NVIDIA Blackwell GB202 GPU For Flagship GeForce RTX 5090 To Measure Around 744mm2, 22% Larger Than 4090’s AD102

NVIDIA’s Blackwell GB202 GPU for the GeForce RTX 5090 Flagship will see a hefty die size increase versus the existing Ada flagship.

NVIDIA GeForce RTX 5090’s GPU Will Be 22% Larger Than RTX 4090, GB202 Chip With 744mm2 Die Size

The NVIDIA GeForce RTX 5090 graphics card will be the flagship solution within the GeForce RTX 50 “Blackwell” gaming lineup. This chip is expected to bring in some huge gains over the RTX 4090, but it looks like the chip will also be gargantuan in terms of die size while making use of TSMC’s 5nm process node.

GB202 Die Size 24*31
Package Size 63*56

— MEGAsizeGPU (@Zed__Wang) November 22, 2024

According to MEGAsizeGPU, the NVIDIA GB202 GPU which will be featured on the RTX 5090 will be roughly 24x31mm in terms of die size, which totals to 744 mm2. The GPU will feature a package size of 63x65mm which is roughly 4095mm2. For reference, the RTX 4090’s AD102 GPU featured a die size of 609mm2 and a package size of 2601mm2 (51x51mm). This is a 22% increase in the die size and a 57% increase in the packaging area.

Now a bigger die size doesn’t mean that the chip will eventually be more power-hungry or run hot, but it should be remembered that it is still a smaller die size than the 754mm2 Turing TU102 chip which was launched with the first generation RTX “20” cards. The chip included Tensor and RT core technology for the first time on an NVIDIA GeForce “Gaming” GPU and since then, we have seen a reduction in die size for flagship chips for each consecutive process node optimization.

NVIDIA GeForce RTX 5090 32 GB Graphics Card

Starting with the details, we first have the NVIDIA GeForce RTX 5090, which is going to be the top GPU in the lineup. The graphics card is expected to feature the PG144/145-SKU30 PCB design & that’s going to incorporate the GB202-300-A1 GPU core. The GPU will have 170 SMs enabled out of the total 192 SMs and will feature 21760 cores instead of the total of 24,576 cores. This marks an 11.4% reduction, which is slightly more than the11.1% reduction of the RTX 4090 versus its full AD102 die.

On the memory side, the NVIDIA GeForce RTX 5090 is going to be a beast with 32 GB of GDDR7 VRAM capacity running across a 512-bit interface. The VRAM will operate at 28 Gbps speeds for up to 1792 GB/s of bandwidth plus a bigger L3 cache and new memory compression techniques are also going to boost the overall bandwidth available to the GPU. The card will also come with a 600W TBP rating, though it should be mentioned that the TBP value shouldn’t be compared with actual real-world or gaming power consumption as that ends up much lower. Previous leaks have suggested a 2-slot cooler for the Founders Edition model of the RTX 5090.

The NVIDIA GeForce RTX 5090 is expected to launch at CES 2025 so expect more information in the coming months.

NVIDIA GeForce RTX 50 GPU Specs (Preliminary):