4070 reviews thread

[DAPUNISHER]

12GB is not only = AMD 14GBs. It's 24GBs actually, because of dlss4.

I swear to god, we will see that on the 5000 series. They will perforate every other pixel on the textures and then replace it by AI. You heard it here first. I want royalties for the idea, mr leather jacket, you hear?

I am happy that most observers say this launch is going poorly. Making me wrong about it being the new 970. If sales suck, or better yet, it costs them market share, we might start moving the right direction again.

On a related note: Tim on HUB now says DLSS is better than native in half the games he tested. I guess this is mostly due to bad AA in most games. That's where we are at; games are so badly done now, that DLSS is "the way it's meant to be played."

 

[Thunder 57]

I am happy that most observers say this launch is going poorly. Making me wrong about it being the new 970. If sales suck, or better yet, it costs them market share, we might start moving the right direction again.

On a related note: Tim on HUB now says DLSS is better than native in half the games he tested. I guess this is mostly due to bad AA in most games. That's where we are at; games are so badly done now, that DLSS is "the way it's meant to be played."

On the games I play I turn AA off (because it usually looks like poo) and if I want to use in game upscaling. Is that not an option anymore? I don't exactly play the most modern of games these days.

 

[Hitman928]

On the games I play I turn AA off (because it usually looks like poo) and if I want to use in game upscaling. Is that not an option anymore? I don't exactly play the most modern of games these days.

TAA is baked into a lot of modern games.

 

[Thunder 57]

TAA is baked into a lot of modern games.

That sounds incredibily stupid. Is there a reason behind that?

 

[VirtualLarry]

That sounds incredibily stupid. Is there a reason behind that?

Something, something, deferred rendering, had to be baked into engine, UE supports it automagically, so why not enable it.

I think that I've hit the major points. @zir_blazer ?

 

[Thunder 57]

Something, something, deferred rendering, had to be baked into engine, UE supports it automagically, so why not enable it.

I think that I've hit the major points. @zir_blazer ?

Well I know deferred rendering is why we can't have good old MSAA, but I don't see why that means forcing some crappy new AA to be always on.

EDIT

In case anyone else is interested, here is a pretty good read.

PC Graphics Settings Explained: MSAA vs FXAA vs SMAA vs TAA

TAA, SMAA, FXAA, MSAA, MLAA, or SSAA, which one should you choose? Four options and this is just anti-aliasing we’re talking about here. Modern games include a slew of graphics settings to choose from, in order to get the best performance for your hardware. Other than Anti-Aliasing (AA), you’ve...

www.hardwaretimes.com

 

[tamz_msc]

Well I know deferred rendering is why we can't have good old MSAA, but I don't see why that means forcing some crappy new AA to be always on.

Because MSAA cannot handle all forms of aliasing.

https://www.reddit.com/r/hardware/comments/jlzbuu/_/gas9dua

 

[beginner99]

The 6950 XT should be 10-15% faster in raster than the 4070 and you get more VRAM. I doubt AMD is happy about selling the 6950 XT at $599 or so but that's what they are doing.

Again, only in US. N21 parts are more or less out of stock here and the few that are are all way to expensive. eg more like >$700.

in my case the 4070 would work with my current PSU and my current case while the 6950 xt would for sure be too large (too long) and PSU would also be maybe too weak. So a short 4070 will end up being much cheaper for me. AMD needs to release N32 bases products ASAP.

 

[Kocicak]

I have been testing a new 2560x1440 monitor. 4070 in Hogwarts legacy, all ultra incl. RT, DLSS Quality, frame generation on, results in a great image quality, most of the time I see more than 80 fps in the most difficult scenes, often more than 100. I am really happy with it. BTW the game crashes occasionally with RT on, I wonder if they will fix this game ever.

What is funny that with this 1708x961 rendering resolution, I saw the game using even 10.3 GB of the card RAM. So this card probably cannot handle rendering natively 2560x1440 resolution with the highest settings in one of the most demanding games of today. But it manages to deliver nice experience with those upscaling and frame generation technologies. I am not sure what to think about it.

Reviewing today what I wrote yesterday, 1920*1080 pixels is 26% more than 1708*961. I do not know how graphic card RAM usage scales with number of pixels of the rendered image, but to me it seems that it is possible, that 4070 cannot run the most extreme game in the most extreme settings in 1920*1080 rendered resolution due to not having enough RAM for that.

Is this a problem given that this card can run this extreme game nicely at 2560x1440 using the upscaling and frame creation technology it has on board? And it is cheap compared to the top model - costs almost a third?

As I wrote in my first message - it would be very good if the cards were available with so much RAM as customers want to pay for. Is Nvidia forcing card manufacturers to make just one RAM configuration? How would customers benefit from that? I believe that there are some state regulatory bodies that protect customers from different practises of businesses that do not benefit customers.

Again, only in US. N21 parts are more or less out of stock here and the few that are are all way to expensive. eg more like >$700.

in my case the 4070 would work with my current PSU and my current case while the 6950 xt would for sure be too large (too long) and PSU would also be maybe too weak. So a short 4070 will end up being much cheaper for me. AMD needs to release N32 bases products ASAP.

The situation in my country is that 6800 XT costs the same as 4070, if it is avilable, and 6950 XT 15% more, if it is available. There is ONE model of each of these cards commonly available here.

Talking about power supplies, the old 6000 Radeons spike to over 600W:



I am not sure how anyone can suggest that 6800 XT or 6900 or 6950 XT can compete with 4070. These are old power hungry monsters. I have a new 750W power supply and I doubt it could handle the 6950XT.

I still do not understand why there is no sign of the mid range 7000 Radeons. The only rational explanation is that 7900 are really broken, really were originally meant to perform much better and have disabled parts in them, and AMD is repairing and respinning the new lower end dies.

 

[Ranulf]

12GB is not only = AMD 14GBs. It's 24GBs actually, because of dlss4.

I swear to god, we will see that on the 5000 series. They will perforate every other pixel on the textures and then replace it by AI. You heard it here first. I want royalties for the idea, mr leather jacket, you hear?

And thus...perforate the pixels is born.

 

[coercitiv]

I have a new 750W power supply and I doubt it could handle the 6950XT.

You would do your homework and pick a 6950XT card that produces 100W lower spikes, such as this Sapphire Nitro+


Meanwhile the 6950XT reference board had even lower requirements.

 

[AtenRa]

Gen to Gen , performance taken from Techpowerup at 1440p

970 vs 770 = 39% faster @ 350 USD

1070 vs 970 = 60% faster @ 375 USD

2070 vs 1070 = 39% + RT faster @ 500 USD

3070 vs 2070 = 50% faster @ 500 USD

4070 vs 3070 = 27% faster @ 600 USD


4070 is the worst xx70 card of the last ~10 years.

 

[Kocicak]

You would do your homework .... the 6950XT reference board had even lower requirements.
View attachment 79615

Well, I got the picture from the newest review available on the site, I am not sure, why they do not have current models in it. It seems that I could run the reference 6950XT with 750W powersupply.

Gen to Gen , performance taken from Techpowerup at 1440p

970 vs 770 = 39% faster @ 350 USD

1070 vs 970 = 60% faster @ 375 USD

2070 vs 1070 = 39% + RT faster @ 500 USD

3070 vs 2070 = 50% faster @ 500 USD

4070 vs 3070 = 27% faster @ 600 USD


4070 is the worst xx70 card of the last ~10 years.

You could try to adjust the prices for inflation the next time. I actually checked and 500 from 2020/21 equals 580 today in US.

If you look closely, the 2070 looks pretty bad too. If you factor in the inflation, it may have been even worse than 4070.

Remember that the designations are just labels, todays product stack and labeling may be incomparable to what existed in the past.

The momentary lack of competition from AMD can play a significant role in what 4070 is.

 

[Kocicak]

...
You could try to adjust the prices for inflation the next time. I actually checked and 500 from 2020/21 equals 580 today in US.

If you look closely, the 2070 looks pretty bad too. If you factor in the inflation, it may have been even worse than 4070.
...

Something like this:



You can see that performance per price is improving all the time. 4070 release is more than twice better than 2070 release, and twice weaker than 3070 release.

The average yearly improvevent of Perf/price from 970 to 3070 is 36%. 23% is nothing dramatic. The reason for outcry (and burning Nvidia down) would be a negative number.

I think that the only reason for whining is that Nvidia probably does not allow sale of different RAM configurations.

And the second reason is AMD, who do nothing in this market segment now.

 

[MrTeal]

It's probably a little worse than that, the way you're calculating it makes Ada seems a better value. For one, there was only a bit under 2.5 years between the 4070 and 3070 for inflation purposes, not 3. Second, inflation is pretty country specific but also really sector specific. I'm guessing you're using something like the officialdata.org calculator, but the primary drivers for inflation in the last two years in the US has been huge increases in the price of housing and food; electronics and other goods hasn't gone up at nearly the same rate.

It's not a huge swing, but the launch cost of a 3070 is probably more like $550 in April 2023, and perf/$ more like 17%.

 

[Kocicak]

Well, if you want 17% improvement in perf per dollar of 4070 compared to 3070, I can provide it by calculating weighted average of inflation unadjusted numbers with adjusted numbers with double the weight. Here you go:


Is its remarkable how quickly I could reproduce your number, how did you arrive at yours???

Even these numbers show twice better improvement compared to 2070, but three times weaker improvement than what 3070 brought.

 

[Saylick]

Something like this:

View attachment 79618

You can see that performance per price is improving all the time. 4070 release is more than twice better than 2070 release, and twice weaker than 3070 release.

The average yearly improvevent of Perf/price from 970 to 3070 is 36%. 23% is nothing dramatic. The reason for outcry (and burning Nvidia down) would be a negative number.

I think that the only reason for whining is that Nvidia probably does not allow sale of different RAM configurations.

And the second reason is AMD, who do nothing in this market segment now.

Yes, perf/$ increasing should be a given. It has to be a given or else there's little to no incentive for people to upgrade.

The amount of incentive is proportional to the magnitude of the perf/$ increase. As you noted, the 4070 has the 2nd lowest perf/$ increase in the last 5 generations, only following behind Turing, which basically offered linear gains in raster performance but tacked on an RT tax with the inclusion of RT.

One can say that the 4070 does the same, except that instead of paying an RT tax, you now pay a DLSS3 tax. That and having to deal with the potential future repercussions of a 12GB frame buffer.

 

[fleshconsumed]

Well, if you want 17% improvement in perf per dollar of 4070 compared to 3070, I can provide it by calculating weighted average of inflation unadjusted numbers with adjusted numbers with double the weight. Here you go:
View attachment 79629

Is its remarkable how quickly I could reproduce your number, how did you arrive at yours???

Even these numbers show twice better improvement compared to 2070, but three times weaker improvement than what 3070 brought.

Looking a the larger picture, nvidia has been screwing 70 series buyers for the past 5 years starting with 2070.

2070 had been abysmally poor value at release date, which pretty much everyone agreed with at the time.
3070 looked great, but only because of how bad 2070 was. However, it also got handicapped with only 8GB of VRAM which essentially relegated it to 1080p, maybe 1440p card in just 2 years.
4070 is only slightly better than the abysmal 2070, and it's also handicapped with 12GB of VRAM which will most likely become a problem in 2 years.

Another "fun" angle to look at, 2070 provided 8.7% perf/dollar improvement, 3070 53.3%, for the sake of looking at the larger picture, averaged out these two generations provided 31% per/dollar improvement, so...

970 70.5% improvement
1070 40.2% improvement
2070 31% improvement (averaged with 3070)
3070 31% improvement (averaged with 2070) and handicapped with 8GB of VRAM
4070 17.4% improvement and handicapped with 12GB of VRAM

The trend is clear. Nvidia has been screwing with gamers for a long time. I hope buyers start wising up.

 

[Thunder 57]

...As I wrote in my first message - it would be very good if the cards were available with so much RAM as customers want to pay for. Is Nvidia forcing card manufacturers to make just one RAM configuration? How would customers benefit from that? I believe that there are some state regulatory bodies that protect customers from different practises of businesses that do not benefit customers.

We don't. It's planned obsolencese. NVIDIA knows it owns the market. If they make sure cards people buy today are only useful for a fairly short time, they know people will have to buy new GPU's sooner. And since they own the market, it will almost certainly be another NVIDIA GPU.

Well, I got the picture from the newest review available on the site, I am not sure, why they do not have current models in it. It seems that I could run the reference 6950XT with 750W powersupply.



You could try to adjust the prices for inflation the next time. I actually checked and 500 from 2020/21 equals 580 today in US.

If you look closely, the 2070 looks pretty bad too. If you factor in the inflation, it may have been even worse than 4070.

Remember that the designations are just labels, todays product stack and labeling may be incomparable to what existed in the past.

The momentary lack of competition from AMD can play a significant role in what 4070 is.

I don't believe in the "inflation defense" game. In the CPU market we have healthy competition and prices are not out of control. NVIDIA essentially owns the GPU market and can get away with a lot. It is little different than Intel during the Bulldozer days.

You want four cores? We can do that. With HT? That'll be about $100 extra.
You want more than four cores? Hope you brought a kidney.

 

[jpiniero]

I don't believe in the "inflation defense" game. In the CPU market we have healthy competition and prices are not out of control.

There's a reason why AMD's client business is no longer profitable.

 

[kondziowy]

You would do your homework and pick a 6950XT card that produces 100W lower spikes, such as this Sapphire Nitro+
View attachment 79614

Meanwhile the 6950XT reference board had even lower requirements.
View attachment 79615

I have a new 750W power supply and I doubt it could handle the 6950XT.

I giggled a little bit here 😉 I was running Vega 64 Nitro for 5 years on 650W power supply, and never had any issues. I was running it stock usually on quiet mode (330W from the wall), or balanced (~360W from the wall). But even if I was doing OC I had no problems ~500W from the wall. Judging from those graphs, 6950XT draws less than Vega. So... yeah... I wouldn't even change my PSU for that "power hungry monster" card at all.

 

[Thunder 57]

There's a reason why AMD's client business is no longer profitable.

For what Quarter(s)? There could be legitimate reasons, including a very slow uptake with Zen 4. I very much doubt anyone is worried about it.

 

[jpiniero]

For what Quarter(s)? There could be legitimate reasons, including a very slow uptake with Zen 4. I very much doubt anyone is worried about it.

A couple quarters now. Main reason I think is because they can't get prices high enough to cope with inflation/TSMC's price hikes. To get profitable, either they are going to have to use cheaper foundries... or do what nVidia is doing and slash die sizes. They seem to be doing the later, which is why Phoenix 2 is so important.

 

[Rigg]

I giggled a little bit here 😉 I was running Vega 64 Nitro for 5 years on 650W power supply, and never had any issues. I was running it stock usually on quiet mode (330W from the wall), or balanced (~360W from the wall). But even if I was doing OC I had no problems ~500W from the wall. Judging from those graphs, 6950XT draws less than Vega. So... yeah... I wouldn't even change my PSU for that "power hungry monster" card at all.

I've been running an Asus TUF GAMING OC Radeon RX 6900 XT with a Fractal Design Ion+ 760P 760 W 80+ Platinum with zero issues since mid January. Large transient spikes shouldn't be an issue for a properly designed power supply.

 

[Thunder 57]

A couple quarters now. Main reason I think is because they can't get prices high enough to cope with inflation/TSMC's price hikes. To get profitable, either they are going to have to use cheaper foundries... or do what nVidia is doing and slash die sizes. They seem to be doing the later, which is why Phoenix 2 is so important.

It would be nice to see Samsung offer so fab competition.