TSMC 5 nm+ Node Manufacturing Goes High-Volume in Q4, AMD one of the Major Customers

[Tomgang]

Holy moly 5 NM already. AMD really going all in now. Ryzen 9 4950X 16 cores or maybe more. Come to papa
This also make good sense as Zen 2 XT models might be the 7 NM+ process then.

Intel still being on 14 NM for desktop. It can only end like this AMD (Green smiley) vs. intel:

And i have been a intel supporter already before X58 came out, but i must say at the rate AMD moving on now (CPU´s before Zen 2, dit not impress me. But Zen 2 has and now with possible 5 NM Ryzen 4000, that is really impressive) and intel loves to stick to there 14 NM. I am beginning to be convince that AMD now are getting ahead of intel the coming years.

 

[Jism]

What a fuss. This is just part of the contract(ual) deals AMD has with TSMC. 5+nm only saves like 7 percent of power based on the same clocks, or a 15% uplift in silicon performance. If we snatch 7% off 125(W) of TDP then compared to the 7nm proces your looking at a 8.75 watts power saving by just going 2nm lower. It's not really alot, but it does offer some extra headroom to raise the bar in clocks and all that.

 

[R0H1T]

Your way off, it's (N5P) as compared to 5nm, not 5nm vs 7nm node ~

 

[Space Lynx]

What a fuss. This is just part of the contract(ual) deals AMD has with TSMC. 5+nm only saves like 7 percent of power based on the same clocks, or a 15% uplift in silicon performance. If we snatch 7% off 125(W) of TDP then compared to the 7nm proces your looking at a 8.75 watts power saving by just going 2nm lower. It's not really alot, but it does offer some extra headroom to raise the bar in clocks and all that.

lol I'll wait for benches rather than trust any common person's opinion on the internet.

 

[Jism]

lol I'll wait for benches rather than trust any common person's opinion on the internet.

It was quoted by TSMC itself. 7% lower power consumption and 15% better silicon performance. Silicon performance is something different then actual CPU performance. Silicon performance could best be expressed by simple more mhz.

 

[Space Lynx]

It was quoted by TSMC itself. 7% lower power consumption and 15% better silicon performance. Silicon performance is something different then actual CPU performance. Silicon performance could best be expressed by simple more mhz.

still can't predict fps increases in games... so... I'll wait for benches. thanks though

 

[watzupken]

If AMD is indeed using 5nm for Ryzen 4xxx, I think this will cause panic over at Intel. With 7nm, AMD is already causing quite a stir in the CPU space. If they gain further advantage in fab, it will take at least Intel to deliver 7nm to become competitive again. However this is only a rumor, so let's wait and see if this is going to be the case.

 

[efikkan]

Wasn't it just a few months ago AMD stated that Zen 3 will be launching in 2020 on 7nm? Things like this can't change so late in the game, just from a completed tapeout it usually takes over 1 year until launch, and a full node change takes ~2 years. So either AMD have been lying, or (more likely) this rumor is untrue.

With Zen 3 to be "launching" in 2020, and DDR5 arriving late 2022, there might be a room for a Zen 3 "refresh" in 2021 though.

I know people have been sitting at home too much lately, but let's not go crazy over every rumor.

 

[Tomorrow]

Wasn't it just a few months ago AMD stated that Zen 3 will be launching in 2020 on 7nm? Things like this can't change so late in the game, just from a completed tapeout it usually takes over 1 year until launch, and a full node change takes ~2 years. So either AMD have been lying, or (more likely) this rumor is untrue.

With Zen 3 to be "launching" in 2020, and DDR5 arriving late 2022, there might be a room for a Zen 3 "refresh" in 2021 though.

I know people have been sitting at home too much lately, but let's not go crazy over every rumor.

March 2, 2017 = Zen1 14nm
April 19, 2018 = Zen+ 12nm (+13 months)
July 7, 2019 = Zen2 7nm (+15 months)
* July 7, 2020 = Zen2+ 7Pnm (+12 months)
* March 3, 2021 = Zen3 5nm+ (+8 months compared to Zen2+ or +20 months compared to Zen2). With the original September 2020 release this would be +14months compared to Zen2.

Like i said earlier. If AMD were to move Zen3 to 5nm+ they would have know this much earlier than we did. And would have had time to prepare.
That being said Digitimes has been wrong before so it's not like im taking this as an absolute truth. Im trying to see if and how this is viable.

In the end the XT refresh might be the strongest evidence in the end if 5nm+ Zen3 ends up being true. It never made much sense to me to launch Zen2 refresh mere 2-3 months before Zen3's launch. But if Zen3 is pushed back to say end of Q1 2021 (31 March) it would make perfect sense as a stopgap.

Assuming AMD initiated this Zen3 change on January (but we learned this now) of this year then a 2021 Q1 launch would give them ~13 months complete and launch Zen3. I find this very much doable and in line what they have done before. Even a full node transition from one fab to another (this is even more radical than a full node change on the same fab) took them 15 months (and that was with a new architecture too) i can't see it being impossible to execute a smaller scale 7nm+ to 5nm+ redesign on the same fab within 13 months.

 

[efikkan]

Assuming AMD initiated this Zen3 change on January (but we learned this now) of this year then a 2021 Q1 launch would give them ~13 months complete and launch Zen3. I find this very much doable and in line what they have done before. Even a full node transition from one fab to another (this is even more radical than a full node change on the same fab) took them 15 months (and that was with a new architecture too) i can't see it being impossible to execute a smaller scale 7nm+ to 5nm+ redesign on the same fab within 13 months.

No way. They must have completed the tapeout by January this year, not initiated the process then.

The reason why they manage to release new products every 12-15 months is they have products in various stages of development simultainiously. Whenever they launch one product, they are taping out the next and developing a third. It is certainly possible to do multiple tapeouts of a singe architecture on various nodes, both AMD and Nvidia have done that in the past, but it must be planned ~2 years or more ahead, they can't change this "last minute".

 

[theGryphon]

If this happens and Zen3 comes in 5nm process, it's good for AMD in many fronts:
1) AMD has the upper hand in performance in pretty much everything but gaming. Move to 5nm will probably help them conquer that front.
2) Being able to squeeze more cores in same die area will not benefit MSDT or HEDT but will hugely boost confidence in the EPYC line. AMD still needs to convince some big players in this market and this move can achieve that.
3) Mobile would also benefit greatly. Smaller CPU die area / higher performance per watt / larger cache with larger GPU are all incredibly valuable here and again, AMD has many big players to win here.
4) Delayed arrival of Zen3 give a longer time to Zen2 owners to enjoy their investments before being dominated. I believe this is very good for the relations with the user base.

 

[TheoneandonlyMrK]

Makes some sense, use 7Nm for first consumer release zen 3 while also building zen3 on 5NmE at reduced initial ramp yeilds only for server.
Then a year later release an enhanced consumer zen3+ on 5 nm for consumer once yeilds improve.

What doesn't make sense is using a new chip design on a new node but they've got zen3 taped out on 7nm so they have translational data to make a shrink happen.

 

[medi01]

RTX 2080 Ti when equalised/normalised for the process is no more than 377 mm2

2080Ti is a 12nm 754mm2 chip.

Vega VII is a node shrink of a 14nm 495mm2 chip.

2080Ti is 35-39% faster (1080p-4k) than Vega VII.

The point is that AMD is not competitive in the graphics market.

The point is that people with mental capacity problems are posting on tech forums.

 

[ARF]

AMD Radeon VII 16 GB Review

The time has come. We're finally allowed to talk about Radeon VII performance numbers. The company's new flagship graphics card is the world's first to be made using a 7 nanometer production process. Also, it has the largest VRAM size of any card below $1000: 16 GB.

www.techpowerup.com

ASUS GeForce RTX 2080 Ti Matrix 11 GB Review

The ASUS RTX 2080 Ti Matrix is the world's first graphics card with fully integrated watercooling. ASUS has managed the seemingly impossible and crammed waterblock, pump, and radiator into a triple-slot graphics card. The Matrix is also highly overclocked, making it the fastest RTX 2080 Ti we...

www.techpowerup.com

OC vs OC RTX 2080 Ti is around 80% faster than the Radeon VII at least in one instance.

 

[TheoneandonlyMrK]

View attachment 157244

AMD Radeon VII 16 GB Review

The time has come. We're finally allowed to talk about Radeon VII performance numbers. The company's new flagship graphics card is the world's first to be made using a 7 nanometer production process. Also, it has the largest VRAM size of any card below $1000: 16 GB.

www.techpowerup.com

View attachment 157245

ASUS GeForce RTX 2080 Ti Matrix 11 GB Review

The ASUS RTX 2080 Ti Matrix is the world's first graphics card with fully integrated watercooling. ASUS has managed the seemingly impossible and crammed waterblock, pump, and radiator into a triple-slot graphics card. The Matrix is also highly overclocked, making it the fastest RTX 2080 Ti we...

www.techpowerup.com

OC vs OC RTX 2080 Ti is around 80% faster than the Radeon VII at least in one instance.

Ah right but that's an old dx11 app similar to E sport's.

I think you are picking idealistic examples though to prove a shit point you got wrong.

Nvidia have in recent years always used more die space and created a more expensive chip.
If they double down on RTX and tensor which they're sure to do they're getting bigger verses performance per watt not smaller.

And the fact they're harassing companies to try and get reduced foundry costs say's it all.
They know they can't compete on efficiency per mm on their designs or costs ie BOM at these price levels and if AMD can bring out a competitive part at a reasonable cost they're in a fix, they're beholden to investors who expect a certain level of growth.

Sorry about that wrong thread and person, I have slapped myself back awake now.

 

[medi01]

Let me cherry pick a couple of benchmark and make a pathetic post instead of being classy and admitting I was wrong.

Whatever makes you feel better, son.