Nvidia’s next Ada Lovelace graphics card is nearly upon us. GeForce RTX 4070 officially arrives Wednesday, April 12, and all the pertinent details are leaking onto the web over a week in advance.
Hot on the heels of a confirmed $599 price tag, the enterprising team over at Videocardz has now spilled the GPU’s key characteristics, courtesy of a slide taken direct from an Nvidia presentation.
While most specifications have already been bandied about in recent weeks, the leak confirms Nvidia’s plans to position RTX 4070 as an energy-efficient choice. Total graphics power is rated at 200W, though Team Green is keen to point out an average power draw of just 186W when gaming, 16W during video playback, and a mere 10W when idle.
Comfortably more frugal than the outgoing RTX 3070 Ti, which chimed in at a 290W TGP yet ought to be soundly beaten on all fronts by the latest-generation successor. The missing details allow us to fill in some more gaps in our Club386 Table of Doom:
GeForce | RTX 4090 | RTX 4080 | RTX 4070 TI | RTX 4070 |
---|---|---|---|---|
Launch date | Oct 2022 | Nov 2022 | Jan 2023 | Apr 2023 |
Codename | AD102 | AD103 | AD104 | AD104 |
Architecture | Ada Lovelace | Ada Lovelace | Ada Lovelace | Ada Lovelace |
Process (nm) | 4 | 4 | 4 | 4 |
SMs | 128 of 144 | 76 of 80 | 60 of 60 | 46 of 60 |
CUDA cores | 16,384 | 9,728 | 7,680 | 5,888 |
Boost clock (MHz) | 2,520 | 2,505 | 2,610 | 2,475 |
Peak FP32 TFLOPS | 82.6 | 48.7 | 40.1 | 29.1 |
RT cores | 128 | 76 | 60 | 46 |
Tensor cores | 512 | 304 | 240 | 184 |
Texture units | 512 | 304 | 240 | 184 |
Memory size (GB) | 24 | 16 | 12 | 12 |
Memory type | GDDR6X | GDDR6X | GDDR6X | GDDR6X |
Memory bus (bits) | 384 | 256 | 192 | 192 |
Memory clock (Gbps) | 21 | 22.4 | 21 | 21 |
Bandwidth (GB/s) | 1,008 | 717 | 504 | 504 |
L2 cache (MB) | 72 | 64 | 48 | 36 |
Power (watts) | 450 | 320 | 285 | 200 |
Launch MSRP ($) | 1,599 | 1,199 | 799 | 599 |
Early impressions going by spec are that Nvidia’s two RTX 4070 Series cards are entirely different beasts. Cleaving a good chunk of firepower from the AD104 blueprint, the upcoming non-Ti card sees core count reduced by almost a quarter, and Nvidia’s desire to give RTX 4070 Ti a wide berth is reflected in boost clock. Not only does 4070 regular feature fewer cores, they’re purposely slower, too.
Still, while the cuts are savage, there are positives, too. The backend retains intact with 12GB of GDDR6X memory via a 192-bit bus, as does Ada’s extra dollop of onboard cache. 36MB of L2 may not sound like much, but compared to just 4MB for RTX 3070 Ti, should help alleviate memory bottlenecks at the target 1440p resolution.
Such credentials naturally lend themselves to a more petite card. While current RTX 40 Series have ranged from large to freaking ginormous, RTX 4070 ought to be a compact proposition well suited to small-form-factor builds. Granted, we’re still likely to see ill-advised partner cards that spill into three or four slots, but finding a cool, quiet, dual-slot model shouldn’t be a problem.
This is a GPU that ought to be cheap to produce, yet the $599 price tag seems about right taking inflation into account. We’ll need in-house performance numbers before passing a truer verdict; stay tuned for that next week.