GeForce RTX 4070 Super, RTX 4070 Ti Super and RTX 4080 Super are coming to market
The GeForce RTX 4000 GPUs run their second year on the market. And like the previous two generations, they’ll get a shake-up for this second year with new models. Nvidia has once again turned to the label used during the Turing generation and will release three new gaming cards this month: the GeForce RTX 4070 Super, GeForce RTX 4070 Ti Super and GeForce RTX 4080 Super. And they should improve the bang-for-buck ratio by quite a bit.
GeForce RTX 4070 Super
These three new products have just been revealed at CES 2024. Starting with the most affordable of the new models, this card, which is a sort of upgrade to last year’s original RTX 4070 model, is again based on the 4nm AD104 chip with 35.8 billion transistors. Of course, it was cut down a lot in the previous card, whereas the RTX 4070 Super has much more of it available to the owner.
There are 56 SM blocks active (in five GPC blocks and 28 TPC blocks), giving it 7168 Ada Lovelace shaders, 56 (third generation) RT cores and 224 tensor cores for AI (these are fourth-generation ones in the Ada Lovelace architecture). The GPU has 224 texture units and 80 ROP units (their number is determined by the five GPCs).
The GPU is clocked at 1980 MHz base and 2475 MHz in boost. The shaders are supposed to deliver 35.5 TFLOPS of theoretical performance in FP32 calculations, and as many as 568 TOPS in AI (which is likely a theoretical throughput for INT4 ops, or for INT8 using the sparsity function).
The card leaves the memory at the same parameters as it was on the RTX 4070 model, i.e. 12 GB GDDR6X (however, it should be said that the same was enough for the more powerful RTX 4070 Ti model). The bus is therefore 192-bit and the effective clock speed is 21.0 GHz, which yields 504 GB/s. However, the memory performance available for graphics applications will also be helped by the large L2 cache, which is 48 MB on the RTX 4070 Super.
The graphics card will have a reference TDP (or TGP, or Total Graphics Power) of 220 W. This of course only applies to reference cards, overclocked non-reference models can have their TGP increased. According to Nvidia, typical power draw in games should be around 200 W, around 16 W in AV1 video playback and 11 W when idle.
The power supply should usually be handled by a 12+4pin and at least a 650W PSU is recommended. Adapters from two 8-pins should be included with the card. When supplying power directly by the native ATX 3.0 / PCIe 5.0 cable, called 12VHPWR or more recently 12V-2×6, the PSU should be able to output 300W through this cable.
The card will be available in a Founders Edition reference design, which is based on the one used with the regular GeForce RTX 4070 and is dual-slot with two 92mm fans, three DisplayPort outputs and one HDMI. The Super cards will use a darker black-themed version of the design. At the same time though, there will of course be a number of models from partners with a variety of custom coolers available.
This card will go on sale as the first one next week on Wednesday 17 January at 3pm CET. On the same day, reviews of cards that will be at a higher price than MSRP are also to be released. However, models with a price not exceeding the suggested amount will be allowed to be reviewed one day earlier, on 16 January. The suggested price is supposed to be 599 USD like what the GeForce RTX 4070 launched at last year. Performance should be better (we’ll see after the reviews, but there’s probably a decent chance it will be close to the more expensive GeForce RTX 4070 Ti), that’s what the price/performance ratio will improve on. The official price should be around 662 EUR in Europe (inclusive of 21% VAT).
For now, the GeForce RTX 4070 should coexist with this newly released graphics card, while selling down a rung at a reduced price, Nvidia is lowering its MSRP to 549 USD (~605 € with VAT).
GeForce RTX 4070 Ti Super
Such an upgraded model will then also replace the RTX 4070 Ti. However, the latter is to be discontinued (it could probably be replaced by the RTX 4070 Super to some extent) and only the new RTX 4070 Ti Super will continue to be sold instead. In this case, Nvidia is swappinge the lower-end AD104 chip for the higher-end AD103 with 45.9 billion transistors (previously only used in the RTX 4080). In this new graphics card, the GPU is cut down to 8448 shaders, which is 66 SM blocks, and therefore 66 RT cores, 264 tensor cores and 264 texture units.
The graphics card has a reference clock speed of 2340 MHz base and 2610 MHz in boost, which should result in a theoretical performance of 44.1 TFLOPS in FP32 calculations. There is no change in the TDP (TGP) for this graphics card compared to the original RTX 4070 Ti, it remains set at the same 285W.
According to Nvidia, typical power draw in games should be just around 226 W, around 17 W in AV1 video playback and 12 W when idle. The gaming power claimed to be much lower than 285 W is however likely a result of basing this number on an average of games that include undemanding titles that do not fully load the GPU. Power values like these have to be taken from grain of salt, consider them similar to officially given performance claims. Also note that OC custom cards can behave differently.
However, along with the upgrade to the AD103 chip, this graphics card also gets a memory bus expansion to 256 bits, so its memory capacity also increases to 16 GB GDDR6X. Its bandwidth will be 672 GB/s (effective clock speed 21.0 GHz). The L2 cache of the GPU has 48MB capacity.
The power supply will be handled by one 12pin connector as well. For this model, there might not be Founders Edition models, or at least we have no information about them yet.
The GeForce RTX 4070 Ti Super will go on sale January 24 at 3pm CET, a week later (but again, reviews will be spread over two days). The suggested price is said to be 799 USD, which should correspond to ~882 EUR in Europe (VAT inclusive). The original GeForce RTX 4070 Ti also launched at 799 USD, but the faster specs of the Super version will again translate into a better price/performance ratio.
GeForce RTX 4080 Super
The top model in the Super series will be the GeForce RTX 4080 Super graphics card, which in turn is an upgrade of the RTX 4080 model which it is replacing on the market. It should be said that the RTX 4090 isn’t being replaced by the RTX 4080 Super, and will remain above it as the top high-end model (and also apparently with an unchanged price of 1599 USD). In China and regions affected by US sanctions, there will be a separate high-end SKU, the GeForce RTX 4090D, also sitting above the RTX 4080 Super.
The GeForce RTX 4080 Super, however, brings perhaps the smallest improvement out of the three models. The AD103 chip used has finally enabled all the remaining units which were left inactive in the RTX 4080 to improve yield. This means a shift from 9728 to 10,240 shaders, 80 RT cores and 320 tensor cores and 320 texture units. The GPU will have all 112 ROPs active (the L2 cache capacity, 64 MB is fully enabled as well).
Nvidia will also increase the clock speeds of this model, the Super card has a base clock speed of 2295 MHz and the boost is listed at 2550 MHz. This means 52.2 TFLOPS of theoretical computing performance in FP32 operations.
Memory capacity remains at 16 GB GDDR6X with a 256-bit bus. Nvidia has increased the clock speed a bit, pushing it to effective 23.0 GHz and increasing the bandwidth to 736 GB/s. On the other hand, again unchanged is the power draw, the TDP (or TGP) has been left at 320 W. Again, a single 12+4pin power connector is used.
Same as with the other cards, Nvidia claims lower typical power draw in games, allegedly around 246 W, (around 22 W in AV1 video playback and 15 W when idle). Again that is likely a result of basing this number on an average of games that include undemanding titles that do not fully load the GPU. The same disclaimer that we already stated is true for all the cards. OC custom cards can also behave differently, with raised gaming but also idle and video power.
This card will have an Founders Edition design sold directly by Nvidia. It is again using the characteristic design of the Ampere and Ada Lovelace cards (in black), but RTX 4080 Super uses the larger version that is three slots thick.
This model will be released another week later, with sales starting January 31 at 3pm CET. Reviews should again come out in the same manner (on the 30th, cards at MSRP and on the 31st, reviews of more expensive models). The official suggested price is set at 999 USD. The MSRP should mean european price around 1103 EUR after VAT. The launch price of this Super card turns out to even be lower than what the “ordinary” GeForce RTX 4080 started at (1199 $), so the price/performance ratio should improve nicely even for this card, despite its specifications undergoing a slightly smaller increase than the two cheaper Super cards.
We have samples of all three models coming in for testing, so you can look forward to HWCooling’s reviews.
Source: Nvidia
English translation and edit by Jozef Dudáš
⠀
⠀
