Rated: 100% (1 votes)
Loading ...Where Nvidia goes next
Alex Watson
There’s an interesting thread on the forums about Nvidia’s ‘8900 series’, and I posted a quick reply this morning, which then got me thinking about the topic in a bit more depth. Not to get too Donald Rumsfeld about it, but there are a few known unknowns and plenty of unknown unknowns when it comes to Nvidia’s next GPU…
It seems likely that it will be released in the Autumn - Beyond3D has some notes from a recent investor briefing from which Michael Hara, Nvidia’s VP of Investor Relations, in which he revealed the company has “no intention from diverging from the cycle they have adopted with the G80, which is to have the high-end part ready at the end of the year and release the lower-end derivatives in the spring.”
Aside from its release date, the other guess which most commentators are making is that G90/G92 (whatever they call it) will be a 65nm chip. AMD has already moved to 65nm for its Radeon HD 2600XT GPU so its a very reasonable expectation, especially as Radeons and GeForces are made by the same Taiwanese foundry, TSMC. Process shrinks usually result in cooler running chips that are capable of higher clock speeds, so Nvidia’s next GPU being clocked higher than the 612MHz GeForce 8800 Ultra is a fairly safe bet.
Beyond this, it’s real ‘here be monsters’ land - there’s not even definite speculation. Fudzilla reckons the card will be PCI-E 2.0, but given his record at the Inquirer (as many misses as hits with GPU scoops), I’m not too inclined to believe this - unless of course, it’s an optional extra. After all, ‘thin on the ground’ would be a very kind way to describe PCI-E 2.0 support at the moment: only Intel’s yet-to-be-released X38 chipset officially supports it, although Bit Tech claims the P35 can, too. Sort of. Nvidia itself lacks a PCI-E 2.0 chipset, so if G90+PCI-E 2.0 were true, nForce 700 would have to be launched alongside any new graphics card. Although this isn’t out of the question; nForce 600 has been around for a while now.
The extremely knowledgable graphics geeks on the Beyond3D have some interesting suggestions in their G90 speculation thread, too: simple spec bumps such as 1GB of memory, DDR4 and faster GPU/shader clock speeds are all popular guesses. Tweaks to the actual architecture - a 512-bit memory interface, more ALUs - don’t get as many votes, and it’s easy to see why. AMD’s Radeon HD 2000-series has failed, at both the mid-range and the high-end, to beat the GeForce 8800s in terms of price and performance, so small tweaks to the process and the amount and type of memory will probably be enough for Nvidia to keep its performance lead.
However, Nvidia needs to be careful about doing ‘just enough’; it is a much more narrowly focussed company than its main competitor, AMD. Nvidia is most definitely not working on a CPU, so for the time being its graphics hardware and chipsets only. The latter isn’t even a separate part of the business, because there’s a strong argument that sales of those chipsets are in part (even largely) driven by the success of the GPUs. Fast graphics card create a performance feel for the Nvidia brand. Which is a good job for Nvidia, because with AMD absorbing ATi and Intel being able to perfectly good chipsets of its own, Nvidia motherboards aren’t the ‘natural’ choice for either AMD or Intel CPUs - so the green team relies on that positive brand value to attract customers.
All this means that fast GPUs are absolutely essential for Nvidia, and as being number 1 for graphics is so critical, an aggressively specified GeForce 8900 - i.e. something beyond a die shrink and clock boost - might not be out of the question. As Ars Technica’s Jon Stokes points out in his well thought-out write-up of the same investor call Beyond3D reported on, increasing the power of the GPU is also important for fuelling the growth of the GPGPU market. His piece is called ‘Nvidia on the high wire‘; it’s well worth picking through the whole thing, as he has some interesting points to make about where the green team might be headed in a more general sense beyond G90.
One thing is for certain: whether G90 is a spec bump, or something a little more radical, come the time, Nvidia will certainly be pitching it as radical. As the only mass market, graphics-centric silicon firm remaining, safeguarding and advocating for the importance of the GPU has now become almost as important to Nvidia as actually designing fast chips.
Nice blog and a good read.
Will be interesting to see how it all turns out.
I could of course wave across the desk at Alex to take his Porta-Pros off to say this but…
The other big advantage of a die shrink (90-65nm) is increased yields. With each chip phyiscally smaller you pack more onto a single wafer of silicon (typically 200mm in diameter, or possibly 300mm). that means each chip is cheaper to make. And will therefore be cheaper to buy.
So, if GeForce 8900 is ‘just’ a die shrink (it ain’t as easy as it sounds) then it’ll offer more peformance for less wattage (and therefore less noise, if the same coolers are used) and for a cheaper price. That’s a big kick in the goolies for ATi…
Well all I can say is roll on 45nm. Current top range cards are far too big and power hungry. Good news for case, psu and cooler vendors but a pain in the arse for the rest of us. Now they’re pushing 1000 watt PSU’s to run SLI! Someone is taking the piss methinks. I used to heat my bedsit with a 1KW fire.
I dont think that the G92 will be top end card. nVidia all readu have complete dominance over the high end and upper mid end of the spectrum with the 8800 series. Whatb they need now to really complete the crown as king of kings and secure the brand will be a decent midrange card, like a 8700, with about 3/4 performance of a 640mb GTS. This would reallt take over the midrange sector, especially at around 140 pounds. I would have brought that card for my rig instead of an 8800, prefectly formed for a 17″ screen and yet good enough to last for about 4 years. That would be, for me, the next logical step.
would agree with cybersolver, 1KW PSU’s are a pisstake, I’ve got a 430W and that’s about as far up as I want to go. It’s interesting to note that with all the talk of global warming and everyone cutting power use/carbon footpronts, how the various manufactureres of computer components will respond with things like smaller die processes.
TFT screens were the first jump, even if it wasn’t for envirnmental reasons. My old 17″ CRT ate 100W and my 19″ TFT eats only 35W so the benefit is obvious there. Would be good to see nVidia making better cards that aren’t as power-hungry.
I think the war on speed in processors (both CPU and GPU) is slowing down quite a bit and the next step (besides the ‘lets-stick-8-cores-on-a-single-die’ one) is to get the same amount of computing power for less consumption. I doubt the price will be that different though, with all the RnD going into it and it is satisfying enough to pay the same price and have a lower electricity bill!