The Fermi Saga Continues

Despite finally having a great DX11-ready graphics solution on the market, NVIDIA is reeling from competition in many ways.  It no longer has a chipset business, the Tegra platform is obviously seeing a much lower adoption rate than expected, the ION 2 discrete solution is lacking the key partners and selling points the first generation had and even the world of professional graphics is being assaulted by the new Evergreen-based AMD FirePro cards. 

The launch of GF100 as the GeForce GTX 480 and GTX 470 were better than many had predicted - our review of both of these cards was fairly positive with aggressive pricing and competitive performance but some undeniable issues with noise and heat.  Even though the prices were decent, the lack of any NVIDIA built DX11-capable cards for under the $350 that the GTX 470 sells for continued to be an issue and prevented them from really gaining ground in the marketplace.  In contrast, AMD has been offering DX11 GPUs from $699 down to under $100 since basically the beginning of 2010.  NVIDIA needed another option and even NVIDIA's own partners were fed up with the constant re-branding that we was being pushed on them.

Enter the GeForce GTX 465 - a Fermi-based GF100 card that will sell for about $279.  No, this won't address ALL of NVIDIA's issues but it definitely is a great option for those of us looking for competition in the mainstream to enthusiast markets. 

The GeForce GTX 465 GPU

For this launch, NVIDIA didn't provide us with any fancy slides or diagrams.  In fact, they left the majority of the details up to its card partners including getting us samples and pushing the marketing side of things.  It was all very...odd...to say the least from the viewpoint of someone that has been in this space for over a decade now.  And it puzzles me as to why NVIDIA would do this - because despite that fact that the GTX 465 is really nothing NEW, it is a pretty good graphics card.

Note: if you want the full run down on the new GF100 architecture and how it differs from both the previous GeForce cards as well as ATI's Radeon 5000-series of cards, you should check out the original review of the GeForce GTX 480 and GTX 470!

The Fermi we may never see

In short, the GeForce GTX 465 takes the above GPU that we were told about in November of 2009, cuts a bit more away from it than the GeForce GTX 480 or GTX 470 does and sells it for a lower price.  This version of Fermi above was never sold to consumers - it is the fabled 512 CUDA core version that will apparently be relegated to the HPC and Tesla lines.

GeForce GTX 480

What gamers got was this: the GF100 GPU with a single Stream Multiprocessor unit disabled providing a total of 480 shader cores.  This is the GeForce GTX 480 that you see today selling for $499.  (Just to create a metric that really doesn't mean much, that is $1.03 per CUDA core.)

GeForce GTX 470

The GeForce GTX 470 simply removed another SM unit as well as one of the 64-bit memory controller units; texture units and ROPs were eliminated as well in the appropriate ratio with the SMs.  This gave the GTX 470 a total of 448 CUDA cores and a price tag that sits today at $349.  (Again, for that metric, the GTX 470 sits at $0.77 per CUDA core.  A steal!)

GeForce GTX 465

Here is the new GeForce GTX 465 - it removes three more SMs as well as another 64-bits of memory bus width.  The resulting specifications include 352 CUDA cores (a drop of 27% compared to the GTX 470), 44 texture units (27%) and 32 ROPs (25%).  The memory bus drops to a width of 256-bits (25%) and the frame buffer is shrunk accordingly to 1GB (25%).  In terms of our new magical, not quite relevant cost measure, the GeForce GTX 465 comes in at $0.79 per core; right on par with the GTX 470. 

I should note that these diagrams were made by me, in Photoshop, in about 3 minutes and may not be completely accurate.  Some of the L2 cache may be disabled and other aspects could have changed but they are just here to provide a quick visual of how the new NVIDIA GTX 465 compares to the other GF100-based cards.  You can see that this is definitely more of a performance drop from the GTX 470 than the 470 itself is from the GTX 480 just based on numbers.  Real world performance testing (that you'll see on the come pages) will determine the viability of the math. 

In terms of clock speeds, the GeForce GTX 465 will run at almost identical frequencies as the GTX 470.

The only variance really is in the memory clock rate that runs at 802 MHz rather than 837 MHz - nearly everything else, including the recommended power supply size, thermal threshold, form factor, etc is the same.  Interesting, even though we have 25% less transistors at work the TDP of the GTX 465 is only about 7.5% lower than that of the GTX 470 and 25% lower than the GTX 480. 

It is important to note from a technological perspective that all of this is being done to the exact same GPU - there isn't a new revision at work here or a new spin of an actual chip with fewer shaders, etc.  The transistor count on the GTX 465 is still the 3 billion behemoth that the GTX 480 and GTX 470 are - just more of them are disabled due to either reliability, power or frequency issues.  The new GTX 465 part will allow NVIDIA to sell more of the dies that come out of TSMC that may not live up to the standards for a GTX 480 or GTX 470.  There is also a chance that the GTX 465 exists so that NVIDIA can sell more of the dies out of TSMC without looking like a shmo by dropping prices on the GTX 480/470 - it all depends on if NVIDIA is having a shortage or an over abundance of GF100 cores. 

There isn't really much else to say about the GTX 465 itself and this is likely why NVIDIA decided to go so "hands off" with it.  There is nothing noteworthy or new to discuss other than the price and NVIDIA PR likely didn't want to have answer questions from the media like "where is NVIDIA Surround?" and "why are companies like BFG leaving?" when on a conference call.  Seems reasonable enough to me - we just want to the goods.