Caporegime
Because it is.... what its not is N5.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
The better late than never OcUK RTX4070 Super review thread!
- Thread starter CAT-THE-FIFTH
- Start date
Because it is.... what its not is N5.
No it’s not N5 but I never said it was, I said it was made on TSMC 5nm and it is.
Caporegime
What you said the RTX 40 series was built on 5nm, that's N5, N4 is not 5nm, its denser and more efficient than that but its also not 3nm, IE N3, its between, hence N4, its 4nm. IE not 5nm.
Last edited:
Caporegime
The clue is in the name, its really simple.
Caporegime
In any case the only reason Nvidia are able to cut the bus width down is due to the quite substantial increase in memory IC speed over the last 2 years, these are increases in performance Nvidia are not passing on to us, instead they use it to increase their margins.
AMD still offer a 256Bit or higher bus at $500+, as they did in previous generations.
AMD still offer a 256Bit or higher bus at $500+, as they did in previous generations.
Last edited:
Again its an enhanced 5nm process not true 4nm, the same with intel optimizing 14nm over a number of years.
The reason Nvidia is able to cut the BUS is because they used a large cache to compensate even though it doesn't fully compensate, they got away with it to an extent though because AMDs performance sucked this generation.
Last edited:
Caporegime
First you use TSMC's papers to strengthen your argument and then dismiss them when they don't, classic...
I don't think it does.. AMD left about 15% on the table with Navi 32, despite this they are still faster than the 4060Ti and 4070, once you unlock that tabled performance the 7700XT is as fast as the 4070 and the 7800XT as fast as the 4070Ti, they are also very much cheaper and the 256Bit bus helps the 7800XT in higher resolutions, as you would expect.
Besides that you should be more concerned with what Nvidia are doing, rather than blaming AMD for Nvidia's choices, AMD have no influence on what Nvidia do, none what-so-ever, so far as Nvidia are concerned AMD don't exist, so by-proxy 1D chess mental gymnastics to make the Nvidia problem AMD's responsibility do nothing other than reinforce Nvidia in what they are doing.
Its something tech jurnoes and NPC PCMR's have been doing for years and keep doing it expecting... (LOL) a different outcome each time.
Last edited:
What I find funny is that it must be cheaper to cut the bus width and add extra cache, otherwise Nvidia wouldn't do it. Yet we've been told that cache is expensive and takes up a lot of die space
The problem with cache is it's only useful until it's filled up, then the game relies on vram with its slow bandwidth. So wide bus, high bandwidth vram will be better than cache overall, until such time where we can have a very large amount of cache on the GPU (like 1GB of cache). Some of the early rdna3 rumours said the 7900xtx would have 512 to 1024MB of cache, but ended up as 96MB
The problem with cache is it's only useful until it's filled up, then the game relies on vram with its slow bandwidth. So wide bus, high bandwidth vram will be better than cache overall, until such time where we can have a very large amount of cache on the GPU (like 1GB of cache). Some of the early rdna3 rumours said the 7900xtx would have 512 to 1024MB of cache, but ended up as 96MB
Last edited:
Soldato
- Joined
- 28 May 2007
- Posts
- 10,158
Yea but it still has benefits as for similar performance AMD use more power to performance. Had AMD paid for this node they may look similar to Nvidia on power consumption. Googled it and it's 11% performance boost at 22% power saving. Probs to steep for AMD in the graphics world but shows that Nvidia had a hand up by going in this direction.
Last edited:
AMD didn't leave 15% on the table, they obviously had QC issues and had to scale back performance for stability. Some cards will OC some will not. If they could have got 15% more on each card then they would have done so and charged more for it.
Last edited:
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 31,006
Damn right they would, AMD are not your friend! Their only friend is money.
Soldato
- Joined
- 28 May 2007
- Posts
- 10,158
I reckon on this node 15% would be possible. To gain 10-15% i need to use 40-50 more watts while undervolting so with 11% improvement off the bat and more power to use it's not far fetched. Also not sure how this would go in games as benchmarks allow you to push further. Mine is only an xt though so not sure how the overclocking goes on the xtx. With those benefits though am sure AMD could get close.
Last edited:
Caporegime
Rumour is Nvidia paid a lot of money to have N4 all to themselves.
Soldato
- Joined
- 28 May 2007
- Posts
- 10,158
Most likely they did. I thought Apple were always first to the new nodes though. Either way Nvidia would have been paying a lot but they can afford to with there margins being so high.
Caporegime
Apple reserved 90% of 3nm already in 2020.... TSMC had some production delays on N3 but yeah, Apple thought they would be knocking out 3nm chips in 2021 or 2022.
Die size is at premium so cache is expensive in terms of die costs. Simple to design though compared to just about anything else.
Power saving too. If they had put in less cache in Ada then for the same performance they would have needed more bandwidth and then team green and team read would have been far closer this round for power consumption - especially if AMD had gone monolith on Navi32.
I would have thought AMD would have had first refusal being TSMCs second largest customer.
Caporegime
Last edited:
What RTX4070 twin fan would you recommend buying for a SFF case which has a GPU size compatibility of Dual Slot, up to 280mm in length, up to 125mm in height?
A lot of the gpu for sale on OC don't seem to show dimensions.
What are the "founder editions" of GPU's?
Cheers
A lot of the gpu for sale on OC don't seem to show dimensions.
What are the "founder editions" of GPU's?
Cheers
Last edited:
FE:
www.techpowerup.com
Asus Dual:
www.techpowerup.com
PNY:
www.techpowerup.com
FE=24CM, Asus Dual=27cm, PNY=25CM
NVIDIA GeForce RTX 4070 Super Founders Edition Review
The NVIDIA GeForce RTX 4070 Super launches today, boasting a solid increase in GPU cores, ROP units and cache. This enhancement brings its performance much closer to that of the GeForce RTX 4070 Ti, at a price point of $600 that signals strong competition for AMD's Radeon RX 7800 XT.
Asus Dual:
ASUS GeForce RTX 4070 Super Dual Review
The ASUS GeForce RTX 4070 Super Dual is sold at NVIDIA MSRP of $600. You still get a greatly improved triple-slot cooler with a dual BIOS feature. Testing in our review confirms that the noise levels with the quiet BIOS are whisper quiet and temperatures are still good.
PNY:
PNY GeForce RTX 4070 Super Verto Review
The PNY GeForce RTX 4070 Super Verto boasts a compact dual-slot configuration, ensuring compatibility with all cases and efficient operation even with a less powerful power supply. However, the design does have its downsides, particularly in terms of temperatures and noise levels.
FE=24CM, Asus Dual=27cm, PNY=25CM
Last edited: