Sneaky Downgrades: Nvidia RTX 4070 GDDR6 Review

Was the clockspeed identical between the cards?

I mean, watt usage seems to be similar, but GDDR6 uses less power than GDDR6X. Maybe they increased core clock to make up for the slower VRAM.
 
Alright, FIRST! Let's take bets, will TechSpot be accused of being Nvidia or AMD shills for this article? Place your bets below!

Frankly, it's on the consumer to refuse to buy this and create an uproar on social media, that is the only way to hold companies accountable. No regulating body will ever care about this market.
 
Alright, FIRST! Let's take bets, will TechSpot be accused of being Nvidia or AMD shills for this article? Place your bets below!

Frankly, its on the consumer to refuse to buy this and create an uproar on social media, that is the only way to hold companies accountable. No regulating body will ever care about this market.
I hope not since AMD was caught doing shady stuff tons of times as well. Gimped PCie speeds etc. 6500XT is among the worst GPUs ever reviewed on this site for example.
 
I hope not since AMD was caught doing shady stuff tons of times as well. Gimped PCie speeds etc. 6500XT is among the worst GPUs ever reviewed on this site for example.
Meh it always happens. When you actually review things that triggers fanbois into calling you a shill. TechSpot has been accused of shilling for every side at one point or another. That's how you know they're actually reviewing things instead of spouting BS. Similar accusations were thrown at Anandtech.
 
Alright, FIRST!

lol

Frankly, its on the consumer to refuse to buy this and create an uproar on social media, that is the only way to hold companies accountable. No regulating body will ever care about this market.

I mean the 4070 is also going to be EoL soon. It would make more sense if they just released the 4070, but imo this is just another way for nVidia to stick it to the cost-conscious consumer who dares to wait for a price drop instead of pre-ordering.

(It’s also possible that G6X supplies are running low, and it doesn’t look like nVidia is gonna keep using G6X for Blackwell)
 
Alright, FIRST! Let's take bets, will TechSpot be accused of being Nvidia or AMD shills for this article? Place your bets below!

Frankly, its on the consumer to refuse to buy this and create an uproar on social media, that is the only way to hold companies accountable. No regulating body will ever care about this market.
Not only will they be accused of being AMD shills, people will also assign them a political affiliation. I think it'll make even less sense than left or right wings. I'm thinking they will be called up-ists against capitalism and someone will throw in something about how this is really Obama's fault.

Thanks Obama for putting GDDR6 on the 4070. If only Elon Musk hadn't sold Taco Bell to the West Koreans we wouldn't be in this situation.
 
lol



I mean the 4070 is also going to be EoL soon. It would make more sense if they just released the 4070, but imo this is just another way for nVidia to stick it to the cost-conscious consumer who dares to wait for a price drop instead of pre-ordering.

(It’s also possible that G6X supplies are running low, and it doesn’t look like nVidia is gonna keep using G6X for Blackwell)
4070 is not going EoL "soon"

5090 and 5080 is Q1 next year.

5070 is Q2 slash Summer and I don't think Nvidia are in a hurry to come out with 5060 series which should make 4070 go EoL.

However 3060 series sold deep into 4000 series and still does actually

I could easily see 4060 and 4070 being sold for years still, it is not exactly like RDNA4 is going to change much about the GPU brackets, just same perf with slightly less power draw probably - AMD left high-end gaming GPUs officially.

Top RDNA4 / Radeon 8000 card will offer 7900GRE performance or so. Thats 4070 series ballpark. Nvidia has no reason to rush anything really.

5090 and 5080 will come out in Q1 next year for 1999+ and 1200 dollars and then the most important cards are out. The remaining cards will simply just overlap with current cards in terms of performance, just like RDNA4.

"Next Gen" won't offer anything really. Just more of the same, except if you buy 5090, which is probably going to beat 4090 by 50-75%

5080 will bring for 4090 performance for less money. All that it needs to do. Expect 1200 dollars.
 
Last edited:
4070 is not going EoL "soon"

5090 and 5080 is Q1 next year.

5070 is Q2 slash Summer and I don't think Nvidia are in a hurry to come out with 5060 series which should make 4070 go EoL.

However 3060 series sold deep into 4000 series and still does actually

I could easily see 4060 and 4070 being sold for years still, it is not exactly like RDNA4 is going to change much about the GPU brackets, just same perf for slightly less power probably (AMD left high-end)


I remember the 3060 getting cut to 8GB. And that’s a x60 family, those tend to be produced/sold in greater quantities. Any time the 4070 has left in production is gonna fly by especially after 5000 reviews drop. I doubt anybody is gonna buy one once they see 5080/5090 numbers unless there’s a sale or if NVIDIA has a Zen 5 moment.
 
I remember the 3060 getting cut to 8GB. And that’s a x60 family, those tend to be produced/sold in greater quantities. Any time the 4070 has left in production is gonna fly by especially after 5000 reviews drop. I doubt anybody is gonna buy one once they see 5080/5090 numbers unless there’s a sale or if NVIDIA has a Zen 5 moment.
Considering Nvidia uses TSMC 5nm again, I am personally not expecting big increases with 5000 series. Gaming GPUs are not really a focus for Nvidia (or AMD for that matter).

5090 is going to be MCM or a massive die. The only SKU that will offer a nice jump over what we have now. However 5080 will deliver 4090 for 1200 dollars instead of 1600-1800 dollars I guess.

GDDR7 will bring some performance on its own too, but mostly in high res meaning 4K/UHD+

Most PC gamers use 1440p and below, and don't really need top tier GPUs anyway. I don't see the problem really. People that care, have options.
 
I hope not since AMD was caught doing shady stuff tons of times as well. Gimped PCie speeds etc. 6500XT is among the worst GPUs ever reviewed on this site for example.

Terrible example imo.
The 6500 XT was launched under its own name. I do agree that it was a terrible product and imo shouldn't have been launched or named the RX 6400XT instead to indicate its performance level was far under the RX 6600XT. But that's not quite 'shady' level, what this article seems to refer to as shady practices is launching a new product under the same name with worse performance than customers would expect. The examples given in the article are:

  • GT 1030 - Later DDR4 model had 40% worse performance (and wasn't any cheaper to add insult to injury)
  • NVIDIA RTX 3060 8GB - Released after the 12GB it had the same name but not just memory, it also had 33% less memory bandwidth and 3584 shading units (down from 3840)
  • NVIDIA RTX 4070 - What this article itself is about, GDDR6 instead of GDDR6X. Seems to be about a 3-5% decrease in performance

However that's not all, some examples I can think of are:

  • GTX 1060GB The later released 3GB version didn't just get gimped in the VRAM department the CUDA cores also went to 1152 (from 1280)
  • GTX 1050GB The 3GB version sounds like it should be superior to the 2GB version. However it had fewer CUDA cores so it wasn't in some rare cases
  • GTX 660 The 3GB model had less memory bandwidth than the 2GB model which like the above example could lead to less performance for the model with more VRAM

Now I'm not saying AMD is a saint free from doing the same, but NVIDIA does this a lot!.
AMD had the RX 560 in both a 1024 and 896 shader units variant. Worth noting that you couldn't buy that 896 variant directly though, it only came in prebuilds.

imo one of the shadiest things NVIDIA has done was the GTX 970 with its 3.5GB (fast) VRAM + 0.5 (slow) VRAM. I got burned on that one myself. It was highly recommended everywhere when I bought it but I had to send it back pretty much the second I got it. Performance was terrible in Skyrim with mods. Replaced it with an AMD RX 280 8GB which was great.
 
On a similar note, I dunno why they gave the RTX 3060 Ti and 3070 just 14Gbps VRAM, same as the outgoing 2060 to 2080 Super Turing cards. Shoulda bumped it up to 16Gbps
 
It just proves that conventional GDDR is as good.
Its not as good, it works tho. Except for high-end GPUs with high res focus where you need the bandwidth, meaning 4K/UHD+ gaming. This is also where GDDR7 will shine on 5090 and 5080 in a few months.

However most 4K gamers use upscaling anyway.
 
On a similar note, I dunno why they gave the RTX 3060 Ti and 3070 just 14Gbps VRAM, same as the outgoing 2060 to 2080 Super Turing cards. Shoulda bumped it up to 16Gbps
3060 Ti and 3070 easily does 1000+ OC on memory, so yeah, they could easily have clocked them higher at stock

The 3070 in my HTPC does +200 Core and +1250 Mem
 
It just proves that conventional GDDR is as good.
Nah its not as good, but conventional works for lower end stuff.

My 4090 would not have 1 TB/s with GDDR6. Runs more like 1.20 TB/s post OC tho.

GDDR7 will make 5090 do 1.5 TB/s, maybe more, with OC, probably 1.6-1.75 TB/s + Teoretically +50% Core Count increase compared to 4090.

28/32GB 448-512 bit bus.

Might upgrade. Might wait for 6090/6080 at true 3nm, will see.
 
Its not as good, it works tho. Except for high-end GPUs with high res focus where you need the bandwidth, meaning 4K/UHD+ gaming. This is also where GDDR7 will shine on 5090 and 5080 in a few months.

However most 4K gamers use upscaling anyway.
ROFL... hell no...

If you play at 2160p, you most likely have an enthusiast GPU.

I play at 2160p natively since 2015.

As for GDDR non x, the review is really clear that it changes nothing. Nvidia only used it to raise the tag prices of their GPUs to create a resentment of FOMO.
 
Nah its not as good, but conventional works for lower end stuff.

My 4090 would not have 1 TB/s with GDDR6. Runs more like 1.20 TB/s post OC tho.

GDDR7 will make 5090 do 1.5 TB/s, maybe more, with OC, probably 1.6-1.75 TB/s + Teoretically +50% Core Count increase compared to 4090.

28/32GB 448-512 bit bus.

Might upgrade. Might wait for 6090/6080 at true 3nm, will see.
You are speculating unless you removed the memory modules on your 4090 to standard GDDR non-x and bench it with replicable and meaningful results.

So far, you will be lucky to have mere percentage of better performances on the speed of memory. It has limited return value and always had. In the end, there is way more important factors to take in consideration before bringing the limitation of the memory speed for gaming performances.

Memory size is way more important than its speed because it really shows when the issue occurs.

2160p-o.png
 
You are speculating unless you removed the memory modules on your 4090 to standard GDDR non-x and bench it with replicable and meaningful results.

So far, you will be lucky to have mere percentage of better performances on the speed of memory. It has limited return value and always had. In the end, there is way more important factors to take in consideration before bringing the limitation of the memory speed for gaming performances.

Memory size is way more important than its speed because it really shows when the issue occurs.

2160p-o.png

Yeah 7900 series are selling so well, I guess thats why AMD left high-end gaming GPU market.

4090 absolute crushes 7900XTX in most games at 4K/UHD+ anyway.


One of the most popular and demanding games right now. 7900XTX is on 4070 Ti S level with inferior visuals due to FSR is beat by DLSS/DLAA with ease.

AMDs next GPU launch will target low to mid-end only. White flag was raised. AMD has like 10% dGPU marketshare right now and dropping. Lets hope RDNA4 will get very aggressive pricing, so AMD can win back some, even tho they will loose money in the process probably.

AMD is simply years behind on features, RT, Path Tracing, everything new. That is the reason AMD users hate upscaling, because FSR is just bad in comparison to DLSS/DLAA. Sadly for AMD and its GPU fanboys, upscaling is here to stay and is replacing AA left and right in new games. Just like RT elements exist in many games, forced that is.

AMD having inferior RT is going to bite them in the *** going forward. Looking at simple raster performance only in late 2024 is a big no-no.
 
Last edited:
ROFL... hell no...

If you play at 2160p, you most likely have an enthusiast GPU.

I play at 2160p natively since 2015.

As for GDDR non x, the review is really clear that it changes nothing. Nvidia only used it to raise the tag prices of their GPUs to create a resentment of FOMO.
An AMD fanboy playing at 4K/UHD natively since 2015? Yet AMD don't even have a good 4K GPU today. Are you a 30 fps gamer?

I use 4090 at 4K OLED 240 Hz. 4090 is the only true 4K solution today. 5090 and 5080 will follow soon.

DLAA improves massively on native. I use DLAA in all DLSS games. Being 600+ for now. FSR is meh and has no widespread support.

7900XTX can't do 4K natively in demanding games without upscaling unless you run games on medium lmao. Enable RT and it blows up. Talking about overall, not a few cherrypicked games.
 
Ah well, a friend bought a 1000 dollar 32 inch 4k monitor being 100% convinced he bought a native Gsync. That was the only real demand he had and he still think the monitor has a Gsync chip. But I know his monitor is Gsync adaptive and does not have a Gsync chip. He is happy and I leave it that way. If he had done some research...
Always do your research...look at specs.

Did it matter to him tho? He got VRR regardless!
 
Back