• 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.

NVIDIA 4000 Series

IT Troll said:
Trying to get a handle on expected raster performance vs the 3090. I think the 4070 4080/12 is about the same or slightly less, the 4080/16 is ~35% faster and the 4090 will be up to twice as fast. Does that sound about right?
Click to expand...
4080 12gb 5% slower, 4080 16gb 15% faster and a 4090 60% faster is my take looking at the specs.

The only card even remotely worth considering at those prices in terms of performance is a 4090.
 
Last edited:
brostradamus said:
overall i still believe the 4090 at 76 billion transistors is going to be a good product, it might be power starved for all we know and some AIB models could address that issue. given how 76 billion was a big surprise, i dont believe AMD would have even planned for it - its just ridiculpus the kind of transistor density nvidia has achieved, i seriously doubt if anyone could have even guess anything around those lines, the broader consensus would have been 55 billion tops, this is a pretty big surprise and i dont believe amd has something even remotely touching that

secondly, nvidia is taking a different approach here, they seem to be relying more on next gen features instead of dedicating more resources to traditional rendering methods.. i feel that nvidia internally believes that rasterization is on the path of diminishing returns and is performing comfortably above reasonable expectations.. for growth to be meaningful


it is rumored to be a mcm design but not in the traditional sense, what they have basically done is separted out the infinity cache on N6 chiplets while the nerve center the actual core will be a different chip, so its still a single die for all practical purposes other than the potential for cost savings

i think that amd officially confirmed that its going to be a MCM design during the zen 4 reveal, the rest of it is speculation but every leaker out there is repeating those lines in unison, igor's lab has a sketch that you could refer
Click to expand...


I don't believe the 76 Billion Transistor count, it's some funny counting or a blatant lie.

Because H100 is a huge die on the same process and has 80 Billion, the ADA GA102 is way smaller 600, I even called out the infamous rumour tweeter on this last week and he didn't reply because he clearly has seen these fake claims from Nvidia or was told them to keep that lie going.


 
IT Troll said:
Trying to get a handle on expected raster performance vs the 3090. I think the 4070 4080/12 is about the same or slightly less, the 4080/16 is ~35% faster and the 4090 will be up to twice as fast. Does that sound about right?
Click to expand...

No. I got that the 4080/12 will be slight slower, the 4080/16 will be just a bit faster and the 4090 will be 50% faster. Now way is it 2x raster.
 
gpuerrilla said:
Yeah, I think even this time, the diehards have thought I wonder what Radeon have to offer then..
Click to expand...

I wonder what AMD will offer you, being you have a 3090. Will be interesting to see after saying these kind of things if you will be buying an AMD 7000 series GPU or do what you did last time around where you bigged up AMD, laughed at Nvidia users, especially ones buying 3090’s as mugs edition and then went on to buy one yourself :cry:

I can see it already, there is a high probability you will just end up sticking with your 3090 and skip a gen :cry:
 
Murphy said:
And the increase in chips per wafer from the fab shrink would probably offset those increased costs, the cost per wafer may have gone up by 20% but the amount of chips, the amount of transistors you can fit on each wafer, goes up because they're smaller. While that may not completely offset the increased wafer costs it does go a long way to offsetting those increases.
Click to expand...

Yes, these chips are on a much denser node compared to 3000 series, and at least 4080 have the same or even fewer cuda cores. So the chips are likely significantly smaller, offsetting the increase in fabrication costs. Nvidia is likely paying less per chip than before.
 
Joxeon said:
Tbh I would have had a hard time considering the 4080 12gb at £500 or the 16gb at £700 with the lukewarm performance on offer so will wait and see what AMD offers and if it's no better then stick with my 3080 for 2 more years.
Click to expand...

Indeed it would look like you would only really gain in RTX DLSS 3 games. Everything else would be little gain, around 25%. So at £1200 thats a poor return.
 
Greebo said:
Indeed it would look like you would only really gain in RTX DLSS 3 games. Everything else would be little gain, around 25%. So at £1200 thats a poor return.
Click to expand...
Not only that they hard lock DLSS 3 to 4000 series which is a really poor move. we are already being shafted for higher GPU prices.
 
Last edited:
apachegoose said:
Review embargo is lifted the day before release, i.e. 11th of October reviews will appear. Normally, having the embargo that close to release is a sure sign of a bad product.
Click to expand...

Surprised its the day before. Wasnt the 3000 series timed with the exact release of stock so if you waited 10 minutes to- read a review everywhere was out of stock?

Having a day before at least means real world game performance will be out there before you can buy the 4090 and a full 2 to 3 weeks before you can buy the 4080.
 
Purgatory said:

:cry: love the opening at the oven .. How true really, that's how I felt the minute he showed the pricing for the 80 class cards and no NVLINK on 90 class.
Click to expand...

I see they changed the title of that video from "The RTX4090 looks SICK". I wonder did viewers give them what for in the comments section... If so, good! The last thing we need is for tech tubers to be normalising these prices.
 
Purgatory said:
I don't believe the 76 Billion Transistor count, it's some funny counting or a blatant lie.

Because H100 is a huge die on the same process and has 80 Billion, the ADA GA102 is way smaller 600, I even called out the infamous rumour tweeter on this last week and he didn't reply because he clearly has seen these fake claims from Nvidia or was told them to keep that lie going.
Click to expand...

It is possible to trade transistor density/area for frequency and performance requirements - don't assume that 80BN at 814mm2 is the max possible - theoretically TSMC 4N can do something like 82-83BN at 600mm2 on paper. You'll often find a non-linear relationship depending on requirements of the chips near the upper end.

I'm not saying either you or Kopi are right or wrong.

EDIT: Also depends on the amounts of different types of processing within the core - you can pack many more transistors into an area for some things and less for others - you might hit a wall at a much lower transistor count in the same area depending on the core config.
 
Last edited:
Joxeon said:
Have a feeling both the 80 class cards are going to get panned in reviews.
Click to expand...

So they should and if they don't we will spot the Nvidia shills. Sorry this time I can't even find one positive with what they have done that benefits the customers, this is nothing more than greed and in a time and age we live in now I hope they learn a very valuable lesson from this generation and people don't hand over their money to such a company or companies that behave this way.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom