As the official launch of the RTX 4090 draws near, photos of add-in board (AIB) RTX 4090 GPUs have emerged. They all share one thing in common: they are absolutely gigantic. The 4090 Founders Edition (FE) is already a substantial card, but it’s dwarfed by Nvidia’s partners’ GPUs. As it turns out, there’s a good explanation for that. According to Igor’s Lab, the AIBs were just following orders from Nvidia.
The German overclocker says his information is based on sources in the AIB supply chain in Taiwan. Designing GPUs, and their coolers, is a very long process. Therefore, Nvidia originally instructed its partners to prepare 600W coolers for its upcoming GPUs. Apparently, when this order was given, Nvidia had planned on once again using Samsung’s silicon for Ada Lovelace, Igor reports. The instructions were included in the Thermal Design Guide Nvidia sent to its partners.
The report states Nvidia figured it would need 600W of juice to achieve the clock speeds needed to boost performance using Samsung’s process. Then Nvidia switched to TSMC’s 4nm node instead. The result was not only better yields but much higher performance than Nvidia expected. This allowed it to lower the total board power (TBP) from 600W all the way down to 450W. Notably, that’s the same TBP as the RTX 3090 Ti. The RTX 4080 16GB also had its power consumption reduced dramatically. It went from 420W initially down to 340W, then finally to 320W. This could be the source of leaks in April saying the RTX 4090 was indeed a 600W GPU.
The end result is over-engineered coolers, even for a super-dense die like AD102. Remember, AD102 has 76 billion transistors, which is triple the number for GA102. Though Nvidia’s design is three slots, most AIB designs are even bigger. This should allow for seriously chilly temps and high clock speeds. According to a report from Wccftech, the RTX 4090 runs between 50-55C with clock speeds over 2.8GHz. It can also allegedly hit 3GHz when overclocked. The roadblock will be voltage, however. There will be ample cooling to keep cranking the clocks, but people won’t have enough volts to take things too far. Nvidia typically doesn’t allow voltage modifications on its GPUs, which appears to be the case here.
Overall, this probably annoyed Nvidia’s partners. They ended up doing a ton of engineering that was unnecessary. It also reportedly cost partners an extra $10 per board. This adds up when you multiply it by millions of boards. It could even be one of the reasons EVGA decided to pull the plug on its GPU business. The company noted in its decision that Nvidia wouldn’t share details on upcoming GPUs until the last minute. This sounds an awful lot like that exact scenario.
Either way, this is both good news and bad news for those looking to buy one. The good news is these new GPUs should offer great temperatures, with minimal noise. The bad news is they might not fit into your case. Cards like the ones above use 3.5 slots, and there are also quad-slot boards too. Some cards are over 13 inches long as well. You can also kiss any plans for PCIe expansion cards goodbye. That might not be an issue for a lot of folks; most add-in features like audio or USB expansion have all moved to the motherboard in the past few years. Speaking personally, I haven’t used a PCIe expansion card in years.
The only issue might be GPU sag, but there are aftermarket options to handle that. In fact, the RTX 4090 FE even has an opening on its edge for case-mounted brackets to support it. It’s unclear if any AIB cards will offer something similar. Nvidia’s RTX 4090 officially launches next week, on Oct. 12. The RTX 4080 cards will arrive sometime in November.
Source by www.extremetech.com