Question 'Ampere'/Next-gen gaming uarch speculation thread

Page 134 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
O

Ottonomous

Senior member
May 15, 2014
559
292
136
How much is the Samsung 7nm EUV process expected to provide in terms of gains?
How will the RTX components be scaled/developed?
Any major architectural enhancements expected?
Will VRAM be bumped to 16/12/12 for the top three?
Will there be further fragmentation in the lineup? (Keeping turing at cheaper prices, while offering 'beefed up RTX' options at the top?)
Will the top card be capable of >4K60, at least 90?
Would Nvidia ever consider an HBM implementation in the gaming lineup?
Will Nvidia introduce new proprietary technologies again?

Sorry if imprudent/uncalled for, just interested in the forum member's thoughts.
 
Sort by date Sort by votes
C

CastleBravo

Member
Dec 6, 2019
119
271
96
CakeMonster said:
A 3080Ti could be 1 year out, its unfortunately a gamble to wait for it, the whole situation could be different by then.
Click to expand...

I wouldn't be surprised if the 3080 ti ends up being a 20 Gb 3080 with maybe a few more SMs enabled and a slightly higher memory clock that gets revealed just after Navi 21 assuming AMD's card beats the base 3080 in some way.
 
Upvote 0 Downvote
swilli89

swilli89

Golden Member
Mar 23, 2010
1,558
1,181
136
Very interesting that NVIDIA used Samsung's revised 10nm process. AMD will likely be using TSMC's 7nm+ for RDNA 2 later this year.

From the looks of it, a 3070 should be about 20-25% faster than a 2080 non-super in normal rendering? Not mind blowing given a 2080 was already available for near $500.
 
Upvote 0 Downvote
M

MrTeal

Diamond Member
Dec 7, 2003
3,569
1,699
136
randomhero said:
I think that is it. Turing could do one int and one fp op per alu per clock. Ampere can probably do 2 fp or one fp and one int op per clock. Ultimately you are always limited by bandwidth. So you see around max 30% uplift in performance if I read that chart correctly(3080 above 2080ti).
Click to expand...
A100 has 6912 CUDA cores at 1.41GHz to give 19.5 TFlops FP32. It would be pretty incredible if 36 "Shader" TFlops for the 3090 was FP32, but the seems very unlikely.
 
  • Like
Reactions: spursindonesia, lightmanek, randomhero and 1 other person
Upvote 0 Downvote
DiogoDX

DiogoDX

Senior member
Oct 11, 2012
746
277
136
Jc4xwEz.jpg


Not even the cards that they were comparing. If this is 3090 vs 2080Ti the performance uplift for 4K with double RT, more that double FP32 and tensor, 50% more bandwidth and 350W is not impressive.

I think will be the 40-50% more in traditional raster like the rumors.
 
  • Like
Reactions: spursindonesia
Upvote 0 Downvote
K

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,249
136
moonbogg said:
Definitely waiting on the 3080Ti. $700 is better than expected, but I can't downgrade on Vram while upgrading in other areas. A next-gen GPU at $700 shouldn't be a pseudo-upgrade in any way. I think Lisa is going to kick his ass. I'd expect a 3080Ti next year or something, possibly sooner depending on the competition. I'm looking forward to those 3090 AIB markups for the epic lulz though.
Click to expand...

In real life I wouldn't be surprised if she could kick his arse!

Pricing seems too cheap at 1st glance coming from Nvidia. Maybe there's more to it?

Not saying AMD will come out on top, but if so they should call the top card " FrameCrusher " lol
 
  • Like
Reactions: spursindonesia and moonbogg
Upvote 0 Downvote
Hi-Fi Man

Hi-Fi Man

Senior member
Oct 19, 2013
601
120
106
randomhero said:
I think that is it. Turing could do one int and one fp op per alu per clock. Ampere can probably do 2 fp or one fp and one int op per clock. Ultimately you are always limited by bandwidth. So you see around max 30% uplift in performance if I read that chart correctly(3080 above 2080ti).
Click to expand...
NVIDIA is historically good at keeping their GPUs fed. I expect a lot of L2 cache on these GPUs.
 
  • Like
Reactions: randomhero
Upvote 0 Downvote
DooKey

DooKey

Golden Member
Nov 9, 2005
1,811
458
136
Kenmitch said:
In real life I wouldn't be surprised if she could kick his arse!

Pricing seems too cheap at 1st glance coming from Nvidia. Maybe there's more to it?

Not saying AMD will come out on top, but if so they should call the top card " FrameCrusher " lol
Click to expand...

I think pricing is based upon the deal they got from Samsung. Also, could be because of the deal they didn't get from Samsung...7nm EUV.

They know they are still behind the foundry curve and had to build big but have a product that will compete with AMD. It's all about competition.
 
  • Like
Reactions: spursindonesia
Upvote 0 Downvote
swilli89

swilli89

Golden Member
Mar 23, 2010
1,558
1,181
136
Konan said:
To be super clear it is a custom 8nm process for NV. Period.
Click to expand...

Yes but anything can make it custom. The smallest, tiny tweak and it can be labeled custom. It costs multiple billions of dollars to research, and commercialize a new process which I think is a very well known fact. I can promise you Samsung didn't make a major alteration to a process just to make a run of video cards for a single customer. This is Samsung's 10nm++ in other words. Very mature at this point and perhaps tuned to build much larger chips versus a typical Samsung ARM SoC.
 
  • Love
Reactions: spursindonesia
Upvote 0 Downvote
M

Macros96

Junior Member
Apr 27, 2017
6
6
81
Insomniator said:
Very little to complain about here from what I saw. Will be funny seeing people trying to sell used 1080TI's for $500 still.

3080 for $700 seems great if they are actually available. Same for the 3070.

3090 looks good as a halo, I'm sure people will cry about it but they can charge whatever the **** they want for the halo card (I might preorder one).
Click to expand...

I may as well. The final decision will be made based on the raster performance of the 3090. I like the specs so far but they better have increased the TMU/ROP throughput to match the increased SM perf.

My MO has always been buy the highest single card available (within reason, no RTX Titan shenanigans) and slap a water block on it. See no reason to change now.
 
Upvote 0 Downvote
mohit9206

mohit9206

Golden Member
Jul 2, 2013
1,381
511
136
I honestly don't think AMD will be able to compete with 3070 and 3080. You guys are being bit too optimistic considering AMD track record in graphics card. I first need to see a price cut on 5700XT to $300 to start things and $200 on 5600XT.
 
  • Haha
Reactions: spursindonesia
Upvote 0 Downvote
D

Dribble

Platinum Member
Aug 9, 2005
2,076
611
136
Well performance looks amazing, price is good, stock cooler looks good, they will be able to churn them out as they will have a lot of capacity on that Samsung node. Downers are power usage and memory - I would look at a 3070, would have prefered 10-12gb not 8 or expensive 16.
 
Upvote 0 Downvote
M

Mopetar

Diamond Member
Jan 31, 2011
7,837
5,992
136
Hitman928 said:
Nvidia website listing 3080 FE price as $699 so I was wrong, looks like that will be the actual price. That's really good pricing for the 3080 but it also seems way out of place with the 3070 at $499 and the 3090 at $1499.
Click to expand...

I think it tells us what NVidia is expecting AMD to have. Competition at the 3070 and 3080 level, but nothing that could comfortably slot in at $1,000.

Now NVidia could be wrong, but they'll look really stupid if there is a good card that fits in that gap.

I still want to see benchmarks before passing final judgment though. NVidia was always going to show the card in the best light possible, and it's clear that's a lighting powered by ray tracing.
 
Upvote 0 Downvote
K

Konan

Senior member
Jul 28, 2017
360
291
106
swilli89 said:
Yes but anything can make it custom. The smallest, tiny tweak and it can be labeled custom. It costs multiple billions of dollars to research, and commercialize a new process which I think is a very well known fact. I can promise you Samsung didn't make a major alteration to a process just to make a run of video cards for a single customer. This is Samsung's 10nm++ in other words. Very mature at this point and perhaps tuned to build much larger chips versus a typical Samsung ARM SoC.
Click to expand...

So should we just call any card on 7nm, '7nm" then and forget a + or any variation? I don't think so.
You can't promise anything mate.
It was clearly labelled 8nm custom process.
 
Upvote 0 Downvote
M

MrTeal

Diamond Member
Dec 7, 2003
3,569
1,699
136
Videocardz is running pretty hard with the 2X CUDA cores. It will be interesting to see how this all plays out when reviews come out.
videocardz.com

NVIDIA announces GeForce RTX 3090, RTX 3080 and RTX 3070 - VideoCardz.com

NVIDIA announces GeForce RTX 30 series NVIDIA today announces its new GeForce RTX 30 series. NVIDIA confirmed that the series based on Ampere architecture are manufactured in Samsung 8nm process technology. The 30 series were announced by NVIDIA CEO Jensen Huang during a special GeForce live...
videocardz.com videocardz.com
 
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom