Hardware & Technical Latest Nvidia Ampere Rumours

Both the 3070 and 3080 are quite a bit cheaper than I feared. The 3090 is a definite no-go for me, and I assume most people. Will have to see how the suggested price translates to local markets but the 3080 looks interesting. And I definitely want a HDMI 2.1 card. Might wait to see what AMD is up to, but I am not that convinced they can reallyo miles above this...
 
Both the 3070 and 3080 are quite a bit cheaper than I feared. The 3090 is a definite no-go for me, and I assume most people. Will have to see how the suggested price translates to local markets but the 3080 looks interesting. And I definitely want a HDMI 2.1 card. Might wait to see what AMD is up to, but I am not that convinced they can reallyo miles above this...

AMD will price the high-end RDNA2 parts at whatever it takes to be competitive, but NVIDIA has to be very confident they have nothing to challenge the 3090.
 
Both the 3070 and 3080 are quite a bit cheaper than I feared. The 3090 is a definite no-go for me, and I assume most people. Will have to see how the suggested price translates to local markets but the 3080 looks interesting. And I definitely want a HDMI 2.1 card. Might wait to see what AMD is up to, but I am not that convinced they can reallyo miles above this...
You want HDMI 2.1 to play 4K at 120 FPS ? Otherwise I read that it is rather a technology for the future which is not really useful currently.

🦠 🦠 🦠 🦠 🦠 🦠 🦠 😷
 
Most recent GPU I've fully unlocked to a higher model was a Radeon HD R9 290, which took a 290X bios and became a full fledged 290. There have been a few cases like this with more recent GPUs, as board partners still occasionally flash higher-end boards to lower-end cards to make up for supply deficiencies in the lower-end models. It's certainly uncommon though.

Anyway, I fully expect the disabled portions of the GPU to be fused off and impossible to reenable with the RTX 3000 series, and I expect different physical memory ICs on the 3080 and 3090 boards as well as two unpopulated memory channels on the 3080.

If the memory tables are compatable, flashing 3090 firmware onto a 3080 could still be the easiest way to increase the maximum power limit of a 3080, without having to resort to hardware mods.



The board power requirements are not all that different from prior generations, and should be just as tunable. Idle power consumption is also negligible and performance per watt has certainly increased. You don't need an 850w PSU for a high-end system with a 3090, and even if you did, it's not going to draw anywhere near peak loads the overwhelming majority of the time.



We don't know how different these processes actually are in efficiency as no GPUs made on Samsung's 7nm process have yet been reviewed.



Not practical configurations. The GPU has a 384-bit memory interface which mandates 12/24GiB in fully enabled configurations. ROP clusters are almost certainly still tied to the memory controllers as well, making it impractical to cut them down without disabling memory channels.

The 3090 has 24GiB to make sure it has a bigger number on it than the competitor's part. People will buy 24GiB cards at the high-end even though it's a meaningless figure and by the time any game benefits appreciably from more than twelve, the 3090 will be quite long in the tooth.

The 3080 has 10GiB because it's only got ten of twelve memory channels enabled and putting 20GiB on the Founders Edition would be an extra cost on the BoM and would eat into AIB's options for making distinctive non-reference products.
Point taken in memory bandwidth but the ampere is 8nm Samsung not 7nm so I am convinced The TSMC part would be significantly lower power for the same performance.
I also think you are mistaken on power draw. I don't think they would say it was a 350w part for the fun of it. My 1080ti is a 250w part and overclocke I reckon I run close it it. With all my stuff on my pc I think a 650w GPU even a good quality one would be too light for that rig with a few hard disks and HOTAS, VR etc
Don't get me wrong I am happy with the reveal it looks a nice step up but I think this could be like the gtx 480fermi. Decent card but Hot and hungry and iirc that was replaced fairly quickly with a more refined 500 series card (which was the one I bought)
 
Last edited:
On my 4770K, I can hardly imagine a 3090 card. :)

🦠 🦠 🦠 🦠 🦠 🦠 🦠 😷
I am going to replace my overclocked 5020k next upgrade as I don't want to bottleneck. Definitely looking to AMD for CPU and depending on bignavi maybe GPU too. Will be my 1st non NV card since ATI 9800 era (that card came with a key for half-life 2 a year or more before HL2 launched
 
AMD will price the high-end RDNA2 parts at whatever it takes to be competitive, but NVIDIA has to be very confident they have nothing to challenge the 3090.
I am surprised on the price of both the 3080 and the 3070, I reckon Nvidia are going to shift a lot of them.

I wonder if that is agressive pre-emptive pricing against Big Navi because if the performance of those cards is a good as quoted, that actually seems like pretty good value. The price of the 3080 FE (£649) is only slightly more than I paid for my 980Ti back in 2015.

Whilst I am itching to buy a new card I will hang fire to see what Big Navi brings, there is the possibilty their cards may have more VRAM at a similar price/performance to 3070/3080 . I dont think they will be able to touch the 3090 although I would love to be proved wrong.
 
Last edited:

Robert Maynard

Volunteer Moderator
Whilst I am itching to buy a new card I will hang fire to see what Big Navi bring, there is the possibilty their cards may have more VRAM at a similar price/performance to 3070/3080 . I dont think they will be able to touch the 3090 although I would love to be proved wrong.
I'm waiting for Big Navi information too - I don't expect AMD to take the performance crown, but strongly suspect that they will get closer to the top this time (than they did with the 5700XT).
 
Last edited:
Point taken in memory bandwidth but the ampere is 8nm Samsung not 7nm so I am convinced The TSMC part would be significantly lower power for the same performance.

Without an actual comparison this is impossible to assess, especially across manufacturers. There is no standard for measuring node feature size across foundries, and each foundry has at least two or three different processes/densities per node.

Pascal was on TSMC 12nm 16nm and has similar or better power/performance efficiency than 1st gen RDNA on TSMC 7nm (I've got a 5700XT as well as two 1080 Ti and normalize for performance, they are almost the same power consumption).

I am surprised on the price of both the 3080 and the 3070, I reckon Nvidia are going to shift a lot of them.

I wonder if that is agressive pre-emptive pricing against Big Navi because if the performance of those cards is a good as quoted, that actually seems like pretty good value. The price of the 3080 FE (£649) is only slightly more than I paid for my 980Ti back in 2015.

Whilst I am itching to buy a new card I will hang fire to see what Big Navi bring, there is the possibilty their cards may have more VRAM at a similar price/performance to 3070/3080 . I dont think they will be able to touch the 3090 although I would love to be proved wrong.

The 3070 and 3080 are only slightly less expensive than I was expecting, while the 3090 is slightly more. Since I was heavily leaning toward a 3080 before, I'm pretty dead set on it now, if I go with NVIDIA.

I do hope the prices are reflective of something NVIDIA knows about RDNA2 that we do not, but they could be trying to undercut AMD right out of the gate, or be taking a long term strategy to undercut AMD's profits. NVIDIA is doing extremely well and can afford to take lower margins than they'd prefer on the volume parts.
 
Last edited:
Impressive, but the computer will have to be high-end also to keep pace ?

🦠 🦠 🦠 🦠 🦠 😷

Well, kinda. To get the most out of them you should have a decent system.
4770k is starting to show its age. I'm about to switch out my 4790k I've had for 6 years(!).
I'm not going high end though. A Ryzen 3700X, X570 chipset mobo for the new PCIE4 lanes, 32GB 3600mhz ram, and two high speed PCIE4 SSDs.

These cards are also the first to take advantage of the higher speed PCIE4 lanes, so even more reason to upgrade (y)
 
3080 looks like a good way to get back on Team Green. So now founders card or partner card? Feel like Evga might be the way to go.
 
Top Bottom