AMD server roadmap, 28nm in H2 2014

boxleitnerb · 12-03-2012, 04:40 AM

A new AMD server roadmap has popped up:

28nm introduction seems quite late. I wonder how AMD thinks they can compete with 14nm CPUs then? If this is any indication for the desktop segment, it surely doesn't look good.

elitejp · 12-03-2012, 05:15 AM

ok ...serious question? Does shrinking the cpu really make any difference for a desktop? I mean theres more than enough real estate on the mobo and in the case to increase the size of the cpu

ShadowVVL · 12-03-2012, 05:22 AM

yeah if it takes amd unit 2014 for 28 nm it is bad.

ShintaiDK · 12-03-2012, 05:31 AM

Quote:

Originally Posted by elitejp

ok ...serious question? Does shrinking the cpu really make any difference for a desktop? I mean theres more than enough real estate on the mobo and in the case to increase the size of the cpu

You also get power savings. IB saw a 20% drop in power consumption with an unoptimized uarch for the process. And thats even with uarch improvements, higher speedbin and bigger iGPU. Not to mention the direct manufactoring cost savings.

The same timeframe on the Intel side is the release of Haswell-E/EP/EX on 22nm in H2 2014.

mrmt · 12-03-2012, 06:03 AM

Quote:

Originally Posted by elitejp

ok ...serious question? Does shrinking the cpu really make any difference for a desktop? I mean theres more than enough real estate on the mobo and in the case to increase the size of the cpu

I think you are confounding the IHS with the die, which aren't the same thing. The chip itself is more or less the size of a coin, or a paper clip in the case of IVB. Die size for the form factor isn't really the issue.

For the consumer and servers smaller dies brings less power consumption, which is far more important for the customers, as they are the critical variable for TCO and are the critical variable for the form factor.

For the manufacturer die size is important as the COGS is a direct function of wafer costs and yields.

What really caught my attention is that this roadmap isn't mentioning anything related to steamroller, something they should be doing if they had plans for the architecture. It seems that the rumors of cancellation of Steamroller and Kaveri were in part true. AMD did indeed cancel Bulldozer projects, but they are fielding something else.

sontin · 12-03-2012, 06:16 AM

I guess there was a reason at all that AMD is paying 1 Billion to get out of the agreement with GF...

Soulkeeper · 12-03-2012, 06:27 AM

I was reading about the loongson CPUs on wiki.
It seems they were planning 28nm 16 core, integrated imc, integrated graphics, 20 watt, etc. for 2012
Apparently the loongson platforms use VIA and AMD chipsets and have all the standard addons like realtech gigabit ethernet, sata, usb, etc.

Will this be a serious contender ?
possibly fabbed at global foundries ?

beginner99 · 12-03-2012, 06:37 AM

Quote:

Originally Posted by ShadowVVL

yeah if it takes amd unit 2014 for 28 nm it is bad.

I thought they skip 28 nm completely (AFAIK payed GF a large penalty for that). So these new server chips could be 20 nm or maybe the make the direct jumps to maybe TSMC 16nm finFET (which should be available in Q1 2014).

there is another thread somewhere with a link to an article. Basically certain companies believe 28 nm to 20 nm jump is not really worth it and might skip 20 nm. Considering that AMD has a lack of resources to quickly create new uArch it will take longer till the have steamroller and hence they just skip those proccess nodes. IMHO would make sense.

This would also explain why there is no SteamRoller APU in 2013. Looks like they design steamroller for 16nm finFET and hence also no APUs base son it before 2014.

Idontcare · 12-03-2012, 06:55 AM

Quote:

Originally Posted by boxleitnerb

A new AMD server roadmap has popped up:

28nm introduction seems quite late. I wonder how AMD thinks they can compete with 14nm CPUs then? If this is any indication for the desktop segment, it surely doesn't look good.

In fairness this is their server roadmap, and if you compare to Intel's server roadmap the process-node vs. release year timeline isn't so badly matched.

Intel only just recently released 32nm SB xeons. They won't be releasing 22nm XEONS until something like Q3 2013. And 14nm XEONs will not come any sooner than 2yrs after that.

ShintaiDK · 12-03-2012, 07:09 AM

Quote:

Originally Posted by Idontcare

In fairness this is their server roadmap, and if you compare to Intel's server roadmap the process-node vs. release year timeline isn't so badly matched.

Intel only just recently released 32nm SB xeons. They won't be releasing 22nm XEONS until something like Q3 2013. And 14nm XEONs will not come any sooner than 2yrs after that.

I think this is what you are looking for:

Roughly, 14nm Broadwell Xeons in summer 2015.

However, AMD will also only release 1P server CPUs in H2 2014 as well. And those are run on the desktop style timeschedule. Aka the Xeons v2 today. That sounds to me like the desktop AMD will also only be 28nm in 2014, competiting with 14nm Broadwell.

Tuna-Fish · 12-03-2012, 07:28 AM

Quote:

Originally Posted by Soulkeeper

I was reading about the loongson CPUs on wiki.
It seems they were planning 28nm 16 core, integrated imc, integrated graphics, 20 watt, etc. for 2012
Apparently the loongson platforms use VIA and AMD chipsets and have all the standard addons like realtech gigabit ethernet, sata, usb, etc.

Will this be a serious contender ?
possibly fabbed at global foundries ?

Those are MIPS cpus. They don't see much use outside really small niches. I doubt they could take significant share in anything that's not subsidized by the Chinese government.

ShintaiDK · 12-03-2012, 07:58 AM

Quote:

Originally Posted by Tuna-Fish

Those are MIPS cpus. They don't see much use outside really small niches. I doubt they could take significant share in anything that's not subsidized by the Chinese government.

They are making their way into routers. But thats mainly at the expense of ARM.

Soulkeeper · 12-03-2012, 08:37 AM

Quote:

Originally Posted by Tuna-Fish

Those are MIPS cpus. They don't see much use outside really small niches. I doubt they could take significant share in anything that's not subsidized by the Chinese government.

They are MIPs yes.
Many linux distros support them, they are apparently popular in china.
Also they have some kinda hardware that allows them to execute x86 binaries at 70% native speed according to wiki.

jones377 · 12-03-2012, 08:53 AM

This pretty much confirms the leaked desktop roadmap earlier. No steamroller in 2013.

beginner99 · 12-03-2012, 08:53 AM

Quote:

Originally Posted by beginner99

I thought they skip 28 nm completely (AFAIK payed GF a large penalty for that). So these new server chips could be 20 nm or maybe the make the direct jumps to maybe TSMC 16nm finFET (which should be available in Q1 2014).

there is another thread somewhere with a link to an article. Basically certain companies believe 28 nm to 20 nm jump is not really worth it and might skip 20 nm. Considering that AMD has a lack of resources to quickly create new uArch it will take longer till the have steamroller and hence they just skip those proccess nodes. IMHO would make sense.

This would also explain why there is no SteamRoller APU in 2013. Looks like they design steamroller for 16nm finFET and hence also no APUs base son it before 2014.

lol. I missed the very small legend at the bottom-right of the picture.

Ajay · 12-03-2012, 12:03 PM

Bizarre, AMD has 20nm in the legend (in red) but nothing in the roadmap

I can't believe that AMD thinks they can release something at 28nm in that time frame and expect anything but trivial server sales. If they seriously plan on staying in the big core x86 market at that time, they need to go to 20nm or go home. AMD needs to skip a half node.

I suppose the a big problem is that design costs for 28nm are already sunk. Wouldn't they be able to write off those costs completely if they weren't able to launch that product (I'm no accountant, so hopefully someone else will comment about the financial side).

I wonder if leaking a Roadmap is just a 'confidence' play?

Edit: didn't see beginner99's post for some reason.

ShintaiDK · 12-03-2012, 12:29 PM

Quote:

Originally Posted by Ajay

Bizarre, AMD has 20nm in the legend (in red) but nothing in the roadmap

The 20nm avaliable might only be useful for very low power devices in that timeframe.

Ajay · 12-03-2012, 01:11 PM

Quote:

Originally Posted by ShintaiDK

The 20nm avaliable might only be useful for very low power devices in that timeframe.

GF was going to have a 20nm SHP process, supposedly around that time frame, but I'm sure the LP/SLP would come first and GFs roadmaps wouldn't bother with such distinctions as they want to look like a top of the line fab. It's really sad that GF can't guarantee 20nm SHP sometime in 2014 - so AMD is put in a worse cash position by paying GloFlo not to make wafers for them

Idontcare · 12-03-2012, 01:29 PM

Quote:

Originally Posted by Ajay

Bizarre, AMD has 20nm in the legend (in red) but nothing in the roadmap

I can't believe that AMD thinks they can release something at 28nm in that time frame and expect anything but trivial server sales. If they seriously plan on staying in the big core x86 market at that time, they need to go to 20nm or go home. AMD needs to skip a half node.

I suppose the a big problem is that design costs for 28nm are already sunk. Wouldn't they be able to write off those costs completely if they weren't able to launch that product (I'm no accountant, so hopefully someone else will comment about the financial side).

I wonder if leaking a Roadmap is just a 'confidence' play?

Edit: didn't see beginner99's post for some reason.

More than likely it is an Excel embedded table within a powerpoint slide. I use them too. In that case they have chosen to format the embedded view for powerpoint to obscure the rest of the table, but if you had the power point file you would just double-click it and it would pop up in edit format and you'd see the rest (likely the 20nm comes in circa 2016).

But the legend is mean to be universal, so it doesn't take away the 20nm just because the view is obscured.

ShintaiDK · 12-03-2012, 03:01 PM

Quote:

Originally Posted by Ajay

GF was going to have a 20nm SHP process, supposedly around that time frame, but I'm sure the LP/SLP would come first and GFs roadmaps wouldn't bother with such distinctions as they want to look like a top of the line fab. It's really sad that GF can't guarantee 20nm SHP sometime in 2014 - so AMD is put in a worse cash position by paying GloFlo not to make wafers for them

Remember GF and TSMC uses factory roadmaps. Thats why they never match with product releases.

Ajay · 12-03-2012, 03:20 PM

Quote:

Originally Posted by Idontcare

More than likely it is an Excel embedded table within a powerpoint slide. I use them too. In that case they have chosen to format the embedded view for powerpoint to obscure the rest of the table, but if you had the power point file you would just double-click it and it would pop up in edit format and you'd see the rest (likely the 20nm comes in circa 2016).

But the legend is mean to be universal, so it doesn't take away the 20nm just because the view is obscured.

Ah, add another item to why I hate PowerPoint

Good point anyways.

Quote:

Originally Posted by ShintaiDK

Remember GF and TSMC uses factory roadmaps. Thats why they never match with product releases.

It's really amazing. I know that GF has already run various test packages @ 20nm, and read a rumor somewhere of them running some ARM SoCs (no word on how they came out) and yet we'll not see a high performance 20nm product till 2016

Seems like @ 16nm & > everyone will be using fewer processes/node and just use voltage to move things along the frequency curve.

AMD must be banking on a significantly improved economy in 2014/15 to get back into the business of running low volume server/desktop CPUs. They better have something really special ready...SR will show about as well as BD did. Maybe they'll actually be able to roll in EX. <sigh>

NostaSeronx · 12-03-2012, 03:35 PM

Fixed it for you guys.

AtenRa · 12-03-2012, 03:50 PM

They will stay five years in 32/28nm ???? no way.
They barely can compete with 32nm XEONs now at 32nm SOI, no way they will compete at 28nm BULK or SOI against 22nm FinFet in 2013-2014.

NostaSeronx · 12-03-2012, 04:09 PM

Quote:

Originally Posted by AtenRa

They will stay five years in 32/28nm ???? no way.
They barely can compete with 32nm XEONs now at 32nm SOI, no way they will compete at 28nm BULK or SOI against 22nm FinFet in 2013-2014.

Don't trust AMD Roadmaps so often particularly the Enterprise/Sever roadmaps.

mrmt · 12-03-2012, 05:13 PM

Quote:

Originally Posted by NostaSeronx

Don't trust AMD Roadmaps so often particularly the Enterprise/Sever roadmaps.

I agree with you that it is not a good idea to trust AMD roadmaps, but I don't think this leak should be completely discarded. Each roadmap revision AMD did brought bad news, we got either delays, or cancelled and watered down designs, never a revision for good. So this roadmap indicates the best case scenario for AMD.

I'm eager to see what AMD is going to field in 2014. When Intel launched SNB-EP Intel toasted AMD server business because the chip was much superior to Bulldozer and 32nm was already very mature and winding down on consumer so they could fill all the relevant price points. We are going to see the same in 2014 but in a much worse fashion. Intel launching Haswell-EP on a very mature 22nm process against whatever AMD throws at it on 28nm.