Date:
Monday , June 16, 2008
Author:
Brent Justice
Editor:
Kyle Bennett
Google +1

BFG Tech GeForce GTX 280 OC & GTX 260

NVIDIA’s next-gen GPU has arrived. We game with BFG Tech’s new GeForce GTX 280 OC in Crysis, Age of Conan, COD 4, and Assassin’s Creed. We’ve included the GeForce GTX 260 as well. Real gameplay advantages are to be had, especially in the brand new Age of Conan.

Introduction

In November of 2006, NVIDIA unveiled its then brand new DirectX 10 unified architecture GPU, the G80 series, specifically known as the GeForce 8800 GTX. The GeForce 8800 GTX has been so kind to the gamers through the years, but it is time to remove its crown and make way for its successor!

The GeForce 8800 GTX introduced a new unified architecture that allowed “stream processors” to carry out pixel, vertex and geometry acceleration functions. The kicker was that there were 128 of these little buggers which meant a LOT of pixel, or vertex shading power was available to the video card for the first time. Not only that, but none of the shading power was wasted since in a unified architecture every shader processor is utilized for something, none sit idle like they use to back in the days of a fixed pipeline GPU. To read more about the unified architecture click here.

The GeForce 8800 GTX also excelled in other ways by being supporting a very wide 384-bit memory bus and have a standard complement of 768MB of it available. This extraordinary shader power combined with memory bandwidth and capacity has allowed the GeForce 8800 GTX to remain at the top of the performance scale in games, especially at high resolutions with AA enabled.

The GeForce 8800 GTX was succeeded this past April by the GeForce 9800 GTX. The GeForce 9800 GTX also has 128 stream processors, and NVIDIA was able to greatly increase the stream processor frequency, making it, until now, the fastest single GPU video card for gaming. We will be comparing the GTX 280 and 260 to the 9800 GTX for this reason. The GeForce 9800 GTX’s stream processors are clocked at 1.688GHz versus 1.350GHz on the 8800 GTX. One area where NVIDIA backslid on though was the memory side of the video card. The GeForce 9800 GTX narrowed the bus to 256-bit and cut the capacity to 512MB. This has hurt the overall performance of the GeForce 9800 GTX when cranking up the resolution with AA, and especially with SLI and Tri-SLI. While that is the case, for the most part the GeForce 9800 GTX did outperform the GeForce 8800 GTX in games, especially the more shader intensive ones.

It has been a year and a half since NVIDIA launched the GeForce 8800 GTX (G80 series), and they are now ready to launch their next generation of GPUs. So what can be done to make a unified architecture even faster? Well add on more stream processors of course! NVIDIA has also improved the memory specifications as you are about to see.

GeForce GTX 200 Series

Firstly, NVIDIA’s new GeForce GTX 200 series does not support DirectX 10.1 (Shader Model 4.1.) AMD still has the leg up on NVIDIA by supporting DX10.1, and you can bet money on AMD continuing to support DX10.1 in their next generation of GPUs. We’ve already seen one game where DX10.1 has provided positive results. NVIDIA’s new GeForce GTX 200 series supports DX10 (Shader Model 4.0) same as the last generation, and is their 2nd generation unified architecture.

Article Image Article Image Article Image

As you can see, NVIDIA has changed the nomenclature for this generation. The “GeForce GTX 200 Series” consists of two video cards NVIDIA is announcing today, the GeForce GTX 280 and GeForce GTX 260. This is reminiscent of the past generation with the simultaneous announcement of the GeForce 8800 GTX and GeForce 8800 GTS.

The new GTX 200 series are high-end enthusiast video cards, and priced accordingly. The GeForce GTX 280 will cost a hefty US$649. Ouch. The GeForce GTX 260 is at a more palatable $399, but in the day and age of über-cheap 8800GT video cards, these new beasts will have to deliver performance where it is important to your specific gaming profile for them to bring the value.

GeForce GTX 200 Series Architecture

Being that this is NVIDIA’s 2nd generation unified architecture, there are core architecture improvements. We are going to keep the focus brief on these improvements and touch on the important aspects for gaming. There are other websites that dissect every little technical feature, but quite frankly, to us gamers, we just want to know how much better it will make our gaming experience!

The design goal with the GeForce GTX 280 is quite simple according to NVIDIA: “Design a processor with up to twice the performance of GeForce 8800 GTX.” Of course the GeForce 8800 GTX is a very fast GPU in its own right, so twice the performance of it in a single GPU form factor for this generation is an impressive goal. To do this NVIDIA has improved upon several computational components in the GPU, basically giving us more of everything.

Article Image

The GeForce GTX 200 series GPUs are based on TSMC’s 65nm fabrication process and include 1.4 billion transistors. As you can see in the chart above, the most obvious performance improvement for shader intensive games will be the increase to 240 stream processor cores from the last generations 128. This in turn means there are more threads in flight throughout the GPU compared to the last generation, from 12,288 to 30,720. GeForce GTX 200 series GPUs also employ a more efficient scheduler for texture processing which allows it to achieve much closer theoretical peak performance in texture filtering from last generation. This is good news; NVIDIA is still putting focus on improving texture filtering.

Article Image Article Image

Above is a breakdown of some important specifications of the GeForce GTX 280 compared to the last generation. You will notice important improvements in the ROP and texture units. You can also see the die block diagram in the second picture.

Various Other Architecture Improvements

NVIDIA has doubled the local register file size, which means longer shader programs can run faster on the GTX 200 series. Dual Issue performance has also been improved, allowing full-speed dual-issue of multiply-add operations. GeForce GTX 200 series GPUs also now support double-precision 64-bit floating point computation. The ROP system has also been improved. The previous generation outputted 24 pixels per clock and blended 12 pixels per clock, the new GeForce GTX 280 can output AND blend 32 pixels per clock. There have also been improvements to the buffer structure for geometry shading and stream out functions, improving performance.

GeForce GTX 280

Article Image Article Image

Let’s get right to the specifics. The GeForce GTX 280 is a 65nm GPU with 240 stream processors and 32 ROPs. NVIDIA still maintains separate clock domains between ROPs, stream processors and memory. The “core” (ROPs and everything else) are clocked at 602MHz. The stream processors are clocked at 1.296GHz.

Here is what you guys are going to like a whole lot; NVIDIA has answered the cry of gamers on the memory specifications. The GeForce GTX 280 has a real 512-bit memory bus! YES! Not only that, but NVIDIA is putting 1GB of GDDR3 on this bus as the standard! YES! With a wide 512-bit memory bus and 1GB of RAM this video card will be ready for high resolutions, AA, SLI and Tri-SLI without the memory being a bottleneck. The memory frequency is set at 2.214 GHz which gives us a bandwidth of 141GB/sec, which is the highest to date!

With a 512-bit memory bus, every MHz of overclocking you achieve with the memory frequency will equate to an exponentially higher memory bandwidth compared to a 256-bit memory bus as well. Meaning a 100 MHz overclock provides a bigger jump in bandwidth with a 512-bit memory bus than it does with a 256-bit memory bus.

The maximum rated board power for the GTX 280 is 236 watts according to NVIDIA, with a maximum GPU thermal threshold of 105c. We will be doing our own power testing later in this evaluation. The GTX 280 requires one 6-pin and one 8-pin power connector.

GeForce GTX 260

Article Image

Now we come to the more affordable, and probably will be the most popular video card out of the two, the GeForce GTX 260. This one is priced at a more reasonable $399, which means more people are going to be able to get their hands on this one than the GeForce GTX 280. As you would expect, the GTX 260 is a cut down version of the 280.

The GeForce GTX 260 is a 65nm GPU with 192 stream processors, and 28 ROPs. Therefore it has 48 less stream processors than the GTX 280, but still 64 more stream processors than the GeForce 9800 GTX. The core clock speed is clocked at 576MHz, and the stream processors are clocked at 1.242GHz.

Thankfully, the GeForce GTX 260 does not disappoint on the memory side of things. NVIDIA could have cut it way down to a 256-bit memory bus and 512MB of RAM, which I’m sure most people were expecting, but, THEY DID NOT. NVIDIA is maintaining a high level of bandwidth and capacity on the GeForce GTX 260! The GeForce GTX 260 has a real 448-bit memory bus and 896MB of RAM, standard! The memory is clocked at 1.998GHz, which gives us 111GB/sec, still much higher than even the GeForce 8800 GTX was! This means the GeForce GTX 260 should also do well at high resolutions with AA, in SLI and Tri-SLI. Maximum rated board power is 182 watts with the same GPU thermal threshold. It requires two 6-pin auxiliary power connectors.