AMD to reportedly skip 20nm entirely for its 2016 GPU family

Scorpus

Posts: 2,156   +238
Staff member

According to the latest reports, AMD will skip the problematic 20nm manufacturing node for their next-generation 'Arctic Islands' GPUs, which the company is expected to launch in 2016. Instead, Arctic Islands will progress directly to 14nm FinFET in the hope of attaining higher levels of efficiency.

Currently, all GPUs in the market are produced using a 28nm process, both from AMD and Nvidia. It's widely expected that AMD's upcoming Rx 300 line of GPUs, set to launch in a few months, will also be manufactured using a 28nm process due to a slew of issues with TSMC's 20nm node that have repeatedly delayed its introduction.

At this stage it's not known whether AMD will stick with TSMC and hope that their 14nm FinFET node is ready on time, or whether they'll switch to a company like Intel or Samsung, both of which have operational 14nm nodes. Currently the node is being used for CPU, SoC and NAND production, but it's expected to suit GPUs just as well.

Considering Arctic Islands is two generations away, details are a little scarce at this stage. With the move to 14nm, these graphics cards are expected to feature low TDPs, similar to Nvidia's Maxwell cards today. 'Greenland' is the codename for Arctic Islands' flagship GPU, which will come with an advanced design featuring SK Hynix's HBM2 memory.

For now though, AMD is focused on getting the Rx 300 line out the door, which should launch around Computex in June.

Permalink to story.

 
Does it really make any difference? As far as I've heard most of the R300 series is mainly going to be a rebrand of the R200 series although I sincerely hope not, Nvidia can do with a shake up, only their high end gaming cards are worth taking note of.
 
Does it really make any difference? As far as I've heard most of the R300 series is mainly going to be a rebrand of the R200 series although I sincerely hope not, Nvidia can do with a shake up, only their high end gaming cards are worth taking note of.
the high end cards will have a new GPU, but most of the other cards will just be rebrands with some small changes. it's kinda to be expected because of the problems they had with switching to a smaller node. Nvidia is in a similar situation although they might release new cards on a smaller node first.
I for one don't really care about the smaller nodes as much as for the new 3D memory that we'll see in the new high end AMD GPU. I really want to see some benchmarks for it. From what I read NVIDIA will also use 3D memory in Pascal in 2016.
 
Last edited:
We are getting so close to being unable to use smaller nodes. I suspect that we'll see a lot of delays in the future. 12/14nm seems to be here to stay for quite a few years.

Hence the 490X has a chance of becoming the next 7970 that stands as a powerful GPU for over 4 years.
 
Does it really make any difference? As far as I've heard most of the R300 series is mainly going to be a rebrand of the R200 series although I sincerely hope not, Nvidia can do with a shake up, only their high end gaming cards are worth taking note of.

That's not true anymore. They delayed the line-up to June specifically so they can do a full redesign on everything 360X and up. This means implementing GCN 1.3 and the new color compression into all the gaming GPU's. Big performance boosts incoming.
 
Does it really make any difference? As far as I've heard most of the R300 series is mainly going to be a rebrand of the R200 series although I sincerely hope not, Nvidia can do with a shake up, only their high end gaming cards are worth taking note of.

Incredibly not likely.

Most of the 200 series is already either a rebrand of old cards from HD 7000, either heavily overclocked cards form HD7000 and some new cards here and there (R7 260X and R9 290, 290X and 295x2).

The 300 series will probably be made on the new GCN 1.2 introduced with the R9 285.
 
We are getting so close to being unable to use smaller nodes. I suspect that we'll see a lot of delays in the future. 12/14nm seems to be here to stay for quite a few years.
I have been hearing that since the mid-90's... And they are still finding new ways to do things.
I suspect that in 20 years, we will be using some technology that we haven't even heard of yet, that runs 1000 times faster on 1% of the power we use today, and the size of the processor will be measured by quantum units.
 
Does it really make any difference? As far as I've heard most of the R300 series is mainly going to be a rebrand of the R200 series although I sincerely hope not, Nvidia can do with a shake up, only their high end gaming cards are worth taking note of.
Since the 200 series was released, AMD has upgraded it several times. The latest available upgrade I believe was the 285. The 300 was supposed to land awhile back, at 20nm, but TMSC has not been able to get anywhere near the volume levels AMD needs to even have a chance of making money on it. So its been pushed back multiple times. Now, its been redone to be a 28nm chip, like the 200 series, but using less power, and having new features that should be faster, while still using lower power levels.
At least thats what the rumor mill seems to think. We shall see soon enough.
 
Most of the 200 series is already either a rebrand of old cards from HD 7000, either heavily overclocked cards form HD7000 and some new cards here and there (R7 260X and R9 290, 290X and 295x2).
The 300 series will probably be made on the new GCN 1.2 introduced with the R9 285.
While I agree they will finally retire the GPUs from the HD 7000 series they still sell today (Pitcairn and Tahiti), they have no reason not to reuse the newer revisions like Bonaire, Tonga and Hawaii. Those are already in production, which means it will cost them less in R&D to deploy, can now beneft from increased efficiency in manufacturing and better binning, and already support all the new features AMD introduced, like FreeSync, Mantle, TrueAudio and DirectX 12 when it's launched. Alongside the new Fiji GPU, that would cover the entire market.
 
There will be newer cards at the top end but themidrange cards will be rebrands BUT rejigged to take advantage of 3d sta ked memory, this will give them a real boost in bandwidth/speed and is a cheap way to get *something new out the door .. So better perf, lower prices, but no real innnov except hi dense memory. Samsung gets their next gig, then buys them out or joint ventures them in 16/17.
 
That's not true anymore. They delayed the line-up to June specifically so they can do a full redesign on everything 360X and up. This means implementing GCN 1.3 and the new color compression into all the gaming GPU's. Big performance boosts incoming.
I hope so.
 
Incredibly not likely.

Most of the 200 series is already either a rebrand of old cards from HD 7000, either heavily overclocked cards form HD7000 and some new cards here and there (R7 260X and R9 290, 290X and 295x2).

The 300 series will probably be made on the new GCN 1.2 introduced with the R9 285.
Not the GCN 1.3?
 
Since the 200 series was released, AMD has upgraded it several times. The latest available upgrade I believe was the 285. The 300 was supposed to land awhile back, at 20nm, but TMSC has not been able to get anywhere near the volume levels AMD needs to even have a chance of making money on it. So its been pushed back multiple times. Now, its been redone to be a 28nm chip, like the 200 series, but using less power, and having new features that should be faster, while still using lower power levels.
At least thats what the rumor mill seems to think. We shall see soon enough.
Yup. Lets wait and see what they throw out of the door. I believe the GTX980Ti is ready and waiting in the wings to see if it can humiliate the 390 when launched.
 
The 290X is as good if not better than the 780 Ti in many gaming tests (like Crysis 3) and is $200 cheaper! My next card will definitely be AMD/ATI.
 
Back