TSMC needs to get their act together...

[Fallen Kell]

I mean seriously, this is truly affecting the entire market. We are now a generation of GPU's behind because TSMC can't get things working properly. I really wish Intel would sell time on it's lines at least for non-CPU chips....

 

[ShintaiDK]

You blame the wrong company.

 

[3DVagabond]

You blame the wrong company.

I'll plat along.

Who should we be blaming?

 

[ShintaiDK]

I'll plat along.

Who should we be blaming?

You can blame nVidia and AMD for not being willing to pay.
You can blame Qualcomm, Apple etc for being willing to pay more with the consumers behind them.

However none of this is TSMCs fault.

Essentially its no different than with the x86 CPUs. Just replace GPUs with desktop CPUs.

 

[3DVagabond]

You can blame nVidia and AMD for not being willing to pay.
You can blame Qualcomm, Apple etc for being willing to pay more with the consumers behind them.

However none of this is TSMCs fault.

Essentially its no different than with the x86 CPUs. Just replace GPUs with desktop CPUs.

Everything I've read has blamed it on the process not being suitable for high performance GPU's.

 

[ShintaiDK]

Everything I've read has blamed it on the process not being suitable for high performance GPU's.

TSMC develops a node for those willing to pay.

 

[dacostafilipe]

You should also blame Samsung.

Read this: http://www.extremetech.com/extreme/...heft-give-samsung-its-14nm-manufacturing-lead

 

[3DVagabond]

TSMC develops a node for those willing to pay.

Your one liners aren't informative.

 

[ShintaiDK]

Your one liners aren't informative.

Because you seem to refuse the previous post.

nVidia and AMD is not willing to pay for a leading edge high performance node. Qualcomm, Apple and so on however is willing to pay for a leading edge low power node.

Had nVidia and AMD been willing to pay, you would have the node. But they are not, hence no node.

Its no secret that nVidia for example have been complaining about transistor cost for a long time.

 

[Borealis7]

i blame the entire Mobile sector for hogging all the wafers.

 

[3DVagabond]

Because you seem to refuse the previous post.

nVidia and AMD is not willing to pay for a leading edge high performance node. Qualcomm, Apple and so on however is willing to pay for a leading edge low power node.

Had nVidia and AMD been willing to pay, you would have the node. But they are not, hence no node.

Its no secret that nVidia for example have been complaining about transistor cost for a long time.

It's not that I refuse. You give no explanation and site no sources. Just forget about it.

 

[ThatBuzzkiller]

No, it makes no difference whether TSMC develops a high performance version or not since they have determined that there was hardly any differentiating characteristics ...

The real issue at hand is transistor cost scaling and I have a feeling that Intel is having a headache about that right now because I highly doubt that they've solved the problem ...

The whole IDM industry is waiting on EUV or another alternative ...

 

[ShintaiDK]

It's not that I refuse. You give no explanation and site no sources. Just forget about it.

http://www.nordichardware.com/Science-Technology/nvidia-20nm-from-tsmc-wont-be-cost-efficient.html

http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless

I doubt this is "news".

GPUs are the losers. Just like desktop CPUs. Something people can just as well get used to.

 

[desprado]

1 company should not be blamed.

 

[exar333]

Because you seem to refuse the previous post.

nVidia and AMD is not willing to pay for a leading edge high performance node. Qualcomm, Apple and so on however is willing to pay for a leading edge low power node.

Had nVidia and AMD been willing to pay, you would have the node. But they are not, hence no node.

Its no secret that nVidia for example have been complaining about transistor cost for a long time.

This.

Fab companies are not going to follow the 'if you build it they will come' philosophy. Companies don't just invest Billions (yes with a 'B') hoping to fill capacity.

Early 20nm was not as promising as hoped, so AMD/NV didn't invest further in that. So they wait for the next opportunity. I guarantee a significant amount of money was spent on 20nm, but the ROI wasn't there to continue the work further on a large-scale.

TSMC 20nm may make sense for smaller-die and lower freq parts (read: mobile) but didn't pan-out for high-performance. Just the way it was. I am sure the issue is more complex than just 'TSMC needs to get their act together'. We are seeing a lot of issues in sub-22nm nodes, across the board.

 

[Exophase]

Everything I've read has blamed it on the process not being suitable for high performance GPU's.

I wonder if this is really true though. GPU's aren't really very high performance devices, they're more very high throughput devices. Most of the difference between a low end GPU and a high end GPU is the high end having more of the low end's parts replicated. You don't need a better process to make that happen.

You can see this in Tegra X1, which is a TSMC 20nm part. It has two Maxwell SMMs, and appears to clock at 1GHz, which is not that much lower than high end desktop 28nm Maxwell parts. Even if they really couldn't get a higher clock speed, which I doubt, if the power consumption was lower enough then the improved density would more than make up for it.

The real reason I think nVidia and AMD haven't been eager to put GPUs on TSMC 20nm is because it doesn't offer enough of a performance/power consumption improvement over 28nm, and they knew 16FF/16FF+ would be available not much longer. They wouldn't be using 20nm for long enough to amortize the cost of development and outweigh the initial very expensive process costs. While in mobile there's more pressure to maintain a very aggressive development schedule because of competition (eg from Intel), and there's a higher volume.

 

[raghu78]

You can blame nVidia and AMD for not being willing to pay.
You can blame Qualcomm, Apple etc for being willing to pay more with the consumers behind them.

However none of this is TSMCs fault.

Essentially its no different than with the x86 CPUs. Just replace GPUs with desktop CPUs.

rubbish. the TSMC 20nm node is just not suitable for high performance CPU and GPUs. Because even though transistor density is almost 2x higher than TSMC 28nm like a traditional full process node the transistor speed gain and power efficiency gain is minimal and more like a half node.

Perf at the same power (15% higher)

TSMC 20nm - 1.15x
TSMC 28nm - 1x

Power at the same perf (30% lower power)

TSMC 28nm - 1x
TSMC 20nm - 0.7x

With a traditional full process node gain we are used to getting a 40-45% speed gain at same power or a 50- 55% lower power at same perf. This is how TSMC 28nm HP was in relation to TSMC 40G. The primary reason for the above problem is planar transistor has run its course and is completely incapable of providing further speed and efficiency gains. For that FINFET is required.

The TSMC 16FF and 16FF+ are what we call a full process node leap from 28nm. 16FF+ is a massive leap as it improves on 16FF and provides a 70% power reduction over TSMC 28nm at same perf or a 65% speed gain at same power.

http://www.tsmc.com/english/dedicatedFoundry/technology/16nm.htm

This is why AMD and Nvidia have chosen to skip 20nm as it does not have the process node characteristics (transistor perf gain, power efficiency gain) required to manufacture power hungry and massive high performance GPUs. The chips manufactured at TSMC 20nm are all sub 5W and sub 150 sq mm chips. TSMC 20nm is just not suitable for 300 sq mm and 200W GPUs. We will get those next year at TSMC 16FF+ .

 

[Arachnotronic]

rubbish. the TSMC 20nm node is just not suitable for high performance CPU and GPUs. Because even though transistor density is almost 2x higher than TSMC 28nm like a traditional full process node the transistor speed gain and power efficiency gain is minimal and more like a half node.

Perf at the same power (15% higher)

TSMC 20nm - 1.15x
TSMC 28nm - 1x

Power at the same perf (30% lower power)

TSMC 28nm - 1x
TSMC 20nm - 0.7x

With a traditional full process node gain we are used to getting a 40-45% speed gain at same power or a 50- 55% lower power at same perf. This is how TSMC 28nm HP was in relation to TSMC 40G. The primary reason for the above problem is planar transistor has run its course and is completely incapable of providing further speed and efficiency gains. For that FINFET is required.

The TSMC 16FF and 16FF+ are what we call a full process node leap from 28nm. 16FF+ is a massive leap as it improves on 16FF and provides a 70% power reduction over TSMC 28nm at same perf or a 65% speed gain at same power.

http://www.tsmc.com/english/dedicatedFoundry/technology/16nm.htm

This is why AMD and Nvidia have chosen to skip 20nm as it does not have the process node characteristics (transistor perf gain, power efficiency gain) required to manufacture power hungry and massive high performance GPUs. The chips manufactured at TSMC 20nm are all sub 5W and sub 150 sq mm chips. TSMC 20nm is just not suitable for 300 sq mm and 200W GPUs. We will get those next year at TSMC 16FF+ .

...and yet, Oracle has no problems building >3GHz, large-die SPARC CPUs on 20nm.

http://www.linleygroup.com/newsletters/newsletter_detail.php?num=5219

 

[therealnickdanger]

...and yet, Oracle has no problems building >3GHz, large-die SPARC CPUs on 20nm.

http://www.linleygroup.com/newsletters/newsletter_detail.php?num=5219

And how much do each of those cost...?

 

[ThatBuzzkiller]

...and yet, Oracle has no problems building >3GHz, large-die SPARC CPUs on 20nm.

http://www.linleygroup.com/newsletters/newsletter_detail.php?num=5219

This ...

People here keep forgetting that the biggest issue isn't the performance characteristics offered by TSMC's 20nm node, it's the cost scaling ...

 

[raghu78]

...and yet, Oracle has no problems building >3GHz, large-die SPARC CPUs on 20nm.

http://www.linleygroup.com/newsletters/newsletter_detail.php?num=5219

That chip has not yet launched.

http://www.theregister.co.uk/2014/08/18/oracle_reveals_32core_10_beeellion_transistor_sparc_m7/

btw thats not a high volume product like a consumer GPU. That server CPU is a low volume very high margin product where Oracle can afford to eat low yields. Thats not the case for AMD and Nvidia which have to sell the chips at a much lower price and need good yields with product volume which is more than an order of magnitude higher than what Oracle Sparc M7 will ever sell.

 

[ThatBuzzkiller]

That chip has not yet launched.

http://www.theregister.co.uk/2014/08/18/oracle_reveals_32core_10_beeellion_transistor_sparc_m7/

btw thats not a high volume product like a consumer GPU. That server CPU is a low volume very high margin product where Oracle can afford to eat low yields. Thats not the case for AMD and Nvidia which have to sell the chips at a much lower price and need good yields with product volume which is more than an order of magnitude higher than what Oracle Sparc M7 will ever sell.

Doesn't change the fact that Arachnotronic proved otherwise about your initial claim ...

 

[Arachnotronic]

That chip has not yet launched.

http://www.theregister.co.uk/2014/08/18/oracle_reveals_32core_10_beeellion_transistor_sparc_m7/

btw thats not a high volume product like a consumer GPU. That server CPU is a low volume very high margin product where Oracle can afford to eat low yields. Thats not the case for AMD and Nvidia which have to sell the chips at a much lower price and need good yields with product volume which is more than an order of magnitude higher than what Oracle Sparc M7 will ever sell.

You think that 20nm yields are worse than 16FF+ yields?

I'm really not sure what point you're trying to make.

 

[raghu78]

You think that 20nm yields are worse than 16FF+ yields?

I'm really not sure what point you're trying to make.

I will quote Joe Macri of AMD who is their product CTO as to why 20nm is unsuitable for high performance devices.

http://www.theregister.co.uk/Print/2014/01/14/amd_unveils_kaveri_hsa_enabled_apu/

Kaveri is baked in a 28-nanometer, planar, bulk silicon process, which is nowhere near as efficient as state-of-the-art FinFET (what Intel calls "Tri-Gate") or even the less-than-TriGate, more-than-bulk – and somewhat expensive – silicon-on-insulator (SOI) process that was used in Kaveri's predecessor.

There were reasons to go with 28nm rather than 22nm, Macri told us, that were discovered during the design process. That process was run by what he identified as a "cross-functional team" composed of "CPU guys, graphics guys, mixed-signal folks, our process team, the backend, layout team."

That cross-functional crew identified a boatload of process variants, and members of the team each ran tests based on their areas of interest, examining such factors as power curves and die-area needs.

"What we found was with the CPU with planar transistors, when we went from 28 to 22, we actually started to slow down," he said, "because the pitch of the transistor had to become much finer, and basically we couldn't get as much oomph through the transistor."

The problem, he said, was that "our IDsat was unpleasant" at 22nm, referring to gate drain saturation current*. In addition, the chip's metal system needed to be scaled down to fit within the 22nm process, which increased resistance.

"So what we saw was the frequency just fall off the cliff," he said. "This is why it's so important to get to FinFET."

the register article also links to a crash course in silicon gate saturation current and related electronic engineering

http://www.eecs.berkeley.edu/~hu/Chenming-Hu_ch6.pdf

btw I am sure Macri was talking of 20nm because there is no foundry with a 22nm process. If you still disagree thats fine but there is enough evidence that TSMC 20nm is not a high performance process. If you just looked at the process flavours and naming you should get a clue. TSMC has only variant of its 20nm process called 20SoC. TSMC 28 nm started off with 28HP (high performance with high k metal gates) and 28LP (low performance polysilicon).

http://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&newsid=6181

Today TSMC 28nm has over half a dozen variants

https://www.semiwiki.com/forum/content/4530-tsmc-unleashes-aggressive-28nm-strategy.html?

20SoC is a short lived node and going to be quickly replaced by 16FF+. So for high performance (at a much higher cost) customers will go to 16FF+ and for cost sensitive requirements 28 nm with its various variants will serve the market for years to come.

 

[ocre]

raghu78,

no offense but what you posted is completely worthless to this discussion.

you cant post something about some unknown 22nm testing and then swap out the entire subject to insert TSMCs 20nm as you please.

If the author is that mixed up on his facts, i would say disregard every last word.

If Marci was that mixed up when he did the interview, then...............i will not even say how messed up that is.

I think there is a much better chance that this is a completely unrelated conversation than some sort of mix up. Like i said, a mix up to that degree where the subject matter itself is lost, why listen to any of the details?