Review Nvidia GeForce RTX 5070 review: $549 price and performance look decent on paper

[ern88]

The RTX 5000 series smells almost like the 2000 series launch. Not a big improvement over Pascal. But new features.

 

[thestryker]

generally speaking generation by generation progress on nvidia is for a 5070 to perform like a 4080 or a 5060 to perform like a 4070.

That was thrown out with the 40 series when the 4070 cost $100 more than the 3070 and wasn't faster than the 3080. I imagine nvidia thinks we should all be grateful they didn't raise MSRP with the 50 series aside from the 5090.

 

[thestryker]

I know it's part of the standard publishing here, but I feel like omitting a star rating until 9070/XT reviews would have been the way to go. The 4070 Super would also have been a better choice than the 4070 Ti, but I suspect it'll be added later so it's not really that big a deal.

Overall this card seems like a more feature filled 4070 Super with $50 off the MSRP than anything to get excited about. I imagine the 9070 will be faster which will just make the pricing on both seem that much more stupid. Whether or not MSRP ends up in retail the price point for this performance really shouldn't be higher than $500.

So far it seems like the 5070 Ti at the hypothetical MSRP is the best choice of the 50 series. Hopefully the 9070 XT will end up being solid competition and provide something people can be excited about.

 

[JarredWaltonGPU]

First time commenting on a Tom's Hardware article. I will say, I've always thought of Tom's Hardware as a trustworthy, very thorough review site, and I make extensive use of their hierarchy charts and benchmarks for my own research and for recommending hardware to my friends and family.

But despite my very positive perception of this site's content, the exclusion of the 4070 Super from this review has me genuinely worried about the quality of journalism here. Tom's Hardware is not the only reviewer I've seen that has skipped the 4070S, but it is the only reviewer I've seen where the absence of the 4070S seems to have artificially boosted the 5070's alleged value proposition.

I understand there is a lot of work that goes into redoing hundreds of benchmarks for a new test bench setup, but the end result of skipping the 4070S on the excuse that its "not too far off the performance of the RTX 4070 Ti" is shaky at best when it results in dubious claims of "decent generational performance increase" (which is the first Pro listed on the first page of this review!!!), a claim which I have not seen any other third party reviewer support when compared to the 4070S.

It’s 20% faster than 4070, with about the same number of SMs. That’s “decent” considering everything else (like staying on TSMC 4N). And the only reason the 4070 Super isn’t in the charts is simply because of lack of time.

Would the 4070 Super radically change things? No. It’s a bit slower than the 4070 Ti, which is in the charts, and it had a $599 price. So it’s about the same performance as the 5070, lacks the new features, and cost $50 more.

There are two more reviews going up tomorrow, and I’ve been benchmarking cards almost nonstop since the start of the year – which includes retesting cards as things keep changing. There are only so many hours in a day, week, and month.

The pricing also doesn’t factor in inflation or tariffs, which isn’t going to help the situation. A $549 price today to me feels like a $449 price from two years ago. Not that my paycheck has kept pace….

 

[Jabberwocky79]

It’s 20% faster than 4070, with about the same number of SMs. That’s “decent” considering everything else (like staying on TSMC 4N). And the only reason the 4070 Super isn’t in the charts is simply because of lack of time.

Would the 4070 Super radically change things? No. It’s a bit slower than the 4070 Ti, which is in the charts, and it had a $599 price. So it’s about the same performance as the 5070, lacks the new features, and cost $50 more.

There are two more reviews going up tomorrow, and I’ve been benchmarking cards almost nonstop since the start of the year – which includes retesting cards as things keep changing. There are only so many hours in a day, week, and month.

The pricing also doesn’t factor in inflation or tariffs, which isn’t going to help the situation. A $549 price today to me feels like a $449 price from two years ago. Not that my paycheck has kept pace….

Can't wait to see the 9070 review. But based on the criticism of every 50 series review you've done so far, you'd better give the 9070s 5 stars or there will be rioting in the streets 😏

 

[palladin9479]

Really now? You know what myself and everyone else was expecting specifically? No. I can only speak for myself, but I personally expected the new gen 60 tier card being the same performance or better than the old 70 tier card at a similar cost to the old 60 tier card.

So you agree you were looking for 20~100% generational improvement for the same price or less...

 

[JarredWaltonGPU]

I know it's part of the standard publishing here, but I feel like omitting a star rating until 9070/XT reviews would have been the way to go. The 4070 Super would also have been a better choice than the 4070 Ti, but I suspect it'll be added later so it's not really that big a deal.

Overall this card seems like a more feature filled 4070 Super with $50 off the MSRP than anything to get excited about. I imagine the 9070 will be faster which will just make the pricing on both seem that much more stupid. Whether or not MSRP ends up in retail the price point for this performance really shouldn't be higher than $500.

So far it seems like the 5070 Ti at the hypothetical MSRP is the best choice of the 50 series. Hopefully the 9070 XT will end up being solid competition and provide something people can be excited about.

Just know that I already have all of the numbers for the 9070 cards. I know exactly where they’re going to land, and I fully expect the same situation with pricing and availability to apply. Maybe I’ll be wrong there, but we shall see.

Honestly, the 80%/4 star score was left in place from the RTX 5070 TI article that I cloned. LOL. I didn’t change it partially because I kind of forgot (ignored it) and partially because I’m really at a loss as to how to score cards that I don’t expect to be sold at MSRP anytime soon. Some people think you should slaughter the score, but that’s just a different form of marketing in my book.

On paper, which will hopefully become reality in the future, the 5070 is supposed to be a $549 card. I’m OK with that and it seems fairly decent overall, which to me is a 3.5 or 4-star rating. If you want, deduct half a star for every $50 above MSRP, maybe? I don’t know.

4070 Super hasn’t been retested yet, because I didn’t have time. 7900 GRE has also not been retested, due to a lack of time. For both the 5070 and the 9070 series, on some level, I view their primary competition as the prior generation $500-$600 cards. I’d like to include both the 4070 super and 7900 GRE, but that’s two solid days of testing doing nothing else. I might add those to the reviews next week once they’re finished.

 

[palladin9479]

Honestly, the 80%/4 star score was left in place from the RTX 5070 TI article that I cloned. LOL. I didn’t change it partially because I forgot, I’m partially because I’m really at a loss as to how to score cards that I don’t expect to be sold at MSRP anytime soon. Some people think you should slaughter the score, but that’s just a different form of marketing in my book.

I'd look at the MSRP situation as a reflection of how much the vendor is allocating to supply. In the case of nVidia they are, for understandable reasons, deliberately undersupplying the market. If at MSRP then score X, otherwise score X -1/2 or more. People use these reviews and scores to make buying decisions and $1000 5070 ti vs $650 9070 XT is a pretty crazy distinction to make.

 

[Gururu]

Honestly, the 80%/4 star score was left in place from the RTX 5070 TI article that I cloned. LOL. I didn’t change it partially because I forgot, I’m partially because I’m really at a loss as to how to score cards that I don’t expect to be sold at MSRP anytime soon. Some people think you should slaughter the score, but that’s just a different form of marketing in my book.

Personally, I like how you do the reviews and cannot fault the score based on the paper MSRP. If a reader can't find the card at MSRP, then they ought to wait or buy something else. I don't understand the NEED to buy it NOW. Simple as that.

 

[JarredWaltonGPU]

I'd look at the MSRP situation as a reflection of how much the vendor is allocating to supply. In the case of nVidia they are, for understandable reasons, deliberately undersupplying the market. If at MSRP then score X, otherwise score X -1/2 or more. People use these reviews and scores to make buying decisions and $1000 5070 ti vs $650 9070 XT is a pretty crazy distinction to make.

And I think the $599 9070 XT MSRP will prove to be just as false as the $749 5070 Ti price, so what then? Let’s say the 9070 XT starts selling for $899. Now it’s not as much of a “crazy distinction,” is it?

 

[Jabberwocky79]

I don't understand the NEED to buy it NOW. Simple as that.

While I agree that having the latest, most shiny toy is rarely necessary, every person's situation is different.

Take me for instance... built a brand new system in September using the budget I had, which did not include the money needed for a new enthusiast-level GPU. By the time I had the money saved for a new GPU, the 50-series release was 3 months away, so I decided to wait. But in the meantime, I've gotten everything else ready for a complete CPU/GPU water cooling loop. Now the clock is ticking on the warranty / return window for every one of those parts. My current GPU is unable to support decent performance at my new monitor's native 4K and also suffers from display port bugs related to the 1080.

My system is the equivalent of having a Lamborghini with a 4-cylinder for an engine. So, yeah, I'm getting a little impatient. 😉

 

[thestryker]

Just know that I already have all of the numbers for the 9070 cards. I know exactly where they’re going to land, and I fully expect the same situation with pricing and availability to apply. Maybe I’ll be wrong there, but we shall see.

Honestly, the 80%/4 star score was left in place from the RTX 5070 TI article that I cloned. LOL. I didn’t change it partially because I kind of forgot (ignored it) and partially because I’m really at a loss as to how to score cards that I don’t expect to be sold at MSRP anytime soon. Some people think you should slaughter the score, but that’s just a different form of marketing in my book.

Oh I know you know the results, but it just seems like it'd be better to not put it up until the direct comparison is available since they're right around the corner. I think that would just allow for slightly better context for the rating. Scoring in general seems like a waste because I don't think it's fair to destroy a product for market conditions, but at the same time they can't be ignored. That's why I generally ignore scoring and and just read the review (or at minimum the conclusions). This is also why I rarely watch YT hardware reviews because they're not allowed nuance on the platform since it's more of a snapshot in time than something that will be a long term reference.

On paper, which will hopefully become reality in the future, the 5070 is supposed to be a $549 card. I’m OK with that and it seems fairly decent overall, which to me is a 3.5 or 4-star rating. If you want, deduct half a star for every $50 above MSRP, maybe? I don’t know.

I suspect the 9070 XT is going to make that $549 price point look worse than it does today.

4070 Super hasn’t been retested yet, because I didn’t have time. 7900 GRE has also not been retested, due to a lack of time. For both the 5070 and the 9070 series, on some level, I view their primary competition as the prior generation $500-$600 cards. I’d like to include both the 4070 super and 7900 GRE, but that’s two solid days of testing doing nothing else. I might add those to the reviews next week once they’re finished.

I think they'd be valuable additions even though performance extrapolation isn't particularly hard.

 

[palladin9479]

And I think the $599 9070 XT MSRP will prove to be just as false as the $749 5070 Ti price, so what then? Let’s say the 9070 XT starts selling for $899. Now it’s not as much of a “crazy distinction,” is it?

I'm not gonna judge until it happens. It really depends on how much supply hits the shelves.

The ridiculous prices are a direct result of the 80+ percent vendor releasing almost no units in the past six to eight months.

 

[helper800]

So you agree you were looking for 20~100% generational improvement for the same price or less...

No, I never made that claim. That is a very specific claim. I was looking for a more general comparative improvement which is different from tier to tier. Where in the following did I say specifically, "looking for 20~100% generational improvement for the same price or less...?"

I can only speak for myself, but I personally expected the new gen 60 tier card being the same performance or better than the old 70 tier card at a similar cost to the old 60 tier card. The new gen 70 tier card being the same performance or better than the old 80 tier card at a similar cost to the old 70 tier card. The new gen 80 tier card being at 15-30% faster than the old 80 tier card at a similar cost to the old 80 tier card. The new gen 90 tier card being 30%+ or better performance at a similar cost to the old 90 tier card.

 

[palladin9479]

No, I never made that claim. That is a very specific claim. I was looking for a more general comparative improvement which is different from tier to tier.

You want a tier higher in performance for no increase in cost, you own words.

That is a 20-100 percent improvement.

 

[Pierce2623]

Less transistors than a 4070 is next level greed.

 

[helper800]

You want a tier higher in performance for no increase in cost, you own words.

You keep overgeneralizing what I am saying to fit the narrative of your claim that I am some reactionary poster while simultaneously telling me how I felt about the release of the product. Why hyperfixate on this one line in my reply and none of the above claims you made that I debunked?

 

[JarredWaltonGPU]

Less transistors than a 4070 is next level greed.

4070 only has "more transistors" because it uses the AD104 die that was also used and created for the RTX 4070 Ti (aka RTX 4080 12GB, LOL). Nvidia made a business decision to have:

GB202 with 192 SMs, replaces AD102 with 144 SMs
GB203 with 84 SMs, replaces AD103 with 84 SMs
GB205 with 50 SMs, replaces AD104 with 60 SMs
GB206 and GB207 are supposed to be 36 and 24 SMs, vs 36 and 20 SMs on AD106 and AD107 I think.

The Blackwell spacing looks slightly more sensible, IMO, other than the massive gulf between GB202 and GB203. There really needs to be a GB202.5 or whatever. Actually, let's see...

Ada:
AD107 at 24 SMs
AD106 at 36 SMs (50% more)
AD104 at 60 SMs (67% more)
AD103 at 84 SMs (40% more)
AD102 at 144 SMs (71% more)

Blackwell:
GB207 at 24 SMs?
GB106 at 36 SMs (50% more)
GB105 at 50 SMs (39% more)
GB103 at 84 SMs (68% more)
GB102 at 192 SMs (129% more)

So in my opinion, having steps of about 40~50 percent in core counts between each level seems about right. Both Ada and Blackwell have some much larger gaps that could potentially get filled with a different die. But the cost of creating a new die is maybe so high that it's better to aim for a 50~60 percent difference. Either way, it's obvious that Nvidia pushed WAY HARDER at the top than in the past. I mean, look at Ampere and Turing.

Ampere:
GA107 with 20 SMs
GA106 with 30 SMs (50% more)
GA104 with 48 SMs (60% more)
GA102 with 84 SMs (75% more)

Turing:
TU106 with 36 SMs
TU104 with 48 SMs (33% more)
TU102 with 72 SMs (50% more)

So Nvidia went from three chips to four chips to five chips in each family. It stretched the range of covered performance A LOT over the past two generations.

 

[HWOC]

This is fundamentally wrong, the 5090 is almost a datacenter GPU in size and power, far beyond anything they've offered before. It is ridiculously large and requires more power then the 12VHPWR connector can provide.

Just as a curiosity, in terms of die size, RTX 5090 is 750 mm2, while RTX 2080 Ti was 754 mm2, 250 W TDP back in 2018, on a 12 nm process, priced at 999 USD (equivalent to about 1,299 inflation adjusted USD in 2025). Titan V was 815 mm2, 250W TDP, on a 12nm process in 2017, priced at 2,999 USD (equivalent to about 3,999 inflation adjusted USD in 2025).

 

[nitrium]

this card performs like a 4070 super. a 5070. it has the same ray tracing performance, same raster performance ends up within 2% on fps, has the same power draw. this is basically a relaunched 4070super.

That isn't quite true. Supposedly the RTX 5070 draws 30W more power than the RTX 4070 S (250W vs 220W TBP). By the looks this is mostly due to more AI features, which might be useful in the future.

 

[smogfactory]

I don't understand how this lemon gets 4 stars. The price (especially considering the likely lack of supply will jack it up) makes no sense for a card with such lacklustre performance even when compared to last gen let alone the competition. Add in lack of memory and really this is something you would only buy if nothing else was available and you needed a card.

 

[palladin9479]

Just as a curiosity, in terms of die size, RTX 5090 is 750 mm2, while RTX 2080 Ti was 754 mm2, 250 W TDP back in 2018, on a 12 nm process, priced at 999 USD (equivalent to about 1,299 inflation adjusted USD in 2025). Titan V was 815 mm2, 250W TDP, on a 12nm process in 2017, priced at 2,999 USD (equivalent to about 3,999 inflation adjusted USD in 2025).

An H100 is 814mm2 and sells for $31,000 and consumes up to 700w.

The 2080 ti was a binned / defective Titan X. The Titan X sold for $2499.

The 2080 was 545mm2.

 

[-Fran-]

Nice data as always, Jarred. Much appreciated.

All things considered, including outside of gaming and in isolation to the rest of the lineup (past, specially), this card could be an interesting alternative at $550, or under. The compute it has on tap seems alright and a decent step up from the 4070 vanilla, but I have to agree the real comparison is the 4070S instead, since that's the "real" 4070 nVidia should've released as the vanilla part. Let's not joke here and say the "mid-refresh" was because they just felt it out of their heart to give better GPUs to the world.

In any case, this sets the floor for the 9070 siblings tomorrow.

Let's see how AMD competes.

Regards.

 

[Alvar "Miles" Udell]

Easiest pass in the world.

 

[Cooe]

🤦 Jarred's RTX 5000 reviews have singlehandedly made me lose and all respect I had for him as a reviewer. THESE CARDS ARE TERRIBLE (as literally every other reviewer makes clear) and yet he's universally giving them good to even great reviews....

I don't think you're a paid off Nvidia shill Jarred, I just think you are absolutely terrible at your job and should do something else with your life. 🤷