Nvidia launches RTX 50 Blackwell GPUs: $2,000 RTX 5090, $1,000 RTX 5080, RTX 5070 / Ti are $549 and $749

Extremely anticipated: Nvidia has lastly unveiled the brand new GeForce 50 collection, together with DLSS 4 know-how. 4 high-end GPUs have been introduced, and it will not be lengthy till you will get your fingers on one. The RTX 5080 and RTX 5090 are set to launch first on January 30, adopted by the RTX 5070 and RTX 5070 Ti in February. Now, it is time to share our ideas on what was revealed throughout the CES 2025 keynote – together with all of the specs, pricing, and efficiency claims.

The brand new GPUs

The brand new RTX 5090 flagship comes packing considerably extra {hardware} over its predecessor. Not solely does this GPU use Nvidia’s new Blackwell structure, nevertheless it additionally packs considerably extra CUDA cores, better reminiscence bandwidth, and the next VRAM capability. The SM depend has elevated from 128 with the RTX 4090 to a whopping 170 with the RTX 5090 – a 33% improve within the core measurement. Nevertheless, GPU increase clocks have decreased barely, from 2.5 GHz to 2.4 GHz.

The reminiscence subsystem is overhauled, now that includes GDDR7 know-how on an enormous 512-bit bus. With this GDDR7 reminiscence clocked at 28 Gbps, reminiscence bandwidth reaches 1,792 GB/s – a close to 80% improve over the RTX 4090’s bandwidth. It additionally consists of 32GB of VRAM, essentially the most Nvidia has ever supplied on a client GPU.

With Blackwell nonetheless constructed utilizing TSMC 4nm know-how, the board energy score has risen to accommodate these cores and reminiscence, growing to a mammoth 575W – 28% larger than the RTX 4090.

The RTX 5080 makes use of a a lot smaller GPU configuration, primarily based on a completely completely different GPU die, with simply 84 SMs – 4 greater than the RTX 4080 Tremendous from the final era.

Nearly every little thing about this mannequin is half of the RTX 5090: just below half the CUDA cores, half the reminiscence bus width at 256-bit, half the VRAM capability at 16GB, and just a little over half the reminiscence bandwidth at 960 GB/s by the usage of GDDR7 clocked at 30 Gbps. Traditionally, this configuration would align with an RTX 5070-tier GPU, so positioning it as a 5080 looks like a disappointing transfer for players.

The RTX 5070 Ti is additional reduce down, that includes 70 SMs – 4 greater than the earlier era’s RTX 4070 Ti Tremendous. Nevertheless, increase clocks have dropped from 2.61 GHz to 2.45 GHz. The reminiscence bus width stays the identical because the RTX 5080’s 256-bit configuration, that means it additionally has 16GB of GDDR7 reminiscence. This time, the reminiscence is clocked at 28 Gbps, leading to 896 GB/s of bandwidth. The 5070 Ti is rated at 300W, whereas the RTX 5080 is rated at 360W.

Then there’s the RTX 5070. This mannequin options 48 SMs, simply two greater than the RTX 4070 from the earlier era. It features a 192-bit reminiscence bus and 12GB of VRAM, which feels inadequate for this tier of product. Increase clocks are solely barely larger than the RTX 4070, growing from 2.48 GHz to 2.51 GHz, with a complete energy score of 250W.

All new GeForce 50 collection GPUs embody up to date connectivity, resembling PCIe 5.0 x16 for the introduced fashions and DisplayPort 2.1b UHBR20, which introduces help for lively cables. Moreover, they characteristic HDMI 2.1 and a single 16-pin energy connector on every mannequin.

GPU efficiency and pricing

Now let’s dive into an important facets of Blackwell. There’s each good and dangerous right here. The GeForce RTX 5090 is outrageously costly, as anticipated, priced at $2,000. In the meantime, the RTX 5080, with half the GPU {hardware}, is priced at $1,000 – much less excessive than many rumors urged, although nonetheless fairly costly.

The surprises for us have been the RTX 5070 Ti at $750 and the RTX 5070 at $550 – each $50 decrease than their counterparts from the earlier era at launch.

As for the efficiency claims… Nvidia has – as regular – used its advertising to obscure precise gaming efficiency. RTX 50 GPUs help DLSS 4 multi-frame era, which previous-generation GPUs lack. This implies RTX 50 collection GPUs can generate double the frames of previous-gen fashions in DLSS-supported video games, making them seem as much as twice as “quick” as RTX 40 collection GPUs. However in actuality, whereas FPS numbers will improve with DLSS 4, latency and gameplay really feel could not enhance as dramatically.

The declare that the RTX 5070 matches the RTX 4090 in efficiency appears doubtful. Maybe it may match in body fee with DLSS 4, however actually not in uncooked, non-DLSS efficiency.

Of their charts, Nvidia highlighted 4 video games and apps supporting DLSS 4: Cyberpunk 2077, Alan Wake 2, Black Fable Wukong, and D5 Render. Nevertheless, we’ll set these apart for now since we lack correct efficiency knowledge for these titles. The efficiency claims for the opposite video games matter extra.

The declare that the RTX 5070 matches the RTX 4090 in efficiency appears doubtful. Maybe it may match in body fee with DLSS 4, however actually not in uncooked, non-DLSS efficiency. Based mostly on Nvidia’s charts, the RTX 5070 appears 20 – 30% sooner than the RTX 4070 at 1440p.

This may place the RTX 5070 barely forward of the RTX 4070 Tremendous for about $50 much less, or alternatively, 20 – 30% sooner than the RTX 4070 for a similar worth. Relying on the way it performs relative to the RTX 4070 Tremendous, this might provide cheap, although barely underwhelming, worth. A ten% efficiency enchancment over the 4070 Tremendous and a 9% worth reduce would lead to roughly a 20% enchancment in value per body.

Nvidia made related claims for the RTX 5070 Ti, estimating a 20 – 30% efficiency enchancment over the 4070 Ti at 1440p (ignoring the deceptive DLSS 4 numbers). This may lead to a good higher cost-per-frame enchancment in comparison with the RTX 4070 Ti, possible exceeding 30%, although it is going to shrink nearer to twenty% relative to the 4070 Ti Tremendous.

The RTX 5080 exhibits an analogous story: a 20 – 30% efficiency enchancment over the RTX 4080. Given the RTX 4080 Tremendous was solely about 5% sooner than the RTX 4080, this ends in a roughly 20% improve in cost-per-frame effectivity.

As for the RTX 5090, the non-DLSS 4 numbers point out a 20 – 40% efficiency acquire over the RTX 4090. Nevertheless, because the RTX 5090 is priced 25% larger, most video games would want to attain nearer to a 50% acquire to yield a 20% enchancment in cost-per-frame.

Nvidia’s testing was performed at 4K utilizing DLSS Efficiency mode, which primarily renders at 1080p. This might result in CPU bottlenecks and underutilization of the huge RTX 5090. We suspect that would clarify the stronger efficiency features in path-traced DLSS 4 video games in comparison with conventional titles. After all, we’ll have to attend for opinions to see how that performs out.

Our intestine feeling right here and preliminary response is that Nvidia appears to be delivering round a 20% cost-per-frame enchancment, and due to this fact a 20% worth improve, throughout the board when evaluating to RTX 40 Tremendous GPUs. That’s actually higher than the RTX 40 collection, which in lots of conditions supplied completely nothing for players by way of worth. Hopefully, this obvious 20% worth improve is extra like 30% in actuality, wherein case this seems like a very good era. If the playing cards fall nearer to twenty% or, worse, decrease, will probably be a disappointment. However thus far, there is sufficient to get at the least just a little bit enthusiastic about.

A rising hole, plus different architectural enhancements

Considered one of our considerations with this era is absolutely the chasm Nvidia has created between the highest-tier mannequin and the following mannequin down. It is fairly absurd that the RTX 5090 is mainly double the {hardware} of the RTX 5080. It is nearly like we’re again within the days of twin GPUs, besides this time it is a huge single GPU die. This makes the RTX 5080 really feel like fairly a weak product compared, though there’s a substantial worth distinction to account for that.

Gone are the times of the 80 class being only a small discount in capabilities relative to the most effective mannequin. That hole is just rising, guaranteeing that players are pressured to pay a daft sum of money to entry efficiency that is materially higher than the most effective of the 40 collection.

This has flow-on results for the remainder of the product stack, which additionally turns into comparatively weak relative to the most effective GPU. Nevertheless, there’s at the least some excellent news right here as a result of, within the earlier era, the RTX 4080 had 26% extra CUDA cores than the RTX 4070 Ti. On this era, the RTX 5080 has 20% extra CUDA cores than the RTX 5070 Ti, in addition to a lot nearer reminiscence specs. So, the 5070 Ti is not reduce down as a lot relative to the 5080 when in comparison with what Nvidia did on the RTX 40 collection launch.

Nvidia’s Blackwell structure is an iteration on earlier architectures, bettering varied facets that existed earlier than, like ray tracing models and Tensor cores. Jensen devoted a ton of time to speak completely about AI – we can’t bore you with that – however mainly, Blackwell is highly effective for AI along with common processing. Shocker, proper? A brand new Nvidia GPU is being marketed as the most effective for AI but.

The structure has a tighter integration with the Tensor cores contained in the SM, improved shader execution reordering (which is twice as environment friendly), RT cores with twice the ray-triangle intersection fee, and enhanced compression to cut back reminiscence footprint. Nvidia claims this may enable for extra geometry whereas ray tracing.

Blackwell additionally improves video encoding and decoding with a brand new era of the {hardware} engine. That is the primary time we’re seeing a video decoding engine enchancment because the RTX 20 to 30 collection improve. The brand new show engine additionally permits for brand new multi-monitor configurations: as much as 4 4K 165Hz displays with out DSC, or two 4K 360Hz displays with DSC. Two 8K 100Hz displays with DSC are additionally potential.

Nvidia is releasing Founders Editions for all fashions besides the RTX 5070 Ti. This FE mannequin seems just like the RTX 4090 FE playing cards, besides now each followers are on the identical facet. Amazingly, Nvidia is utilizing only a two-slot design for the 5090 regardless of the cardboard being rated for 575W. The RTX 5080 seems to make use of a really related Founders Version design, whereas the RTX 5070 makes use of a scaled-down mannequin that is not as lengthy: 242mm lengthy as an alternative of 304mm lengthy, in response to Nvidia’s specs.

DLSS 4 next-gen upscaling tech

Nvidia additionally introduced DLSS 4, the following era of their upscaling know-how. The headlining characteristic right here is body era to generate 3 frames for each 1 actual body, as an alternative of producing 1 body for each 1 actual body.

There are vital query marks right here over latency and picture high quality. When producing each second body, 50% of the time you might be seeing an actual body. When producing 3 frames, solely 25% of the time you might be seeing an actual body. So, these generated frames had higher be superb high quality; in any other case, you will notice a ton of artifacts plenty of the time – one thing that shall be a difficulty until the standard of generated frames is considerably overhauled relative to DLSS 3.

The opposite concern is utilizing multi-frame rendering to hack our approach to high-frame-rate gaming. Nvidia confirmed a demo the place, with out DLSS 4, a sport was working at 30 FPS, however then over 240 FPS with DLSS 4 enabled. Now, after all, not even DLSS upscaling was enabled within the 30 FPS instance, however we actually do not need to see any conditions the place producing 3 frames is used to get a 30 FPS sport as much as 120 FPS – that may be completely horrific by way of enter latency.

DLSS 4 is unique to Blackwell GPUs, however just for the Multi-Body Rendering element. DLSS 4 additionally options enhancements to common body era, tremendous decision, and ray reconstruction, with the identical help for these options as we at the moment have.

The restriction on multi-frame rendering is because of what Nvidia calls “enhanced {hardware} flip metering capabilities” within the Blackwell structure. Nvidia says that is required for a clean expertise, which we hope is true; in any other case, RTX 40 collection house owners are getting screwed.

The standard enhancements to upscaling and different facets of DLSS 4 are vital, in response to Nvidia, because of the use of transformers. Nvidia claims this may “enhance picture high quality with improved temporal stability, much less ghosting, and better element in movement.”

Truthfully, we’re extra within the upscaling high quality enhancements in DLSS 4 than multi-frame rendering. Nvidia additionally says the brand new body era mannequin is 40% sooner and makes use of 30% much less VRAM.

That sounds fairly thrilling to us, and actually, we’re extra within the upscaling high quality enhancements in DLSS 4 than multi-frame rendering. Nvidia additionally says the brand new body era mannequin is 40% sooner and makes use of 30% much less VRAM. For producing a number of frames, the AI algorithm solely must run as soon as per rendered body. Optical circulation is changed with an environment friendly {hardware} mannequin, which raises questions as to why body era continues to be unique to the 40 collection, given it was {hardware} restricted because of optical circulation accelerators in Ada Lovelace.

What’s not clear is whether or not the brand new transformer mannequin for DLSS has a efficiency affect and if this efficiency affect is identical for all GPUs. For instance, they declare body era is now sooner, however is that on each RTX 50 and RTX 40 GPUs? Is there a efficiency affect from utilizing the transformer mannequin for upscaling? Many inquiries to be answered in testing.

Nvidia’s gaming software program moat

The truth that 75 video games and apps are coming at launch with DLSS 4 help is sort of spectacular, displaying Nvidia has discovered that widespread sport help is required to make options worthwhile.

We do not know if this implies there shall be vital sport updates at launch for all these titles or whether or not Nvidia has discovered some approach to mechanically add DLSS 4 to those video games. However after a fast look by the checklist, it is not each DLSS 3 sport, and a few video games are solely including DLSS 4 help later. This means a widespread effort to implement the characteristic.

As well as, Nvidia is constructing an override characteristic into the Nvidia app, which can act like a DLL swap, permitting upgrades (or overrides, as they name them) to the newest DLSS fashions for body era, tremendous decision, and ray reconstruction. It would additionally enable body era to show into multi-frame era for 50-series house owners.

Nvidia has introduced Reflex 2, which primarily brings body warp from VR gaming to common PC gaming for the primary time. The concept right here is that as an alternative of displaying the rendered body primarily based on outdated enter, Reflex 2 will take a newer enter pattern, carry out calculations on the CPU, and warp/shift the rendered body primarily based on that enter.

In a way, this “fakes” a body replace and applies it to the rendered body, lowering perceived latency for actions that shift the digital camera as a result of actual enter is being taken into consideration to warp the body by the correct quantity. It won’t cut back latency for sport actions like firing your gun, for instance, as a result of that can’t be warped.

It is a cool idea, nevertheless it stays to be seen the way it really works in video games. Will it really feel good to play a sport with Reflex 2 enabled, or will it really feel a bit floaty and bizarre if the warp calculations aren’t correct? It is arduous to say at this level, though it is good to see Nvidia claiming features for each CPU- and GPU-limited eventualities.

There’s additionally the query of picture high quality, and whereas Nvidia claims they’re utilizing inpainting to unravel points with gaps within the warped render, each single body is affected by this warp. So, if the standard of the warp is dangerous, this may make the complete sport look dangerous on a regular basis.

Reflex 2 is first coming to The Finals and Valorant quickly, with help for RTX 50-series GPUs initially and different RTX GPUs later. So, it would not sound like this shall be one thing restricted to new Blackwell fashions. Nevertheless, it is essential to notice right here that body warp in Reflex 2 won’t clear up latency issues with DLSS frame generation, particularly multi-frame era, as a result of there is no such thing as a indication will probably be built-in into video games that help DLSS 4 multi- or single-frame era. It won’t even work collectively, for all we all know.

Blackwell additionally helps RTX Neural Shaders and RTX Neural Faces, although we’re undecided how quickly we’ll really be seeing these in video games, so it appears extra like a future-looking characteristic for now.

Nvidia claims that RTX Neural Shaders can be utilized to compress textures by as much as 7x, saving reminiscence, which is probably why they’re nonetheless being stingy with VRAM this era. However after all, this characteristic is totally ineffective in video games that do not use it, so this is not an precise excuse for the low VRAM quantities on the RTX 5070 specifically.

That is every little thing Nvidia has introduced at CES 2025 for PC players. The brand new line-up of GeForce RTX 50-series graphics playing cards seems cheap at first look, nevertheless it actually did not blow us away. Efficiency features look cheap when you strip out all of the BS advertising about new options, the worth improve seems modest to okay, and the {hardware} itself is spectacular.

The exception to this may be the RTX 5070, which has simply 12GB of VRAM at $550 – too low for that class of graphics card in 2025, though Nvidia is claiming diminished reminiscence utilization by new options. Some video games already present that 12GB is inadequate even with out DLSS utilized, and since a lot of the VRAM financial savings come by DLSS 4 mannequin updates, if the bottom sport nonetheless would not work with 12GB of VRAM, it would not matter if body era is a bit more reminiscence environment friendly.

It is all the time optimistic to see that pricing is not ridiculous or outrageous. This does not look like a repeat of the RTX 40 collection, the place players have been handled to little to no improve in worth and shockingly excessive costs for some fashions. After all, we nonetheless have to see wholesome efficiency features from these fashions to get the required worth improve.

DLSS 4 sounds good on paper, and we’re trying ahead to see these claims about bettering the standard of upscaling. The flexibility to improve the DLSS mannequin in current video games by an Nvidia app characteristic can be neat as many players have been performing handbook DLL swaps.

As for multi-frame era, we’re not bought on that characteristic but given the implications for picture high quality and latency… particularly if it is used as a crutch to permit for native rendering at even decrease body charges. And we completely hate how we’re now in a state of affairs the place Nvidia can simply declare ridiculous efficiency features by producing an increasing number of frames. Relaxation assured we’ll examine this additional when we’ve the GPUs readily available and get to testing.