Nvidia's RTX 5080 is Actually an RTX 5070

The GeForce RTX 5080 is definitely an RTX 5070, or no less than it must be based mostly on its {hardware} configuration. That is the explanation the RTX 5080 is so underwhelming: it must be sitting within the 70 class at a lower cost level, delivering way more spectacular positive factors era over era. As an alternative, Nvidia has turn into grasping, and that is what we can be exploring on this article.

The rationale we are able to confidently say the RTX 5080 is definitely an RTX 5070 in all however identify is as a result of not so way back we explored the historical past of Nvidia’s GPU configurations, courting again to the GeForce 700 collection from 2013 (see our characteristic: Nvidia’s GPU Lessons By means of the Years).

We appeared on the {hardware} configurations in every class and in contrast them relative to the flagship of the period. This allowed us to create a “typical” Nvidia GPU era – a median of what we noticed over the past six generations.

Now that we all know for positive what the {hardware} of the RTX 5080 and RTX 5090 is, we are able to slot this into the comparability and see how Nvidia’s newest era is stacking up.

Spoiler alert: it does not stack up properly in any respect.

Evaluating GPU Core Configurations

For core configuration, the RTX 5090 is an impressively massive GPU with 21,760 CUDA cores, however the RTX 5080 cuts this down to simply 10,752 CUDA cores, and the RTX 5070 is even additional lower to six,144 CUDA cores. Which means the RTX 5080 packs simply 49% of the cores seen within the flagship, and the 5070 packs simply 28%.

Nvidia GPUs: Shader (CUDA) Core Depend

Traditionally, that is fairly a weak {hardware} configuration, even for the RTX 5080. Throughout the six prior generations, on common, the 80-class GPU featured 72% of the cores of the flagship. Among the higher generations, just like the RTX 30 collection and RTX 3080, this quantity was over 80%. The 40 collection was one of many weakest showings at 59%, however the RTX 5080 is even worse at simply 49%.

This truly places it beneath the relative configuration of each 70-class GPU up to now six generations, excluding the RTX 4070. It is not fairly as lower down because the RTX 4070, however the common core configuration throughout six years for a 70-class GPU is 54% of the flagship.

With the RTX 5080, it is simply 49%, so you would even make the case it is a 60 Ti-type product. It definitely would not look misplaced as a 60 Ti within the 30 collection, for instance, the place the RTX 3060 Ti had 46% of the core configuration of the RTX 3090. That stated, generally, the 5080 would often match between a 60 Ti and 70-class product.

Evaluating VRAM

Issues do not look significantly better when analyzing reminiscence bandwidth. The RTX 5080 has the most important discount in reminiscence bandwidth relative to the flagship that we have seen within the 80-class.

With the 5090 packing 1,792 GB/s of bandwidth, however the 5080 simply 960 GB/s, it has obtained simply 54% of the reminiscence bandwidth. Traditionally, that places it beneath the typical from the 70-class, although there are three generations the place the bandwidth discount is much like what we’re seeing from Blackwell.

Nvidia GPUs: Reminiscence Bandwidth (GB/s)

The RTX 5070 can also be shaping as much as be a lower-than-60-class configuration, although once more, there are some generations the place it matches the 60-class.

VRAM capability varies rather a lot between generations, and there have been some funky configurations through the years. Nevertheless, once more, the RTX 5080 does not stack up notably properly, that includes half the VRAM of the flagship mannequin. That is extra sometimes what you’d see from a 70-class card, though the 80-class does not all the time obtain considerably extra reminiscence than the 70-class.

The RTX 5070, with 12GB of VRAM, is extra in step with a 60-series product, and that aligns with our normal sentiment on the minimal stage of VRAM that is appropriate for a $550 GPU.

Nvidia GPUs: VRAM Capability (GB)

Evaluating GPU Costs

Lastly, we are able to check out pricing. Now, GPUs are getting dearer annually, with the flagship mannequin (beforehand the Titan playing cards, now the 90-class playing cards) rising from $1,000 to $2,000 within the house of 10 years. That has additionally dragged up the pricing of the opposite fashions all through the lineup, however even nonetheless, is a $1,000 RTX 5080 an appropriate worth lower relative to the flagship?

Nvidia GPUs: Launch MSRP Pricing

That is the one metric the place Blackwell is sitting extra favorably than the typical era since 2013. Typically talking, the 80-class card is priced at 60% of the flagship’s worth, whereas the RTX 5080 is 50% of the worth.

Now, there are some generations the place this aligns – the ten collection and 30 collection, for instance, come to thoughts the place the 1080 and 3080 have been roughly half the worth of the Titan X and 3090, respectively. However it’s a little higher than common. Have been the configurations additionally in step with expectations from an 80-series card, the 5080 might be priced at $1,200, however sadly, the configuration is properly beneath regular.

Equally, the RTX 5070 is definitely cheaper than the historic relative common worth of a 70-class GPU. With the flagship now costing $2,000, the 70-class GPU could be extra like $750. However once more, the 5070 being cheaper than common relative pricing is simply good if the configuration is similar because the historic common, which it is not.

Nvidia GPU Configuration Abstract

This abstract of the six-generation common Nvidia GPU configuration exhibits the problem completely. The GeForce RTX 5080 has a configuration that might usually sit round or barely beneath the extent of a 70-class GPU, but the worth is extra in step with what’s charged for a mannequin between the 70 and 80-class. In impact, this makes the RTX 5080 a barely worse-than-normal 70-class GPU in its {hardware} configuration, for which Nvidia is charging 70 Ti relative cash.

The RTX 5070 has an identical drawback. Its configuration is worse than a typical 60-class GPU, in all probability extra across the mark of a theoretical “50 Ti” mannequin, and Nvidia is charging 15% greater than typical relative 60-class pricing for it.

This mismatch between {hardware} configuration – which drives efficiency – and the worth is why most reviewers and many individuals in the neighborhood really feel Blackwell is a particularly underwhelming era.

Fixing the RTX 5080

Now, there are two methods Nvidia may have mounted this era to make it extra in step with the historical past of the GeForce collection. The primary is to regulate the configurations. For the RTX 5080, this may have required a bigger GPU die that slots between GB202 and GB203. It could require someplace round 47% extra shader cores, so as a substitute of packing 84 SMs, it could want extra like 124 SMs, so 15,872 shader cores.

It could additionally want 43% extra reminiscence bandwidth, and the simplest method to obtain that might be rising the reminiscence bus from 256-bit to 384-bit. This might additionally permit for a configuration with extra VRAM. In return, this configuration would justify the next worth level of $1,200 to be in step with historic pricing developments.

Based mostly on the configuration, this correctly configured RTX 5080 could be roughly equal to an RTX 4090 besides priced $400 cheaper, extra for those who go by road pricing for the 4090.

That also would not be a massively compelling generational uplift relative to elements just like the RTX 4080 Tremendous – within the vary of 35% quicker for 20% more cash – however quite a lot of that might be because of the improve in worth for the flagship mannequin, dragging up the remainder of the lineup.

On the identical $1,000 worth level, this legendary RTX 5080 could be fairly compelling, and we suspect that is what most individuals have been hoping for – RTX 4090 efficiency for 4080 pricing.

This might repair one of many largest points with the 5070 as we see it proper now, which is its 12GB VRAM buffer. Rising this mannequin to be extra in step with the 5080 would see the cardboard outfitted with no less than 16GB of VRAM.

For the RTX 5070, the configuration would want to considerably improve to justify 70-class branding based mostly on historic developments. The 5070 would really want a barely bigger configuration than the present RTX 5080, although once more, the worth may additionally improve to match developments.

The second, less complicated strategy to fixing the configuration concern is reducing the worth for the present fashions and doubtlessly renaming them as properly. However let’s simply go together with reducing the worth to make issues easy.

$700 or $800 RTX 5080, Anybody?

Based mostly on the {hardware} configuration of the RTX 5080, it must be priced round 35% of the flagship mannequin, which might put it at $700. That will lead to a price per body of simply $7.69 per body, a whopping 38% discount relative to the RTX 4080 Tremendous, making it a extremely compelling era. That will put it on par with, perhaps barely worse than, the fee per body uplift the RTX 3080 offered relative to the RTX 2080, among the best current generations we have had till crypto mining ruined it.

Possibly that is a bit too beneficiant for the present GPU market, so perhaps $800 is extra life like for this type of {hardware} configuration in 2025. That is fairly costly for what’s traditionally extra like 70-series {hardware}, however it could nonetheless current a 29% discount in value per body relative to the RTX 4080 Tremendous.

Is a Higher Configuration or Value Attainable?

The apparent caveat is that it is turn into dearer to supply a graphics card in 2025. Wafers at the moment are considerably dearer, reminiscence is dearer, board designs have turn into extra advanced – mainly, every little thing is dearer.

The apparent caveat that goes with all of that is that it is turn into dearer to supply a graphics card in 2025 in comparison with earlier years, particularly if we examine again to the GeForce 900 or 10 collection eras. Wafers at the moment are considerably dearer, reminiscence is dearer, board designs have turn into extra advanced – mainly, every little thing is dearer. There’s additionally normal inflation to cope with over the past decade.

That is Nvidia’s main protection when requested in regards to the rising value of graphics playing cards through the years. So the query is whether or not it is affordable for 70-class {hardware} to value considerably extra in 2025 and now be rebranded as 80-class {hardware}.

First, let’s discover die sizes. There’s some fluctuation through the years with how a lot GPU silicon is in every product, and the way lower down every mannequin is, however the development has remained comparatively related. The present RTX 5070 die, GB205, is definitely the smallest 70-class die we have seen throughout these seven generations. GB203 used within the RTX 5080 can also be smaller than the typical 80-class die, however barely bigger than the typical 70-class die.

Nvidia GPUs: Die Dimension (sq. mm)

The issue Nvidia has confronted is that TSMC wafers have turn into considerably dearer. Producing 378 sq mm of 4N silicon is drastically dearer than creating 398 sq mm of 28nm silicon, as was seen within the GeForce 900 collection. However does this actually clarify why the GTX 980 value $550, the GTX 1080 value $600, and now the RTX 5080 prices $1,000?

We did some calculations based mostly on die sizes, yield estimates – slotted into the wonderful SemiAnalysis Die Yield Calculator – plus rumored prices for numerous TSMC wafer households. We additionally set a worth for faulty dies relative to good dies and assumed that good dies could be wanted for every product.

Based mostly on this, we got here up with a price of $140 for every GB203 die utilized in an RTX 5080. To be clear, it is a tough estimate, not the factual value of the die. It is also clearly not the one element that goes into an RTX 5080 – there’s additionally the reminiscence, PCB, cooler, silicon packaging, bodily packaging, analysis and improvement prices, and so forth.

Nvidia GPUs: Die Value Tough Estimate

We additionally did the identical factor for the GP104 silicon that Nvidia used within the GTX 1080 and GTX 1070, constructed on TSMC’s 16nm course of, and adjusted for inflation.

Our tough estimate for that die is a price of $40. Which means within the almost 9 years since Pascal, we estimate the price of 70-class silicon has risen from $40 to $140, a 3.5x or almost $100 improve.

This might be on the upper finish of the dimensions based mostly on the numbers we enter into the calculation, however even with extra favorable numbers for TSMC 4N silicon, we’re nonetheless probably greater than a 2.5x value improve.

Inflation vs. {Hardware} Value vs. Greed

The GeForce GTX 1080 launched at $600 again in 2016, and the GTX 1070 launched at $380. Adjusted for inflation, these costs in the present day could be extra like $780 and $500. Then you must consider about $100 of further GPU silicon bills, plus the added value of 16GB of GDDR7 reminiscence versus 8GB of GDDR5-class reminiscence, and different related prices with a extra advanced graphics card design. This consists of needing a cooler and energy phases to deal with 360 watts, in comparison with simply 180 watts for the GTX 1080 and 150 watts for the GTX 1070.

By way of worth improve, Nvidia has raised the worth of the mannequin with the complete die – so GTX 1080 to RTX 5080 – by $220 above inflation. And for the cut-down model – so GTX 1070 to RTX 5070 Ti – we’re a $250 improve above inflation. That is ignoring for a second {that a} full GB203 configuration is extra in step with a cut-down GP104 configuration relative to the flagship.

You’ll be able to see why, from a enterprise perspective, Nvidia desires to cost $200 to $250 extra for his or her newest merchandise in comparison with almost decade-old GPUs. The price of producing graphics playing cards has gone up, and Nvidia desires to take care of or develop fats margins on every product.

What portion is attributed to every of these issues is hard to find out, particularly as a result of a few of the rise in element costs must be a part of the worth rise attributable to inflation, and different elements can be attributable to bills outpacing inflation.

It is also potential that whereas the margin as a greenback quantity has grown, the margin as a proportion has decreased – these Pascal dies would have been very low cost to supply. That is all primary enterprise although: promote extra stuff at the next margin for those who can.

From a shopper perspective, we see little motive why a relative {hardware} configuration that used to value $380 in 2016 or $500 in 2018 now completely must value $1,000 in 2025.

With all elements included, we expect it is affordable for graphics playing cards to value extra, however a doubling of the worth in lower than a decade is unreasonable. It’s not costing Nvidia $400 – 500 extra to supply an RTX 5080 now than a GTX 1070 or RTX 2070 within the years prior.

The margin on these GPUs continues to be large, even when in the present day they may not be the place Nvidia desires them to be – in any case, the margins on these datacenter AI playing cards nonetheless far outstrip GeForce GPUs.

Does Nvidia need to promote an RTX 5080 at a worth extra in step with 70-class {hardware}? Completely not. May they accomplish that in the event that they actually wished to? Yeah, and possibly fairly comfortably.

Additionally they may have made the RTX 5080 a bigger GPU configuration, which might have been extra pricey to supply, however once more, not outrageously so (except you care solely about margins, by which case a $100 – 150 value improve is devastating) – and if it led to a considerably higher product, there would have been extra worth flexibility.

We are able to look to different elements of the GPU market to see whether or not that is affordable too. The Intel Arc B580 makes use of a GPU die that’s related in measurement to the RTX 5070, constructed on an identical TSMC 5nm node, and with an identical 12GB VRAM buffer. Clearly, there are some variations – GDDR7 would improve the price of the 5070, as would the 60W increased TDP.

However Intel has priced their GPU at $250, in comparison with $550 for the RTX 5070. There have been some recommendations that the B580 is being offered for close to value worth, and there are inventory points anyway, so it is arduous to say it is truly out there at $250, however we expect you get the image. Many of the distinction in worth between the B580 and RTX 5070 is because of margin, not the {hardware} itself.

The opposite consider all of that is competitors. Nvidia is promoting avid gamers a 70-class configuration for $1,000 as a result of they don’t have any competitors. There isn’t any stress to cut back costs or lower into their margins as a result of the overwhelming majority of avid gamers purchase Nvidia GPUs.

They’re solely competing with themselves on this worth vary, and based mostly on the GeForce 40 collection versus AMD’s RDNA3, you would make a powerful case they have been competing with themselves throughout most of that product stack. What would possibly drive Nvidia to cease creating poor GPU configurations is a compelling various, however proper now, there is not one.