NV Super refresh reviews

[BFG10K]

I'd expect most rational people upgrade when they have games they want to play, that their current HW can't adequately handle, and economic conditions are favorable for them.

All of these apply to me and I'm not upgrading. Give me a 2070 killer with 12GB and <=130W @ $199 and then we'll talk. If it takes 5+ years, that's just fine with me. I have hundreds of older games I can play, so I never need to buy a new game again.

If NV gets really bad, I'll switch to AMD APUs for my old games. Then neither NV or new games will get a cent out of me. Once you stop kowtowing to "argh-tee" and "aye-eye" and remember that entertainment is what gaming's actually about, all these first-world problems simply melt away.

I'm currently replaying Fear 1 from 2005 and it's still a lot of fun, with a trivial load on my 2070. The fans don't even turn on, which is yet another benefit of older games.

 

[MrTeal]

4070 Super is +2GB VRAM and about 5%/10% better performance at 2160/1440p compared to the RTX 3080 for $100 lower MSRP. That was 1217 days ago. Progress but at a snail's pace.

CPI rose from 262.35 to 308.85 or about 17% over that time period. TDP reduce from 320W to 220W too. But who cares.

You're just using the wrong settings. Turn on frame generation, the 4070 Super is twice as fast as the 3080. It's better than two 3090 Ties in SLI; that's a $4000 value for just $600.

 

[DavidC1]

Most sales dropped significantly after the forced lockdowns were up.

Don't you think the overall computing market drop had anything to do with that? The sales increased abnormally during that period, so we're now experiencing the opposite. How did AMD fare compared to previous generations? Isn't Nvidia at one of the highest marketshare in the history of the company? AMD barely at 10%?

Saying Super is failing without taking that into account is just bias, that's all.

However, they don't break out laptops and PCs in this, so its hard know if that is really the comeback of the tower or just more laptops.

You could compare the data for Intel. They reported bigger drops on Notebook than Desktops.

 

[beginner99]

MLID claims that sales are really bad, especially for the over-MSRP cards. He suggests that Nvidia solved the wrong problem, by releasing a more powerful card with still just 12 GB, which makes sense to me.

Indeed but the failure was made all back in the design phase

f they increased the amount of vRAM that would be "admitting" that they launched cards with too little.

For the 4070 TI Super they did increase it? And because they could because the chips does have a 256-bit bus. Therefore that really the best super card of them all with an actual relevant gain.

The 4070 however uses a different chips with just a 192 bit bus so no way to increase the ram. This decision was made during design phase and no way to correct it until next-gen.

 

[igor_kavinski]

This decision was made during design phase and no way to correct it until next-gen.

I assure you that they were being very charitable when they made those decisions.

"We will bless them with FOUR GIGABYTES(!!!!) of extra VRAM over the 3070/3070 Ti!" and then hugged and patted each other on the backs all around and went back to generating their AI manga.

So there is hope for the 5070 having 16GB. If not, well, another gen to skip.

 

[Aapje]

How did AMD fare compared to previous generations?

AMD failed with the design of this generation and despite this they are doing relatively well, probably in large part due to giving their cards more VRAM.

Also keep in mind that the GPU market is really a duopoly, which means that if one of the companies falters the competitive pressure is gone immediately on the other company.

Indeed but the failure was made all back in the design phase

It was a business decision to put lower tier chips in their cards, compared to previous generations.

I also strongly suspect that Nvidia signed a horrible contract with TSMC, just before the mining boom ended, so they are paying way too much. They should sign a new contract for the 5000-series, hopefully with better prices, that we can then benefit from.

You can't blame the designers for the TSMC contract.

The 4070 however uses a different chips with just a 192 bit bus so no way to increase the ram. This decision was made during design phase and no way to correct it until next-gen.

The choice which chip to put on what cards was not made during the design of the chips, but is purely a business decision made by leather jacket guy.

 

[igor_kavinski]

is purely a business decision made by leather jacket guy.

Yup. He should've put the 4060 chip in a 4050 but ego got in the way.

 

[jpiniero]

NVIDIA GeForce RTX 4070Ti SUPER is 8% faster on average than RTX 4070Ti in 3DMark tests - VideoCardz.com

NVIDIA RTX 4070 Ti SUPER shows 8% higher performance on average in 3DMark tests While the RTX 4070 SUPER showed a large increase in performance compared to the RTX 4070, the RTX 4070 Ti SUPER appears to offer a smaller improvement. Reviewers are currently conducting tests on the RTX 4070 Ti...

videocardz.com

Seems the 4070 Ti Super isn't going to be that close to the 4080 NS. Didn't realize they had cut the L2 to 48 MB (same as 4070 Ti).

 

[igor_kavinski]

You people took Jensen for a fool! Cunning, greedy fat fox he is, I knew all along! 😀

 

[Aapje]

Seems the 4070 Ti Super isn't going to be that close to the 4080 NS. Didn't realize they had cut the L2 to 48 MB (same as 4070 Ti).

I expected more due to the wider bus. Well, I guess that it was a good choice to get a 6700 XT and skip at least this gen.

 

[TESKATLIPOKA]

4070 Super is +2GB VRAM and about 5%/10% better performance at 2160/1440p compared to the RTX 3080 for $100 lower MSRP. That was 1217 days ago. Progress but at a snail's pace.

CPI rose from 262.35 to 308.85 or about 17% over that time period. TDP reduce from 320W to 220W too. But who cares.

RTX 3080 doesn't support Frame generation. 😉

 

[psolord]

RTX 3080 doesn't support Frame generation. 😉

Fake news, it supports fsr3+fg! xD

 

[TESKATLIPOKA]

Fake news, it supports fsr3+fg! xD

But that's from AMD.
I still find It funny how Ampere(Turing, etc) owners have to use AMD's technology to extend their card's usefulness.

 

[beginner99]

You can't blame the designers for the TSMC contract.

In the end it's NVIDIAs fault and they are to blame either for the design or being bad at negotiation. After all if what you say is true, they only were in that spot because they wanted to cheap out with Samsung. process on 3000-series where we saw the exact same lack of VRAM on a cheap process.

 

[Aapje]

In the end it's NVIDIAs fault and they are to blame either for the design or being bad at negotiation. After all if what you say is true, they only were in that spot because they wanted to cheap out with Samsung. process on 3000-series where we saw the exact same lack of VRAM on a cheap process.

I think that's a little simplistic. AMD obviously has an intimate relationship with TSMC and thus gets better deals, also because they do technology transfers to TSMC. But the amount of wafers that Nvidia bought during a shortage and the timing just before the end of Ethereum mining was always going to result in high prices.

I don't even blame Nvidia/Jensen that much because the timing of the proof of stake transition was bad for him and for an outsider it was very hard to judge whether ethereum mining was going to end overnight (which it did) or more slowly. He had to sign the contract in time to be able to launch Ada on schedule. He couldn't just wait indefinitely.

Signing a smaller contract would obviously have been wiser in hindsight, but the auto makers suffered heavily from canceling their contracts with the chip makers, to the point where they couldn't make the cars due to a lack of chips.

But as a smart consumer you can try to let other people make unwise purchases. My guess is that at the very least we'll get a more normal price/performance increase with Blackwell, especially if RDNA4 is any good. Although I'm rather uncertain about AMD. Their good showing with RDNA2 seemed to be in large part due to a node advantage. So they still seem to have a lot of catching up to do and I'm unsure to what extent RDNA3 was a weak design or whether their architecture in general is not that great.

The question marks are also huge for Intel. They are way behind AMD, but also obviously can be expected to learn fast. So will Battlemage be any good? I have my doubts, but on the other hand they should be willing to accept very low or zero margins.

 

[KompuKare]

I don't even blame Nvidia/Jensen that much because the timing of the proof of stake transition was bad for him and for an outsider it was very hard to judge whether ethereum mining was going to end overnight (which it did) or more slowly. He had to sign the contract in time to be able to launch Ada on schedule. He couldn't just wait indefinitely.

Surely trying to play off Samsung versus TSMC was a (mostly failed) strategy which only the CEO could have tried?

As it happens, AMD didn't have the volume or inclination but RDNA2 versus Ampere could have turned out very differently if AMD had the volume and desire to take a bit less margin.

 

[MrTeal]

Wow, Ti Super is really poor. 10% more shaders and 42% more BW gives <10% improvement at 4k, and like <5% at 1080p?
That's really rough.

Should qualify that the 16GB is definitely better for the price. I'm just really surprised that the performance is so close to the vanilla 4070 Ti.

 

[jpiniero]

AMD failed with the design of this generation and despite this they are doing relatively well, probably in large part due to giving their cards more VRAM.

Doesn't seem like RDNA3 is selling all that well... plus they are basically wiped out in mobile. The 7900 XTX might be the one exception cuz AI hype.

RDNA2 is selling better.

 

[SolidQ]

RDNA3 is selling all that well

Say to 7800XT

 

[blckgrffn]

Wow, Ti Super is really poor. 10% more shaders and 42% more BW gives <10% improvement at 4k, and like <5% at 1080p?
That's really rough.

Should qualify that the 16GB is definitely better for the price. I'm just really surprised that the performance is so close to the vanilla 4070 Ti.

TFLOPS are a 10% improvement and all the other hardware - tensor cores and RT cores - seems like these are going to be really workload dependent and are in the same ~10% increase. So I guess the cache is working well and most loads are currently core dependent. All the performance uplift that was more than 10% was bandwidth and vram amount related.

For my part, I had hoped it would land closer to the 4080 too, but I guess the 4080 Super at $1k might be the true range topper for mere mortals.

 

[Aapje]

Surely trying to play off Samsung versus TSMC was a (mostly failed) strategy which only the CEO could have tried?

How so? Supposedly Nvidia got a great deal from Samsung with the chip prices. The low yields meant that they had to sell a lot of GA102 chips as 3080's, but with the mining boom the profits on those cards were huge and they sold all they could make. It was easier for them to increase production than for AMD, since few companies had designs for Samsung N8, so Nvidia profited more from the mining boom than AMD.

Then just as the AI boom took off and they benefited from having very fast chips that they can sell for many times the cost of production, they switched back to TSMC. The high costs and bad sales for the gaming cards are a minor issue for Nvidia, who are getting richer than rich from the data center cards.

If it is failing when the stock price of a company triples in a year, then what is success?

As it happens, AMD didn't have the volume or inclination but RDNA2 versus Ampere could have turned out very differently if AMD had the volume and desire to take a bit less margin.

RDNA2 vs Ampere was all about the mining boom and thus margins didn't matter at all. All cards sold for inflated prices for most of these cards lives and thus market share was all about production volume.

You can just as easily turn your argument around where Nvidia could also have bought more volume (perhaps even making chips both with TSMC and with Samsung) and could have taken even more market share from AMD. But neither company predicted the immense shortages during the mining boom.

 

[GodisanAtheist]

Yeah, a startlingly poor showing from the TI Super.

For most purposes, you're getting a 4070Ti with some extra ram.

4070 Super looks like it will be the star of refresh-o-rama. Don't have high hopes for the 4080 Super (Guessing it will be within 1-2% of the 7900XTX)...

 

[DamZe]

Anytime nVidia does a "Super" refresh I am reminded of how bad that generation is and how it forces nVidia to put lipstick on a pig to generate some hype/sales.

The only saving grace is that now nVidia buyers can get 256-bit card starting at $800, LMAO!

What a time to be alive.

 

[TESKATLIPOKA]

RTX 4070Ti Super is super fast.😀
I see 1000% increase in FPS, certainly worth the money.

Yet you all are disappointed. 😛

 

[jpiniero]

Then just as the AI boom took off and they benefited from having very fast chips that they can sell for many times the cost of production, they switched back to TSMC. The high costs and bad sales for the gaming cards are a minor issue for Nvidia, who are getting richer than rich from the data center cards.

Except Gaming revenue has been going back up...