NVIDIA GeForce RTX 4070 Ti Super GPU Review & Benchmarks: Power Efficiency & Gaming
video description
Date: 2024-01-23
Related videos
Comments and reviews: 20
viniciusmr
Any chance we could see something like GPU efficiency at a locked FPS, like you guys did in the The Intel Problem: CPU Efficiency & Power Consumption video
Not sure if many people would care about this, but that's what I find more interesting, considering I don't care for 'wasteful' FPS above what the display is going to be showing. That's just:
- more heat in the room (and summer's already hot enough where I live without any help)
- more fan noise
- more expensive $$ electric bill
Given that for higher FPS we need higher boost frequencies and those don't scale anywhere near linearly regarding power consumption X work delivered, I think those uncapped FPS efficiencies may differ significantly from the efficiency the card would deliver at what I would call reasonable playing conditions.
Like, even if I purchase a 4090, I would still play a whole lot of games that wouldn't get even close to maxing out their capabilities (unless I wanted to waste money on extra frames to be discarted before they even get to the monitor), so most of the time it would be running at a different scale of efficiency than it's maxxed out one.
I realize though that would mean even more work on your already really well made and thought out reviews, for perhaps a not-so-exceptional return in terms of data, so I'm not keeping high hopes, but one can dream, right
reply
Any chance we could see something like GPU efficiency at a locked FPS, like you guys did in the The Intel Problem: CPU Efficiency & Power Consumption video
Not sure if many people would care about this, but that's what I find more interesting, considering I don't care for 'wasteful' FPS above what the display is going to be showing. That's just:
- more heat in the room (and summer's already hot enough where I live without any help)
- more fan noise
- more expensive $$ electric bill
Given that for higher FPS we need higher boost frequencies and those don't scale anywhere near linearly regarding power consumption X work delivered, I think those uncapped FPS efficiencies may differ significantly from the efficiency the card would deliver at what I would call reasonable playing conditions.
Like, even if I purchase a 4090, I would still play a whole lot of games that wouldn't get even close to maxing out their capabilities (unless I wanted to waste money on extra frames to be discarted before they even get to the monitor), so most of the time it would be running at a different scale of efficiency than it's maxxed out one.
I realize though that would mean even more work on your already really well made and thought out reviews, for perhaps a not-so-exceptional return in terms of data, so I'm not keeping high hopes, but one can dream, right
reply
alexatkin
I don't think this IS comparable to the IO die base load, as if you reduce clocks on the 4090 it can become far more efficient, especially for compute. Also the Ada cards idle power consumption can get really low, which is the big thing AMD chiplet CPUs cannot do due to the SoC consumption.
For example when running Folding at Home I limited my 4090 FE power until it was around the same as my 4070 Ti (for the same project work unit, not the cards TDP). The 4090 is about 66% faster at the same power consumption in this scenario.
Of course the 4070 Ti can also be improved by reducing it clock rate too, Ada seems to be most efficient around 2.6Ghz. I'd love to see a more scientific test of this to see exactly where the sweet-spot is.
reply
I don't think this IS comparable to the IO die base load, as if you reduce clocks on the 4090 it can become far more efficient, especially for compute. Also the Ada cards idle power consumption can get really low, which is the big thing AMD chiplet CPUs cannot do due to the SoC consumption.
For example when running Folding at Home I limited my 4090 FE power until it was around the same as my 4070 Ti (for the same project work unit, not the cards TDP). The 4090 is about 66% faster at the same power consumption in this scenario.
Of course the 4070 Ti can also be improved by reducing it clock rate too, Ada seems to be most efficient around 2.6Ghz. I'd love to see a more scientific test of this to see exactly where the sweet-spot is.
reply
SquintyGears
I find it pretty weird that you suggest that someone would upgrade from a 3060 to this card( a 4070TiS). Not that it can't happen, but I think unless someone's life situation changes significantly their reasons for choosing a particular market segment doesn't change. So this seems like a pretty steep upsell to suggest...
I do realize that it's probably part of your processes to talk about the comparison 1 tier up and down and a quick mention of the halo product sprinkled in. So this particular instance has mostly nvidia's naming to blame... But I think you should catch this case... remember the 4070Ti was going to be called 4080, if we consider Ti as nvidia does, than it's 100% it's own tier. so ti/-1 is the comparison not XX70/-1.
reply
I find it pretty weird that you suggest that someone would upgrade from a 3060 to this card( a 4070TiS). Not that it can't happen, but I think unless someone's life situation changes significantly their reasons for choosing a particular market segment doesn't change. So this seems like a pretty steep upsell to suggest...
I do realize that it's probably part of your processes to talk about the comparison 1 tier up and down and a quick mention of the halo product sprinkled in. So this particular instance has mostly nvidia's naming to blame... But I think you should catch this case... remember the 4070Ti was going to be called 4080, if we consider Ti as nvidia does, than it's 100% it's own tier. so ti/-1 is the comparison not XX70/-1.
reply
Mopantsu
Based upon specs I think I can safely say that most people were expecting more. It should have at least been as fast as the 7900XT in rasterisation performance. It would seem that Nvidia are using their worst AD103 silicon and much like they are trying to upsell you from the 4070 Super using more VRAM on the Ti Super model they are trying to upsell anyone considering the Ti Super to the 4080 Super. The 4070 Ti Super should have been MUCH closer to the the 4080 in terms of performance. No wonder AMD won't cut the official price of the 7900XT to less than $749. They probably knew what the performance would be in advance.
reply
Based upon specs I think I can safely say that most people were expecting more. It should have at least been as fast as the 7900XT in rasterisation performance. It would seem that Nvidia are using their worst AD103 silicon and much like they are trying to upsell you from the 4070 Super using more VRAM on the Ti Super model they are trying to upsell anyone considering the Ti Super to the 4080 Super. The 4070 Ti Super should have been MUCH closer to the the 4080 in terms of performance. No wonder AMD won't cut the official price of the 7900XT to less than $749. They probably knew what the performance would be in advance.
reply
Woodzta
My issue is just how obvious it is that they're keeping so much in their back pocket these days. When they released the GTX Titan, everyone knew they threw everything at it, and you were overpaying for the overpowered beast for that period of time. A few months later they used the GK110 and a GTX 780 was a reasonable, cheaper alternative. The 4090 AD102 came out in 2022 and now 16 months later this generation is still drip feeding us scraps from the AD104 for well over half the MSRP of the original 4090.
reply
My issue is just how obvious it is that they're keeping so much in their back pocket these days. When they released the GTX Titan, everyone knew they threw everything at it, and you were overpaying for the overpowered beast for that period of time. A few months later they used the GK110 and a GTX 780 was a reasonable, cheaper alternative. The 4090 AD102 came out in 2022 and now 16 months later this generation is still drip feeding us scraps from the AD104 for well over half the MSRP of the original 4090.
reply
brianm.595
Its a weird age when consoles easily have better video specs than a lot of the cards you can buy. Just for reference I have like a 10 year old evga 1070ti SC and it has 8gb of vram. How are there cards coming out still with 8gb or less. The consoles theoretically can have 16gb of vram available. Its problematic when games are made for console because they are looking for more vram than a lot of cards have.
reply
Its a weird age when consoles easily have better video specs than a lot of the cards you can buy. Just for reference I have like a 10 year old evga 1070ti SC and it has 8gb of vram. How are there cards coming out still with 8gb or less. The consoles theoretically can have 16gb of vram available. Its problematic when games are made for console because they are looking for more vram than a lot of cards have.
reply
SLiX-thatguy11m-
Was considering the 4070Ti with possible price drop based on the 4070Ti release, but with these benchmarks, I'm considering the cheaper 3080Ti instead given they seem so similar. The 4070Ti review last year didn't include the 3080Ti and it somehow kept up with the 3090, but now it seems it's on par with the 3080Ti in some games Also this means the 3080Ti surely is more value than the 4070 Super.
reply
Was considering the 4070Ti with possible price drop based on the 4070Ti release, but with these benchmarks, I'm considering the cheaper 3080Ti instead given they seem so similar. The 4070Ti review last year didn't include the 3080Ti and it somehow kept up with the 3090, but now it seems it's on par with the 3080Ti in some games Also this means the 3080Ti surely is more value than the 4070 Super.
reply
bront_catherder
GN, in the 4K RT testing, the 4070 had better 1% and lows than the 4070 Super, by a significant margin and didn't suffer the 1% and .1% issues the 4070 TI or 4070 Super did. I'd be interested in you digging into that again or double checking the results. Seems off, given memory was likely the cause of the 2 better cards having issues and the 4070 normal should have the same memory issues.
reply
GN, in the 4K RT testing, the 4070 had better 1% and lows than the 4070 Super, by a significant margin and didn't suffer the 1% and .1% issues the 4070 TI or 4070 Super did. I'd be interested in you digging into that again or double checking the results. Seems off, given memory was likely the cause of the 2 better cards having issues and the 4070 normal should have the same memory issues.
reply
wencosa
So 4060 = 4050
4070 = 4060
4070 Ti Super = 4070
4080 Super = 4080
What a waste of a generation completely flipping everything they worked on with greed and confusion for everyone. :) My bad, that makes too much sense to have everything a stack cheaper and having a 50 class with 8GB, 60 class with 12 and 70 with 16. Silly me not making millions in profit.
reply
So 4060 = 4050
4070 = 4060
4070 Ti Super = 4070
4080 Super = 4080
What a waste of a generation completely flipping everything they worked on with greed and confusion for everyone. :) My bad, that makes too much sense to have everything a stack cheaper and having a 50 class with 8GB, 60 class with 12 and 70 with 16. Silly me not making millions in profit.
reply
59.9fps
A suggestion on the power efficiency charts. Since most of the audience here are typical gamers/ enthusiasts, a more practical metric could be power consumption per 60 Frames. So 0.27 FPS/ Watt -> 3.68 Watts/ Frame -> 222.2 Watts/ 60 FPS. This makes the numbers more relatable rather than a theoretical metric which needs to be normalized by the viewer.
reply
A suggestion on the power efficiency charts. Since most of the audience here are typical gamers/ enthusiasts, a more practical metric could be power consumption per 60 Frames. So 0.27 FPS/ Watt -> 3.68 Watts/ Frame -> 222.2 Watts/ 60 FPS. This makes the numbers more relatable rather than a theoretical metric which needs to be normalized by the viewer.
reply
Jwasin3_1
I can’t wait to see what the 4000 series can do with Hellblade 2, Death Stranding 2, and GTA 6 at 1440p and 4K . Hard to push to Alan Wake 2 at a good 120fps at 4K and on ray tracing it was 60fps and lower . I want to see the 2024 -2025 games tested not the games available for PS 4 . Gamer nexus broke it down so far.
reply
I can’t wait to see what the 4000 series can do with Hellblade 2, Death Stranding 2, and GTA 6 at 1440p and 4K . Hard to push to Alan Wake 2 at a good 120fps at 4K and on ray tracing it was 60fps and lower . I want to see the 2024 -2025 games tested not the games available for PS 4 . Gamer nexus broke it down so far.
reply
MeGaDwarf2008
I'd like to see a summary chart similar to 7900 XT to 4070 TI Super, but for the all 3 new cards, in the final review.
4070 Super -> 4070 Super Ti -> 4080 Super. Should make the value more evident for price-comparison shopping, especially later on when prices start to fluctuate due to promos and all that.
reply
I'd like to see a summary chart similar to 7900 XT to 4070 TI Super, but for the all 3 new cards, in the final review.
4070 Super -> 4070 Super Ti -> 4080 Super. Should make the value more evident for price-comparison shopping, especially later on when prices start to fluctuate due to promos and all that.
reply
animateclay
When it comes to using graphics cards for AI purposes (language models and image generation), I think showing how cards perform this way will be an important metric in the near future. There are a lot of people using Stable Diffusion already, it's becoming less of a niche and more mainstream nowadays.
reply
When it comes to using graphics cards for AI purposes (language models and image generation), I think showing how cards perform this way will be an important metric in the near future. There are a lot of people using Stable Diffusion already, it's becoming less of a niche and more mainstream nowadays.
reply
todorkolev7565
efficiency is a waste of time, you need to be playing for 10 hours every single G-D day to make a significant difference at the end of the year... Most gamers will save maaaybe $5 from switching to a more efficient card and would have probably spent a lot more money upfront...
reply
efficiency is a waste of time, you need to be playing for 10 hours every single G-D day to make a significant difference at the end of the year... Most gamers will save maaaybe $5 from switching to a more efficient card and would have probably spent a lot more money upfront...
reply
danielcobia7818
Of course, the irritating thing about this launch is that this card is in the 2080 Ti price class and not any previous XX70 series card. I suspect there are very few people that have purchased XX70 series cards in the passed will be able to stomach the much higher price.
reply
Of course, the irritating thing about this launch is that this card is in the 2080 Ti price class and not any previous XX70 series card. I suspect there are very few people that have purchased XX70 series cards in the passed will be able to stomach the much higher price.
reply
gamersnexus
So... Unless the 4080 Super ends up cheaper than the 7900xtx (or better performance per dollar) I should basically buy an AMD card if I am not buying the 4090. That is, disappointing. But at least the Super cards seem better val... it's a lack of price drop thing isn't it.
reply
So... Unless the 4080 Super ends up cheaper than the 7900xtx (or better performance per dollar) I should basically buy an AMD card if I am not buying the 4090. That is, disappointing. But at least the Super cards seem better val... it's a lack of price drop thing isn't it.
reply
StormagedonBOD
I'm really looking forward to efficiency testing for lower end cards. In most of Europe most cards are way more expensive than in the US and the power's more expensive too, so seeing how cards stack up in power efficiency should make buying a cheaper card easier.
reply
I'm really looking forward to efficiency testing for lower end cards. In most of Europe most cards are way more expensive than in the US and the power's more expensive too, so seeing how cards stack up in power efficiency should make buying a cheaper card easier.
reply
GiGaSzS
Hmm, the marketed TDP would suggest Nvidia 4070 SUPER to blow out 7800 XT, but they actually offer very similar efficiency XD
I would suggest you use FPS/kW, so you will not have decimal points and since we are paying for kWh it is more intuitive ;)
reply
Hmm, the marketed TDP would suggest Nvidia 4070 SUPER to blow out 7800 XT, but they actually offer very similar efficiency XD
I would suggest you use FPS/kW, so you will not have decimal points and since we are paying for kWh it is more intuitive ;)
reply
olnnn
For efficiency tests, please look at total system power usage and not just the usage of the cards themselves as there are differences in how much driver/system scheduling overhead AMD, nvidia and intel cards have so that also plays in to power usage.
reply
For efficiency tests, please look at total system power usage and not just the usage of the cards themselves as there are differences in how much driver/system scheduling overhead AMD, nvidia and intel cards have so that also plays in to power usage.
reply
jassco2
Did they only enable 48MB of cache 103 die has 64MB. If so this probably creates more of the gap shown to the 4080. Also need to update specs video with pin. Nvidia corrected L2 on 4070S from 36MB to 48MB. This was a sneaky move by NVidia if true.
reply
Did they only enable 48MB of cache 103 die has 64MB. If so this probably creates more of the gap shown to the 4080. Also need to update specs video with pin. Nvidia corrected L2 on 4070S from 36MB to 48MB. This was a sneaky move by NVidia if true.
reply
Add a review, comment
Other channel videos
