News AMD Navi GPUs: Rumors, Release Date, All We Know

abryant

Asst. Managing Editor
Staff member
May 16, 2016
183
17
18,685
Here's everything we know about AMD's upcoming Navi GPU architecture. Read more here.
navi.jpg

ZHIYE LIU
@zhiyeliu

Zhiye Liu is a Contributing Writer for Tom's Hardware US. He covers CPU, GPU and motherboard news.
 
If the specs and pricing is true, Nvidia will be scrambling to lower prices. They'll still hold the performance crown, but price/performance could sway towards AMD.
 
If the 3060 uses the same die as the 3070, then it means the 3070 is down to 128bits-wide bus. I hope AMD pulled a small memory efficiency miracle to make it work out fine. I was hoping for a 192bits bump and 6GB baseline for the lower-midrange as 4GB seems too low for RX580-class performance on the 3060.
 
On paper, the 3080 doesn't sound bad, but I'm wondering about the power consumption. Looks like this will be another generation of hot cards. Not that it's a deal breaker or anything, but at 250W being "mid-range" is a hard sell for the expected performance of the high end... With just 50W extra to work with, I don't have much faith on seeing anything that can touch the 2080ti, like at all.

Cheers!
 
On paper, the 3080 doesn't sound bad, but I'm wondering about the power consumption. Looks like this will be another generation of hot cards. Not that it's a deal breaker or anything, but at 250W being "mid-range" is a hard sell for the expected performance of the high end
Rumored TDP listed in the article for the 3080 is 150W, not 250W.
 
If the specs and pricing is true, Nvidia will be scrambling to lower prices. They'll still hold the performance crown, but price/performance could sway towards AMD.
Yep Nvidia has the performance crown but most that buy GPUs are looking for cards in the $200 to $300 range that is where the majority spend their money. If AMD wins that segment they will be doing great.
 
If the 3060 uses the same die as the 3070, then it means the 3070 is down to 128bits-wide bus. I hope AMD pulled a small memory efficiency miracle to make it work out fine. I was hoping for a 192bits bump and 6GB baseline for the lower-midrange as 4GB seems too low for RX580-class performance on the 3060.

I read a leak the other day for Linux drivers that seems to point Navi as being just another iteration of GCN. If thats the case then they are basically just shrinking and improving GCN to 7nm. The benefits will mostly be the power draw from 7nm but I don't know if performance will be much better, or at least enough to compete evenly model for model with nVidia.

They will probably have to keep costs low to compete better.

I do hope the rumored names are not correct though but considering AMD took similar Intel chipset names only one gen ahead I would not be surprised if they are planning on doing the same in the GPU market. I actually had to explain to someone the other day why they couldn't use a X299 board with a Ryzen CPU.
 
I do hope the rumored names are not correct though but considering AMD took similar Intel chipset names only one gen ahead I would not be surprised if they are planning on doing the same in the GPU market.

I don't care which manufacturer, Intel or AMD they are both stupid when it comes to motherboard naming
For instance A320, B350, X370
Why change both the Letter AND the number? If you change just the Letter eg A350, B350, X350 - no need to change the numbers. If you change the numbers, No need to change the letter, eg A320, A350, A370 - this makes so much more sense, then the revision would be the LETTER going up.... Anyway gripe over.

As for Navi and other rumours, they will be out in Q3 this year, according to Guru3d website posted today....
 
No need to change the letter, eg A320, A350, A370 - this makes so much more sense, then the revision would be the LETTER going up....
Why? If all your chipsets have a base model of Axx0, the 0 is superfluous and could be used for revisions which would make more sense than changing the letter which typically stands for product line.

Aside from that, I agree that model numbers are getting stupid. Not as stupid as Intel's decision to drop model suffixes in favor of ofuscated model numbers because feature suffixes were "too confusing" for normal people... now we need look-up tables to figure out which model numbers support what instead of looking for the appropriate suffix letter.