www.abithelp.com

NVIDIA GT200 Architecture

Introduction

It’s been a long time coming – despite NVIDIA’s assurances that G92 was more than just a die shrink of the well accepted G80 architecture we knew better. G92 was, and continues to be, a great GPU for your dollar but we are hungry for something new, something that would really push gaming into the next generation. As it turns out both NVIDIA and AMD have something planned for the month of June but NVIDIA’s new GT200 GPU is the first up to bat. NVIDIA GeForce GTX 200 series of graphics cards based on this design are incredibly powerful and incredibly expensive. Be prepared to be impressed.

The GT200 Parallel Processing Architecture

One of the big changes to the NVIDIA’s presentation at the most recent technology summit was how the company promoted and positioned the GT200 design. Going back nearly a decade to when these events started there was no doubt you were there to see a graphics card and GPU at work but now with the well-documented public fight with Intel looming over everything the company does their angle shifted. Now being called a “parallel processor” more often than a GPU, NVIDIA is deathly serious about pushing the premise of their GT200 being used for more than just gaming: video encoding, high-performance computing, folding and more. We are covering all of these aspects of the GT200 as well in a separate article: Moving Away From Just a GPU.

At its heart though, the GT200 shares a lot in common with the theory and design of the G80 architecture. It is NVIDIA’s second generation of unified shader design, their second GPU to use HybridPower technologies and the second to offer 3-Way SLI to wealthy gamers. The GT200 does have new tricks though including a drastically increased number of shader cores, double precision floating point and is also much, much bigger than its predecessor.

The NVIDIA GT200 is 1.4 billion transistors strong and being built on TSMC’s 65nm process technology (at least for now) makes it the largest chip the manufacturer has ever built. The shaders can produces as much as 933 GigaFLOPS of horsepower at top reference speeds while maintaining impressively low idle power consumption. But packing in 1.4 billion transistors into a 65nm design makes this one BIG chip: XX mm^2 to be exact. When the chip was being held up by a GeForce product manager to display, NVIDIA CEO Jen-Hsun Huang jokingly said, “That chip’s as big as my head!” An exaggeration to be sure, but an interesting one with important business implications we’ll discuss in our conclusion.

Below is a complete (as complete as NVIDIA has revealed) block diagram of what makes the GT200 tick:


Click to Enlarge

All of those squares and colored sections are very specific components of the core that give the design such power. No doubt frequent readers of our articles that are in to the detailed, technical discussion here will recognize some key features: top setup block, lots and lots of stream processors, ROPs, memory controllers and more.

Let’s zoom in on one of the ten blocks of shader processors for a closer look:



Each of the ten divisions is made up of 24 separate shader processors bringing the GPU total to 240 SPs for those counting at home. The collection of 24 shaders is divided again into three sets of eight; these 8 SPs share a small block of localized memory for data sharing at 16K for each core. The larger L1 cache is shared between all 24 of the shader units and is used to increase memory performance and bandwidth in much the same was a primary L1 or L2 cache is used on a CPU.



Moving in even closer we can take a look at each of the individual SPs and find double register file (to help ever increasing shader program sizes) and three arithmetic units: one for FP, one for integer and one for moves and comparisons. This makes the GT200 a dual-issue design, much like the G80 was, that can handle both a MAD and MUL operation in a single clock, and puts the new design at 3 FLOPS per core per clock.

The shaders on the GT200 are unique though in that they also contain a floating point unit capable of handling double precision computing completely separate from the FP unit illustrated above. This feature probably won’t be used for gaming purposes at all since single precision calculations are more than adequate for any visual representations but it IS crucial for NVIDIA to have double precision support in the HPC market.
The filtering section of the design, shown as the red and black section at the bottom of the full block diagram, is updated as well for GT200. There are now 64 total samples for color and Z on each clock, divided up into eight segments of 8 each.

What does it all add up to? Here is a comparison of the theoretical bandwidth and performance numbers comparing GT200 to G80:



The GT200 has large advantages in ROP performance – the new GPU is able to handle 32 pixels per clock compared to the G80 that could only handle 12 pixels. That same performance increase is seen in the ROP blending numbers that put the GT200 at 19 giga-blends/s while the G80 pulled in only 7 GBL/s. The increase in cores (from 128 to 240) and frame buffer bandwidth give the new GT200 chip a huge leap forward in theoretical performance.

One area that did not increase much was the texture addressing performance: it jumped from 64 texels/clock to 80 texels/clock. Why did NVIDIA leave this area largely unchanged, at least in comparison to pixel shading performance? With each architecture design, GPU companies have to essentially “guess” which type of horsepower programmers will be using for years to come: shader processing or texture processing. Before the advent of unified shaders this task was even more difficult as designers had to decide between pixel and vertex shaders but it is still a risk. With G80, NVIDIA admitted that they overplayed their hand with regards to texture performance – they put more in than game developers needed, sacrificing some shader power to do so. AMD leaned more heavily in favor of shader power with their RV670 design which is why they consistently had a performance lead in synthetic shader processing benchmarks. The GT200 attempts to correct that by moving the ratio of floating power to texture power from 14:1 on the G80 to 19.4:1. We will soon see if that decision pans out as they hoped.

Not only does NVIDIA claim the GT200 has higher theoretical maximums on those key areas in processing and bandwidth, but they also say the GT200 pushes the limits on efficiency as well – basically how closely those maximums can be met. Our 3DMark06 and 3DMark Vantage GPU tests should be able to detail this more closely.

Another area of improvement on the GT200 core is in geometry shading. Thanks to a 6x increase in the output buffer compared to G80, NVIDIA can now claim to have geometry shading performance above any current AMD solution – the 8800 GTX was a far lower performer previously

We've told you that Nvidia said to its investors and the rest of the world that they are shipping 55nm in volume, but we were surprised to learn that Nvidia expects that more than 70 percent of GT200 chips shipped in Q4 to be 55nm.

Nvidia plans to do this quietly, as it doesn’t plan to change the specification of the existing GTX 260 216 Shaders and GTX 280 cards. Nvidia counts on the Jack in the box effect and you might be surprised that the card you buy in December will come with a 55nm GT200 chip and likely overlclock better.

The 55nm chip will be a bit cooler and smaller, and the first cards based on them are expected in early December.

Nvidia has stopped producing the 65nm-based Geforce GTX 260 with 192 shaders and all GTX 260 that are coming from now on are 55nm and have 216 shaders.

These cards will be available in early December and we are quite sure that they will overclock much better than the old ones. Nvidia doesn’t want to let ATI walk away without a fight and it's preparing a surprise or two for this month.

The old GTX 260 with 192 shaders, the one that is clearly slower than Radeon HD 4870, is going for retirement and it will be replaced by the new GTX 260 216 shaders and most of these cards will end up with 55nm chips.

The chips are shipping

Nvidia is planning to have its 55nm GT200 cards in retail / etail by early December. The company has started shipping these chips a few weeks ago and evaluation boards have been manufactured a few weeks ago, too.

Now the wait is purely logistic as it takes a long time to get the cards from Taiwan and China back to the western world. We learned that new cards should overclock better, but the first wave will be clocked at the same speeds as Geforce GTX 260, 216 Shaders and we suspect that GTX 280 cards will also get a new 55nm chip.

The new chip should have drastically lower power consumption and it should overclock higher and luckily we won’t have to wait much more to find that out.

According to the image we have on hand, the mark of GT200b on GeForce GTX260 graphic card will change from 65nm “G200-103-A2″ to be “G200-103-B2″. The “B” means latest 55nm processing technology, and its production cycle happened during the 33rd week of this year (August 11th to 17th), only 5 weeks after G200-103-A2 which was manufactured in the 28th week. Therefore, obviously NVIDIA had prepared for a long time.

Remark: The earliest version of GeForce GTX260 utilized GT200 with 192SP, and it was marked as “G200-100-A2″.

The first of new GTX260 graphic cards will hit the market in January next year. The number of its stream processors will maintain as 216, and its memory frequency is unchanged as well. The reference and non-reference versions cards will be on sale at the same time. Thanks to application of new processor and the circuit designed by manufacturers themselves, the price will hopefully keep declining. The GTX260 card is already available at $264 at the lowest right now.

More high-end products based on G200b, including GTX280 and dual-core 260GX2 will also emerge in Jan 2009.

Based on brand-new P654 reference design, the latest version of GeForce GTX260 Graphic Card utilizes 26.5CM PCB. All the 14 video memories are put on the front side of PCB, with 10 layers of PCB design. Previous P651 PCB shared by GeForce GTX260/280 was 14 layers. Previous version of GTX260’s features 3+2 phase power module (The 3 CSP sealed Volterra VT1195SF MULTIPHASE chip would be most eye-catching).

New P654 power modules has upgraded to 4+2 phase, equipping solid-state capacitor, magnetism shield inductance and MOSFET. Where the two 6Pin are connected features a large input inductance separately. Therefore, the reduced difficulties in design will bring down the cost.

Conclusion

XFX really knows their stuff, and these guys overclocked the card so high that even GTX 280 can feel the ground shaking beneath its feet. Running at evil 666MHz, the card can take anything you throw at it and it beats the old GTX 260 with 192 shader processors by 20%.

The card comes with a gift game FarCry 2, a gift that anyone buying this graphics card will surely like. XFX stuck to reference cooling, but don’t worry as the card runs quiet and cool. You’ll need two PCI Express power connectors to power this card, but that’s usual in this performance range. We’ve no other choice than to recommend this card to anyone, so that it can keep them warm trough cold winter nights.

EVGA has informed us that they will be the first with EVGA Geforce GTX 260 Core 216 / 55nm cards, at least in Europe. This doesn’t come as a surprise, as EVGA is not the only big one to support Nvidia only, as XFX has defected to ATI, and BFG is not that big as it used to be. There will be at least three models of the new 55nm GTX 260: the reference clocked, Super Clocked and Super Super Clocked cards.

Flextronics make the cards, and these cards should have some very good overclocking capabilities. Nvidia usually uses Flextronics for its first NVTTM (Nvidia Time To Market) production batch and EVGA decided to get the first ones, and we don’t believe that they had any choice.

We found the card listed in Austria, Grafikkarte PCI-E 2.0 EVGA e-GeForce GTX 260 Core216 55nm Superclocked, Retail, 896MB listed at ditech.at for €299 and with 626MHz clock speed.

SuperSuper clocked should work at, at least 666MHz, while we suspect that it might even reach 700MHz.

The cards are shipping as of early next week and there is a tiny chance you might have one to put under the Xmas tree. This will be the third version of GTX 260 product, as the first one has 192 shaders, the second one is again 65nm chip with 216 Shaders, and finally the one that is available as of week 52, the GTX 260 with 55nm chip and 216 shaders.