Introduction
Today, March 4th, 2025, the NVIDIA GeForce RTX 5070 Founders Edition is launching, with the availability of video cards, as well as factory overclocked non-MSRP video cards, on March 5th, 2025. The GeForce RTX 5070 is NVIDIA’s $549 GeForce RTX 50 series GPU based on the Blackwell architecture. The GeForce RTX 5070 is a 12GB GDDR7 video card with features such as DLSS 4 and Multi-Frame Generation. If you haven’t checked out our recent GeForce RTX 50 series reviews, we have the GeForce RTX 5090 FE, GeForce RTX 5080 FE, and GeForce RTX 5070 Ti video cards reviewed on our site.
The GeForce RTX 5090 sits at the top of the GPU stack as a flagship video card with a $1,999 MSRP. The GeForce RTX 5080 sits much farther down the stack as a $999 MSRP video card with 16GB of GDDR7. The GeForce RTX 5070 Ti fits in under the GeForce RTX 5080 also as a 16GB GDDR7 video card at $749 MSRP. The new GeForce RTX 5070 being launched today fits in under the GeForce RTX 5070 Ti, but as a 12GB GDDR7 video card for $549, and uses a smaller GB205 die compared to the RTX 5070 Ti’s GB203.
The GeForce RTX 5070 utilizes the GB205 die, which interestingly is not the full specification of GB205, it is cut down by 1 TPC, and 2 SMs from the full specification of GB205. The GeForce RTX 5070 GPU is based on the NVIDIA Blackwell architecture, and the GB205 is based on the TSMC 4nm 4N NVIDIA custom process, which is the same as the previous generation Ada Lovelace. The NVIDIA GeForce RTX 5070 supports a PCI-Express 5.0 (Gen5) interface.
The GeForce RTX 5070 at $549 can be equated as a successor to the GeForce RTX 4070, which launched in April of 2023 at an MSRP of $599. It can also be equated as a successor to the more recent GeForce RTX 4070 SUPER launched in January of 2024 with an MSRP of $599 as well, in which the GeForce RTX 4070 officially moved down to $549 at that time. Both the GeForce RTX 4070 and GeForce RTX 4070 SUPER were based on the same AD104 die, with the SUPER getting a large upgrade in shader and raster performance. It is, therefore, a departure, in comparison to the previous generation, for the RTX 5070 to utilize the “xxx05” die, while the previous generation in this model used the “xxx04” die.
The NVIDIA GeForce RTX 5070 consists of 5 GPCs, 24 TPCs, and 48 SMs with 128 CUDA Cores per SM. When you add all this up, the NVIDIA GeForce RTX 5070 has 6,144 CUDA Cores, 192 5th Gen Tensor Cores, 48 4th Gen RT Cores, 80 ROPs, and 192 Texture Units. The L1 Data Cache/Shared memory size is 6144KB, and the L2 cache size is 49152 KB. The GPU Boost Clock is set at 2512MHz. Memory consists of 12GB of GDDR7 on a 192-bit memory bus at 28Gbps, giving it 672GB/s of memory bandwidth. The TGP (Total Graphics Power) is 250W.
| GeForce RTX 5070 | GeForce RTX 4070 SUPER | GeForce RTX 4070 | |
|---|---|---|---|
| GPU Codename | GB205 | AD104 | AD104 |
| Architecture | Blackwell | Ada Lovelace | Ada Lovelace |
| Process | TSMC 4N NVIDIA | TSMC 4N NVIDIA | TSMC 4N NVIDIA |
| SMs | 48 | 56 | 46 |
| CUDA Cores | 6144 | 7168 | 5888 |
| RT Cores | 48 (4th Gen) | 56 (3rd Gen) | 46 (3rd Gen) |
| Tensor Cores | 192 (5th Gen) | 224 (4th Gen) | 184 (4th Gen) |
| ROPs | 80 | 80 | 64 |
| Texture Units | 192 | 224 | 184 |
| GPU Boost | 2512MHz | 2475MHz | 2475MHz |
| VRAM | 12GB GDDR7 | 12GB GDDR6X | 12GB GDDR6X |
| Memory Interface | 192-bit | 192-bit | 192-bit |
| Memory Clock | 28Gbps | 21Gbps | 21Gbps |
| Memory Bandwidth | 672 GB/s | 504 GB/sec | 504 GB/sec |
| TGP | 250W | 220W | 200W |
NVIDIA GeForce RTX 5070 Pictures
The NVIDIA GeForce RTX 5070 Founders Edition is a small video card, and not very heavy, it will fit very well in a small case or small build. The video card measures 242mm in length (9.5″), 112mm (4.4″) in width, and is a 2-slot video card. This actually makes the GeForce RTX 5070 just a hair shorter than the GeForce RTX 4070 and GeForce RTX 4070 SUPER, which have a length of 244mm (9.6″) in length, with the same width and 2-slot nature. In terms of power, NVIDIA recommends a 650W PSU using 2x PCIe 8-pin cables (adapter in box) or a 300W or greater PCIe Gen 5 cable. The video card does support PCI-Express 5.0 x16.
The GeForce RTX 5070 Founders Edition utilizes NVIDIA’s custom dual-fan cooling solution, with half of the GPU venting air passing through the heatsink completely, but the other half not, it has a solid backing. The backside of the vents do have the curved nature, however, to improve air pressure and noise levels.
Both of the fans pull in air and push it out, there is a bit of openness on the I/O shield where hot air can escape as well. The 12VHPWR 2×6 connector is positioned at an angle atop the video card, and the included power adapter cable is quite bulky on the connector end; it’s built very sturdy. The video card also features the top and bottom air deflectors to keep hot air from circulating back into the video card.
In terms of display support, you’ll find 3x DisplayPort 2.1b with UHBR20: up to 4K 480Hz or 8K 165Hz with DSC and 1x HDMI 2.1b: up to 4K 480Hz or 8K 120Hz with DSC, Gaming VRR, and HDR. There is no RGB or LEDs on this video card, which should make for a very nice stealthed-out kind of build.
Discussion (11 replies)
Join Discussion →umm, trying to be positive about this card, DLSS4 (multi frame-gen)?, umm. :unsure:
Edit: I thought of a positive, it could help promote RX 9070 / B580 sales!
Wow RTX 4090 performance for just $549, right?.... right? :rolleyes: :rolleyes: :rolleyes: ;)
Lol the ROPS comment at the end was golden.
👇
4090 at $549... lol!
And on top of all the fun being made....
[URL unfurl="true"]https://videocardz.com/newz/nvidia-says-geforce-rtx-5070-founders-edition-wont-be-availble-at-launch[/URL]
Wow... just..... wow...
This is nVidia being worried about the AMD launch
It's also a way for nVidia to screw over their AIBs by dumping the burden of selling a "not too hot" card onto them (and having them shoulder the costs -- and projected losses).
It's noteworthy that the "desired" custom AIB cards of the 5070Ti/5080/5090 class all came with very hefty premiums over the FE "MSRP-ish" pricing -- and there weren't too many of either the FE or AIB models, to boot. That's nVidia prioritizing AI (no surprise there) while also proceeding to mess around big with the AIBs. Jensen has long wanted to cut the AIBs right out of the market (he's frequently stated his admiration of Apple's closed ecosystem model, and also see what happened to EVGA, for example), and he already had the good connections (along with considerable amounts of R&D on the FE designs) with OEMs like Foxconn, etc to be halfway there. And now that Jensen has the $$$ with AI (and nVidia is a multi-trillion USD company), he has the raw capital to aggressively move strongly to cut the AIBs right to the core (by being financially able to ignore their complaints, and tie them up in courts as needed, should they sue) ...
IMHO, Jensen himself has probably envisioned the "demise" of the AIB market for nVidia gaming GPUs in 2-3 years (maybe a bit longer for the commercial/AI stuff).
I wouldn't mind them 'undercutting AIBs' if I could walk into Best Buy / Microcenter and get whichever FE model I budgeted for.
But that ain't the case.
AI priority, of course (nVidia shareholders would revolt if Jensen dedicated TSMC GPU inventory to the VASTLY less money-making gaming market), but once nVidia chases the AIBs into bankruptcy (of their GPU divisions, a la EVGA), Jensen will provide more inventory, but at increased prices, too (virtual monopoly, as expected).
With that in mind, we can expect theoretical "RTX 8060s" (relabeled 8050s?) to hit the $500-700 USD range ...
Side note: I really hope I don't end up getting the mediocre 5070 as my GPU upgrade in the days/weeks to come (there's a Microcenter near me, so I'm planning to be there a few hours early this Thursday for a shot at 9070 XT or 9070) -- I just sold my RX 6700 XT, which has been showing its age as of the past year or so ...