cM :: MORLEY|DIGITAL

When chip developers, be it Intel, AMD, or NVIDIA, design new technology, they tend to stratify the product in a way that will appeal to various markets.

They do this several ways.  They can speed-bin the parts and cull the fastest performers and market them under a premium label, or they can go the other way and take lesser performing parts and market them to the mainstream.  Everyone, in fact, does both.  Either way they are able to get the most out of their research and development dollars pumped into the product while delivering value to the consumer.

Focusing on the graphics card market for a moment, take a look at AMD/ATI and NVIDIA’s products.  They feature various amounts of stream processors operating at varying frequencies.  These processors are paired up with different types of memory configured in varying capacities and bandwidth.

Now, there seems to be accusations (mostly by fanboys in forums - but also by snide remarks from tech press) levied against NVIDIA right now that they are somehow being dishonest in what is being called a “rebranding” of their G92 chip.  This chip formed the basis of the 8800GT and the 9800 series solutions.  It’s a powerful chip that delivers great performance.  It was derived from the G80 chip, the original 8800 GTX, which was the reigning speed king for 18 months.  Since then NVIDIA has been able to shrink the chip, increase clock speeds, reconfigure the types of memory it works with, and even remove stream processors to create products that are priced to meet market needs and compete with AMD/ATI.  The whole time making more peformance available to more people at better price points.

Now that NVIDIA has had a whole new architecture available to them for several months now, the GT200, NVIDIA needs to be able to do the same thing - stratify.  The 65nm GT200 is used in the GTX 260 and 280.  The 55nm GT200b are utilized in the GTX 285 and 295.  These are high end components that serve a small market niche and are large, power hungry, and expensive.  They’re also silly fast.  

One thing to keep in mind at this point is that the GT200 does not introduce any new feature sets.  Compared to the GeForce 9-series, it “merely” features more stream processors and memory bandwidth.  It’s not a DirectX 11 part.  It’s just pure muscle based on the same architecture dating back to the G80.  And that’s not a bad thing in the least.  AMD does the same thing.  The very successful Radeon HD 4800 series are built upon the same basic philosophy as the not-so-successful Radeon HD 2900.  They just tweaked it, shrunk it, dded stream processors, and repositioned it.

So NVIDIA has to make the choice on how to “fill down” their product stack.  Given their GTX 200 naming convention, the obvious choices would be something like the GTS 250 or the GTS 240, which is what the tech press is reporting.  Given that the current GTX 260 has 216 stream processors and the GTX 285 has 240, how would you design the value parts that NVIDIA needs to compete in the mainstream if you were in their product group?

Well, let’s start by subtracting stream processors and some memory bandwidth.  128 stream processors is pretty hefty, wouldn’t you think?  I mean, that’s as many as the recently released HP Firebird has.  And that machine is getting a lot of attention!  And let’s chop that memory bandwidth up - boy those 512-bit traces make for an expensive PCB.  Let’s go 256-bit.  OK…next step.  Do we take our large, expensive GT 200 chip and disable stream processors or…wait a moment, don’t we have a chip that already meets this feature set?

Yeah, it’s the G92.  Which, btw, will get a nice performance boost as well.

It’s good for the consumers.  They now have a product lineup that will make sense from top to bottom, both pricing and performance.  There is nothing nefarious about what NVIDIA is doing.  They are pricing and positioning these new products against their GT200-based bretheren.  And the G92 nips at the heals of the GTX 260, so it’s a smart chip to start with.  Consumers win.

It’s good for the channel.  This economy sucks.  But we still need to move new product.  These parts are new.  They make it easier for us to talk to the customer about value vs performance.  Boutiques, system builders, retailers, e-tailers, and distys win.

It’s good for NVIDIA and their shareholders.  NVIDIA is able to continue to make money on a product that they originally designed over 2 years ago.  Their architecture has legs and as long as they can continue to deliver value to the customer and bring in profits, it will only help them be a stronger company.  And that’s a good thing for everyone because:

It’s good for competition.  Do you think that AMD stands idly by and lets NVIDIA make these movements in a vacuum?  No, they change their pricing and products accordingly.  Who knows, maybe it’ll spark a new mid-range GPU throwdown.  

And who doesn’t like a good cage fight?