This week in our GPU Flashback Archive series we cast our minds back to a very popular and well loved graphics card series, the GeForce 400 series. NVIDIA launched the GeForce 400 series in March 2010 armed with a new Fermi architecture that it hoped would help it compete with the successful AMD Radeon 5000 series. Let’s look at the new features that Fermi offered, the cards that were popular and the scores that were submitted to HWBOT in this era.
NVIDIA GeForce 400: Overview
Compared to previous product launches from NVIDIA, the GeForce 400 series launch did not go as smoothly as hoped. September 2009 saw AMD come out with their Radeon 5000 series which made a solid case against NVIDIA 200 series offerings. It would be January before NVIDIA really started wooing tech media with tales of its forthcoming Fermi architecture lineup. It would be March 2010 before tech media actually got their hands on the new cards and several weeks after that before enthusiasts would be able to actually buy one. This was not the typical NVIDIA launch. Reasons for the delay certainly seemed to lie with issues with actual fabrication at TSMC who were not providing the yields expected on their new 40nm process. This was a problem that particularly hurt NVIDIA due to the fact that the new Fermi GPU, the GF100, was actually very large. When the GeForce 400 series finally arrived in the form of the GeForce GTX 480 and GTX 470, by most calculations they were six months late.
The NVIDIA GF100 was in fact the first major architectural overhaul since the first Tesla-based GPU, the G80. The NVIDIA marketing team touted the new chip as the next evolutionary step in GPU development. One thing is certainly true, the GF100 was the biggest GPU the company had ever produced. Its 3.4 Billion transistors made the previous gen GT200 with 1.4 billion transistors look rather small and weak by comparison. In terms of physical size it was also pretty monstrous, arriving in a 529 mm² die package that was considerably larger than its predecessor at 470 mm².
So let’s get down to some of the physical features present on the GF100, the features that actually make it such a monster. The Fermi GF100 was design to have a whopping 512 Stream Processors, more than double the previous generation. However, either due to manufacturing issues or poor yields, NVIDIA never actually brought to market a fully equipped GF100. To the surprise of many, the GF10 arrived mounted on the GeForce GTX 480 with one SM disabled just 480 Stream Processors. On the GTX 470, two SMs were disabled meaning only 448 Stream Processors were present. In terms of ROPs the GTX boasted 48 compared to 28 on the GTX 285, further boosting the pure horsepower available.
Aside from a considerable bump in Stream Processors, the new GeForce 400 series also offered improved memory bandwidth. Fermi GPUs were coupled with a GDD5 compatible memory controller which was configured with a memory bus of 385-bit on the GTX 480 and 320-bit on the GTX 470. The jump from GDDR3 to GDDR5 was considerable with capacities also rising to 1,536MB (6x 256MB) on the GTX 480 and 1,280MB (5x 256MB) on the GTX 470 card.
Here’s a shot of a GeForce GTX 480 PCB. You can see the GDDR5 ICs and the considerable presence of the GF100 GPU itself.
In terms of gaming and API support, the Fermi-based GF100 was the first NVIDIA chip to be DX11 compatible. This meant it had support for Shader Model 5.0, Multi-threaded rendering, DirectCompute 11 with Physics and AI support plus hardware-based Tessellation. All of these features would debut on DX11 capable machines running the new Windows 7 OS.
In terms of pricing, the new flagship GeForce GTX 480 arrived asking for considerable $499 USD, pretty much equal with the dual GPU GeForce 295 card. The GeForce 470 arrived with a more modest $349 USD tag. We already mentioned the fewer Stream Processors on the GTX 470, it also had lower clock speeds with a GPU clocked at 607MHz compared to the flagship at 700MHz, a Shader clock of 1,215Mhz compared with 1,401MHz on the GTX 480 and a lower memory clock, 837MHz compared with 924MHz.
In many ways the arrival of the GeForce 400 series saw a change in the way overclockers would approach GPU purchases. With the previous 200 series it was entirely possible to push a cheaper GTX 275 card to similar performance levels to a GTX 285 card. Remove default clocks and pretty much one GT200 chip was similar to another. However with the GF100, we saw the GTX 470 card hobbled just enough to prevent such an outcome.
The Most Popular NVIDIA GeForce 400 Card: The GeForce GTX 480
It’s time to take a look at the most popular NVIDIA 400 series cards in terms of submissions to the HWBOT database:
- -GeForce GTX 480 – 29.46%
- -GeForce GTX 460 (256-bit) – 23.65%
- -GeForce GTX 470 – 16.47%
- -GeForce GTS 450 (GDDR5) – 8.46%
- -GeForce GTX 460 (192-bit) – 5.35%
- -GeForce GTX 465 – 4.33%
- -GeForce GT 430 – 2.03%
- -GeForce GTX 460 SE – 1.38%
- -GeForce GTX 460M – 0.89%
- -GeForce 310M (DDR3) – 0.83%
Having mentioned the clear difference between a GTX 470 and a GTX 480 in terms of potential performance, it’s perhaps not so surprising to see that the majority of GeForce 400 entries involve a high-end, flagship GTX 480 card. If you wanted to have a chance of actually competing at the highest levels, a GTX 480 was a must. For a significant period, it ruled supreme as a single GPU solution, being used in almost a third of all GeForce 400 series submissions on HWBOT.
The iconic NVIDIA GeForce GTX 480 reference design graphics card:
Let’s just hone in on the GTX 480 card for a moment. Whereas the GTX 470 used a very similar full shroud cooling design to previous generation, the GTX 480 cooling array was considered quite unique. The card is one of the first to feature exposed heatpipes protruding from the GPU cooler almost like a race engine grill. The effect is magnified by the a heatsink grill on the front of the card, again adding a race car like edge to things. In reality however, the GTX 480 and its GF100 GPU were rated with a TDP of 250W. It was one of the first to require both 6-pin and 8-pin connectors from the PSU. Yes this baby was powerful, power-hungry and hot as hell. The cooler design wasn’t just cosmetic, it was entirely necessary.
Here’s a shot of the cooler that lies within that long shroud.
Both the GTX 470 and GTX 480 arrived with dual DVI outputs plus a mini-HDMI port. Although some AMD cards had started using DisplayPort outputs, NVIDIA preferred HDMI due its apparently broader adoption with display manufacturers. At 10.5 inches, the GTX 480 was a substantial card. It also shone in benchmarks, offering best in class frame-rates. In most games and benchmarks the GTX 480 completely wiped the floor with its main rival, the Radeon HD 5870. Likewise the GTX 470 dealt a massive blow the relevance of the upper mainstream Radeon HD 5850.
In terms of overclocking we see a new era emerge around this time with specific GPU BIOSes allowing for more options from the card. This was particularly true because the GPU was increasingly served by digital PWM controllers from manufacturers like ChiL, Intersil and International Rectifier. Unlocking the full potential of the now digitally delivered power would only be possible with the right BIOS. Few members of the overclocking scene had access to these special card BIOSes although many eventually were shared among the community on popular forums such as XtremeSystems. The arrival of digital controllers meant a major shift in VGA card overclocking, a shift that many overclockers rue to this day.
NVIDIA GeForce 400 Series: Record Scores
We can now take a look at some of the highest scores posted on HWBOT using NVIDIA GeForce 480 card, the fastest single GPU card in the 400 Series lineup.
Highest GPU Frequency
Although technically speaking, GPU frequency (as with CPU frequency) is not a true benchmark, it does remain an important metric for many overclockers. Looking through the database, we find that the submission with the highest GPU core frequency using a GeForce GTX 480 card comes from Swedish Elmor. He pushed a GeForce GTX 480 to 1,450MHz, a massive +107.14% beyond stock settings. The rig used also included an Intel Core i7 Extreme 980X ‘Gulftown’ processor clocked at 5,923MHz (+77.87%).
You can find the 3DMark06 submission from kayl here on HWBOT: http://hwbot.org/submission/1061564_elmor_3dmark_vantage___performance_geforce_gtx_480_39281_marks
3DMark Vantage – Performance
The highest 3DMark Vantage – Performance score submitted to HWBOT using a single NVIDIA GeForce 480 card was made by ikki from Japan. He pushed an MSI GeForce GTX 480 Lightning card to 1,300MHz (+85.71%) on the GPU core and 1,148MHz (+24.24%)
on the graphics memory. With this configuration he managed a hardware first place score of 40, 684 marks. The submission was actually fairly recent and was helped by an Intel Core i7 5960X ‘Haswell-E’ chip clocked at 5,376MHz (+79.20%).
Here’s a close up of the LN2 cooled rig as pushed by ikki..
You can find the submission from ikki here on HWBOT: http://hwbot.org/submission/2924638_ikki_3dmark_vantage___performance_geforce_gtx_480_40684_marks
In the classic Aquamark benchmark we find that ikki (Japan) is again the highest scorer with a single GeForce GTX 480 card. This time he pushed the GPU clock to 1,154MHz (+64.86%) to hit a score of 608,081 marks. The score was made just in February of this year and will have benefited greatly from being joined by an Intel Core i7 7700K ‘Kaby Lake’ chip clocked at 6,800MHz (+61.90%).
You can find the submission from ikki here on HWBOT: http://hwbot.org/submission/3462382_ikki_aquamark_geforce_gtx_480_608081_marks
Thanks for joining us for this week’s episode of the GPU Flashback Archive series. Come back next week and join us for a look at the NVIDIA GeForce 500 series of graphics processors and cards.