More details about what will probably be marketed as Nvidia’s GeForce RTX 40 series has seeped through the cracks. Serial leaker Kopite7kimi today shared some insight into the full-fat AD104 GPU. This GPU is probably destined for an upcoming RTX 4070 Ti graphics card. According to the leaker, with a good track record with regard to these things, this next gen 70 Ti product will “easily match RTX 3090 Ti.” As ever with leaks, take the news with a pinch of salt.
To see if Kopite7kimi’s assertions hold water, it is worth constructing a comparison table of the specs we know about the current gen champ, vs the purported specs of the upcoming RTX 4070 Ti. Who knows, the RTX 4070 Ti may just make be a future contender for our best GPUs.
|
Nvidia GeForce |
RTX 3090 Ti |
RTX 4070 Ti |
|---|---|---|
|
Architecture |
Ampere |
Ada Lovelace |
|
GPU |
GA102-350-A1 |
AD104 full fat |
|
Transistors |
28.3 billion |
30 billion |
|
Node |
Samsung 8nm |
TSMC N4 |
|
CUDA cores |
10,752 |
7,680 |
|
Memory quota |
24GB GDDR6X |
12GB GDDR6X |
|
Memory speed |
21Gbps |
21Gbps |
|
Memory bus |
384-bit |
192-bit |
|
TDP |
450W |
400W |
Most of the above AD104 specs have come from our ‘everything we know’ article about the Nvidia Ada Lovelace and GeForce RTX 40-Series. Of course the RTX 3090 Ti specs are common knowledge.
If you consider the table above, the upcoming Ada Lovelace GPU clearly has advantages in transistor count, and a big change in process node. It is also expected to be pushed to significantly higher clock speeds compared to the RTX 3090 Ti, which is why the 4070 Ti presumably gets such a high TDP. High TDP averse folk might look at the regular RTX 4070 with a purported TDP nearer to 300W. However, the purported GeForce RTX 4070 Ti should exceed or draw parity to the RTX 3090 Ti with a much lower bill of materials (e.g. smaller silicon die) and help bring 2022 flagship performance to much more affordable levels.
One area where the previous gen graphics card is much stronger, is in the memory department. They are both seemingly use GDDR6X at 21Gbps, but the expected configuration of the RTX 4070 Ti has half the memory and half the bus width, which again will help bring down costs. There will be a certain resolution and quality setting in every game where the greater memory bandwidth of the RTX 3090 Ti will make itself known, but we need thorough comparative testing in a host of games to find it.
In summary, it is exciting to think of Ampere flagship+ performance coming to the almost ‘midrange’, but we have some worries about power and cooling. AMD with its Radeon 7000 series and 5nm RDNA 3 GPUs, seems to be keeping a lid on power consumption – so the upcoming graphics card battle is going to be pretty interesting.
