In a few months, Nvidia will bring a new generation of GeForce to the world, the GeForce RTX 40 series. The GeForce RTX 4090 Ti and GeForce RTX 4090 will be the flagships. It will probably reach a new level in terms of performance.
Like Ampere, the Ada Lovelace architecture (GeForce RTX 40 series) will primarily focus on the high-end. Rumors point to two graphics cards.
- Geforce RTX 4090 Ti
- GeForce RTX4090
GeForce RTX 4090 Ti and RTX 4090 specifications.
The GeForce RTX 4090 Ti and RTX 4090 will likely be the only references to use the AD102 GPU. This will be Nvidia’s flagship project based on the Ada Lovelace architecture.
The RTX 4090 Ti usually benefits from the “full” configuration of this GPU with the presence of 144 SM, enough to give rise to 18,432 CUDA cores. The GPU will carry 96MB of L2 cache and will have 384 ROPs. At the moment there are still unknowns around the frequencies. Assuming that Nvidia relies on an N4 engraving from TSMC, values between 2 and 3 GHz are possible.
The card should carry 24GB of GDDR6X at 24Gbps operating over a 384-bit memory bus. Such a configuration provides a bandwidth of 1,152 TB / s. The set is accompanied by a respectable 600W thermal jacket. Its power supply is based on a single additional connector, a 16-pin connector based on the PCIe 5.0 standard. Note that some rumors speak of 48GB GDDR6X and not 24.
GeForce RTX 4090 Ti, a leak reminiscent of 48 GB GDDR6X and a TGP of 900 watts
Custom models can be even more brutal with two 16-pin PCIe 5.0 connectors to offer some overclocking opportunities.
Snapshots of his hypothetical heatsink have surfaced recently. It looks imposing, suggesting that a map needs space and space. Four-slot models are even possible for Custom Editions.
For its part, the GeForce RTX 4090 will also be positioned in the high-end, but through a slightly less “violent” configuration. We should find a “lighter” AD102 GPU where 126 SMs bring 16,128 CUDA cores to life. The L2 cache would also be less important, as would the number of ROPs. Again, the frequencies are unknown. They are said to be lower than those of the RTX 4090 Ti, but just as important as those of the current GeForce RTX 3090 Ti.
It will be packed with 24GB of GDDR6X at 21Gbps and powered by a 384-bit memory bus. The bandwidth should thus climb to 1,008 TB/s. Its thermal envelope should be around 450 watts, i.e. with the current RTX 3090 Ti.
Ada Lovelace’s graphics architecture will likely enable 4th Gen Tensor cores, 3rd Gen RT cores, optimized and improved versions of the NVENC encoder and NVCDEC decoder while adding support for the latest APIs and technologies such as DLSS, Reflex, Broadcast and Resizable feature -BAR, Freestyle, Ansel, Highlights, Shadowplay and G-SYNC.
GeForce RTX 4090 Ti and GeForce RTX 4090 – alleged specifications
|Geforce RTX 4090 Ti||GeForce RTX4090|
|GPU’s||AD 102-350 – Ada Lovelace||AD 102-300 – Ada Lovelace|
|delicacy of engraving||TSMC4N||TSMC4N|
|TMU / ROP||? / 384||? / 384|
|Tensor/RT cores||? /?||. /?|
|Computing power (FP32)||~103 TFLOPs||~90 TFLOPs|
|VRAM||24GB GDDR6X||24GB GDDR6X|
|memory speed||24 Gbps||21 Gbps|
|Start||July 2022?||July 2022?|
GeForce RTX 4090 Ti and RTX 4090, performance
As far as performance is concerned, we can only give theoretical numbers for the time being. For several months, the rumors on this topic have all been going in the same direction. The GeForce RTX 40 series lineup should allow Nvidia to exceed 100 TFLOPs of performance in the “consumer” market. This limit would be exceeded by the GeForce RTX 4090 Ti (103 TFLOPs), while the GeForce RTX 4090 would be at 90 TFLOPs. For comparison: The GeForce RTX 3090 Ti has 40 TFLOPs and the GeForce RTX 3090 has 36 TFLOPs (FP32).
All of this is encouraging, as 100 TFLOPs means double the processing power compared to the current flagship (the RTX 3090 Ti). But be careful, the computing power cannot be directly converted into “in-game” performance.
GeForce RTX 4090 Ti and RTX 4090, prices.
Finally, on the price side, don’t expect miracles. The GeForce RTX 3090 Ti & RTX 3090 are currently the most expensive single GPU graphics cards. Nvidia announces suggested retail prices of $1999 and $1499. So we have to be prepared for a similar or even higher positioning if there are small price increases here and there.
Given the current economic environment, this is likely.
#GeForce #RTX #GeForce #RTX #price #performance #specifications #GinjFo