipsets going forward to follow this design logic.
Expect to see more chipsets like this in the future
And there will be more chipsets – it seems inevitable that this story from Digitimes is going to cause a lot noise without a lot of proof. To me there just seems to be too much going on inside the NVIDIA MCP division to even think the company is going to pulling out of the business. Does it seem like the drive and push of NVIDIA’s chipset marketing team, and the MCP products themselves, have lost of a lot of their glamour? Yes, but we honestly expected this once Intel formally announced its move to an on-die memory controller thus negating the last vestige of traditional chipset differentiators. The chipsets at NVIDIA might be less exciting than they were in the original nForce/nForce 2 days, but I just don’t think we’ve seen the end of the nForce name.
UPDATE: NVIDIA sent along their official response to the Digitimes news:
- The story on Digitimes is completely groundless. We have no intention of getting out of the chipset business.
- In fact, our MCP business is as strong as it ever has been for both AMD and Intel platforms:
- Mercury Research has reported that the NVIDIA market share of AMD platforms in Q2 08 was 60%. We have been steady in this range for over two years.
- SLI is still the preferred multi-GPU platform thanks to its stellar scaling, game compatibility and driver stability.
- nForce 790i SLI is the recommended choice by editors worldwide due to its compelling combination of memory performance, overclocking, and support for SLI. In fact, a recent article on Tom’s Hardware recently came to the same conclusion: http://www.tomshardware.com/reviews/NVIDIA-790i-SLI,1977-29.html
- We’re looking forward to bring new and very exciting MCP products to the market for both AMD and Intel platforms.