NVIDIA's next-gen GeForce RTX 4080 graphics card has some fresh new specs, with leaker "kopite7kimi" providing some updates on the second-fastest Ada Lovelace GPU.
The new GeForce RTX 4080 graphics card was first rumored with a rather large 420W, but the last purported power consumption on the RTX 4080 was scaled down to 320W. But now, kopite7kimi is saying that the GeForce RTX 4080 will have total card power of 340W, but more interestingly: 16GB GDDR6X at a now faster 23Gbps bandwidth.
NVIDIA's current-gen Ampere-based GeForce RTX 3080 has 10GB of GDDR6X, which isn't enough in its class against AMD's fastest Radeon RX 6900 XT (at the time) with 16GB GDDR6, and the replacement of the RX 6900 XT in the Radeon RX 6950 XT continued with 16GB GDDR6X. NVIDIA's faster GeForce RTX 3080 Ti only has 12GB of GDDR6X, while the real dominance happens with the GeForce RTX 3090 and GeForce RTX 3090 Ti both with 24GB of GDDR6X memory.
- Read more: NVIDIA GeForce RTX 4080 latest: 320W power, previously rumored at 420W
- Read more: NVIDIA GeForce RTX 4080 + RTX 4070: early 3DMark estimates teased
- Read more: NVIDIA's next-gen GeForce RTX 4080: AD103 GPU with 420W TDP rumored
- Read more: Chinese GPU supplier: we're getting GeForce RTX 40 cards in August
- Read more: NVIDIA GeForce RTX 4080 rumored specs: same 16GB GDDR6X, now at 21Gbps
The larger 16GB of GDDR6X on the GeForce RTX 4080 will be great to see, but the faster 23Gbps bandwidth is very welcomed. Micron will be providing NVIDIA with faster GDDR6X memory modules for the Ada Lovelace GPU family, with 23Gbps dies reportedly being used on the GeForce RTX 4080.
Previously, rumors had NVIDIA's new GeForce RTX 4080 featuring 16GB of GDDR6X memory but at 21Gbps. So the upgrade to the faster GDDR6X memory at 23Gbps is a decent bump, with what I'm sure is going to be some seriously good overclocking headroom on custom AIB models of the GeForce RTX 4080 graphics cards.
In the tweet, kopite7kimi explains:
The use of 23Gbps GDDR6X memory on a 256-bit memory interface will provide up to 736GB/sec of memory bandwidth, which is actually a little less than the GeForce RTX 3080 and its 10GB GDDR6X on a 320-bit memory bus with 760GB/sec memory bandwidth. We should expect NVIDIA to use an integrated next-gen memory compression suite that will help boost memory bandwidth, even on the lower 256-bit memory bus.
NVIDIA is expected to use the AD103-300 GPU inside of the GeForce RTX 4080 graphics card, with 9728 CUDA cores, or 76 SMs enabled in total (out of the possible 84 SMs). Previously, the GeForce RTX 4080 was expected to have 10,240 CUDA cores, or 80 SMs.
We are expecting the GeForce RTX 4080 to have 48MB of L3 cache and lower ROPs count, with the full AD102 GPU expected to have 64MB of L2 cache and up to 224 ROPs.
NVIDIA GeForce RTX 40 series GPU power consumption:
- NVIDIA GeForce RTX 4090: around 450W
- NVIDIA GeForce RTX 4080: around 340W
- NVIDIA GeForce RTX 4070: around 285W
- NVIDIA GeForce RTX 4060 Ti: around 250W
- NVIDIA GeForce RTX 4060: around 220W
GeForce RTX 4080 (new rumored specs +new power +23Gbps GDDR6X)
- GPU: AD103-300-A1
- GPU clocks: Unknown
- CUDA cores: 10240
- SMs: 80
- VRAM: 16GB GDDR6X @ 23Gbps (previous rumor @ 21Gbps, non-X)
- Memory bus: 256-bit
- TDP: 340W or so (previously 320W, before that 420W)
GeForce RTX 4070 (new rumored specs +new power)
- GPU: AD102-275 (used to be AD104-400)
- CUDA cores: 7168
- SMs: 56
- VRAM: 12GB GDDR6 (up from 10GB GDDR6) @ 18Gbps
- Memory bus: 192-bit (up from 160-bit)
- TDP: 280W or so (down from 300W)