Wow, IBM creates 7nm chips!

[brainhulk]

https://www.yahoo.com/tech/s/ibm-computer-chips-obliterate-everything-else-market-154808759.html

Wait, I don't care anymore. I have two i5/i7 + gtx 780 computers i hardly even use... :'(

 

[Markbnj]

https://www.yahoo.com/tech/s/ibm-computer-chips-obliterate-everything-else-market-154808759.html

Wait, I don't care anymore. I have two i5/i7 + gtx 780 computers i hardly even use... :'(

It matters a lot for mobile SOC packages and power consumption. Pretty soon the damned things won't even exist in the physical world. We'll just have to take their word for it.

 

[rh71]

come on stock price...

 

[Sonikku]

Benchmark it and we'll talk.

 

[dave_the_nerd]

Finally! The Powerbook G5 will be a reality!

 

[IronWing]

When they get down to 2nm they can have Mork do the ads. Oh, :'(.

 

[mmntech]

It matters a lot for mobile SOC packages and power consumption. Pretty soon the damned things won't even exist in the physical world. We'll just have to take their word for it.

I'd say this matters even more for data centres. A powerful chip that sips electricity and doesn't output a lot of heat could be a big game changer for large scale enterprise systems. Which are still IBM's bread and butter.

 

[FerrelGeek]

Read the article on Ars. It's amazing that the tech keeps finding a way to go smaller.

And yes, this could be a big deal for data centers.

 

[Sho'Nuff]

Article makes some wild claims.

E.g., "Obviously, 7nm chips would be even faster than today’s 14nm processors and next year’s 10nm CPUs. Comparatively, a strand of DNA measures 2.5nm while a red blood cell has a 7,500nm diameter."

Maybe I'm off base. But to my knowledge feature size is just one component of processor speed. http://www.extremetech.com/computin...its-of-smaller-processes-and-new-foundry-tech

And in this case, we don't even know what features IBM is claiming to have manufactured at the 7nm scale. Are we talking about source drain distance or some other dimension?

As for the claim that the chips are 4 times faster than what is currently on the market - that may be true but I will believe it when I see it. Particularly as the baseline for that performance claim (i.e., what was actually measured) was not provided. 4X faster transistor switching speed, while impressive, does not necessarily compute to a processor that is, overall, 4x faster than its closest competition.

Edit: ARS report is better and provides more detail - http://arstechnica.co.uk/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

 

[Rakehellion]

Finally! The Powerbook G5 will be a reality!

I understood that reference.

 

[Rakehellion]

Weren't they making some 16-core 64-threaded server CPU?

 

[Dirigible]

IBM is really good at press releases. I'll care once they're selling product.

 

[KeithTalent]

Bah, probably can't even taste them at that size.

KT

 

[Cogman]

Article makes some wild claims.

E.g., "Obviously, 7nm chips would be even faster than today’s 14nm processors and next year’s 10nm CPUs. Comparatively, a strand of DNA measures 2.5nm while a red blood cell has a 7,500nm diameter."

Maybe I'm off base. But to my knowledge feature size is just one component of processor speed. http://www.extremetech.com/computin...its-of-smaller-processes-and-new-foundry-tech

And in this case, we don't even know what features IBM is claiming to have manufactured at the 7nm scale. Are we talking about source drain distance or some other dimension?

As for the claim that the chips are 4 times faster than what is currently on the market - that may be true but I will believe it when I see it. Particularly as the baseline for that performance claim (i.e., what was actually measured) was not provided. 4X faster transistor switching speed, while impressive, does not necessarily compute to a processor that is, overall, 4x faster than its closest competition.

Edit: ARS report is better and provides more detail - http://arstechnica.co.uk/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

Yup.

Basically what a smaller node buys you is higher transistor density and lower power consumption/per transistor (usually). It doesn't generally say much if anything about performance. Without architectural changes, the only way going to a lower node helps performance is it allows for higher clock speeds (usually). But that only buys you so much.

 

[poofyhairguy]

Wait, I don't care anymore.

I do. Smartwatch with three day battery life and slim profile, yes please.

 

[preslove]

Meh, my money is on IBM's cell processor revolutionizing the server space...

 

[CountZero]

Article makes some wild claims.

E.g., "Obviously, 7nm chips would be even faster than today’s 14nm processors and next year’s 10nm CPUs. Comparatively, a strand of DNA measures 2.5nm while a red blood cell has a 7,500nm diameter."

Maybe I'm off base. But to my knowledge feature size is just one component of processor speed. http://www.extremetech.com/computin...its-of-smaller-processes-and-new-foundry-tech

And in this case, we don't even know what features IBM is claiming to have manufactured at the 7nm scale. Are we talking about source drain distance or some other dimension?

As for the claim that the chips are 4 times faster than what is currently on the market - that may be true but I will believe it when I see it. Particularly as the baseline for that performance claim (i.e., what was actually measured) was not provided. 4X faster transistor switching speed, while impressive, does not necessarily compute to a processor that is, overall, 4x faster than its closest competition.

Edit: ARS report is better and provides more detail - http://arstechnica.co.uk/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

They make a lot of assumptions given it is also a new material being used. It could be the same speed as 10nm but smaller allowing for physically shorter datapaths which would appear faster but not really be faster at the device level.

You hit a good question on what feature is 7nm. Since we hit finFet the feature size has become more hand wavey then it was for planar. 22, 16, 14, 10, 7 are practically marketing terms anymore. I don't doubt something is 7nm but what it is they haven't said.

What I found most interesting was that it used EUV and quadruple masking. The costs have to be through the roof. My understanding was that EUV would help so that we didn't have to do double/triple patterning let alone quadruple patterning but the EUV costs are currently so high and the technology so touchy that we've continued down the path of double/triple patterning. Closing DRC on quadruple patterns should be a lot of fun...

 

[BoberFett]

Yup.

Basically what a smaller node buys you is higher transistor density and lower power consumption/per transistor (usually). It doesn't generally say much if anything about performance. Without architectural changes, the only way going to a lower node helps performance is it allows for higher clock speeds (usually). But that only buys you so much.

The clock speed wars are over. It's all about parallelism now (task dependent obviously) and smaller processes means the ability to pack more on a chip.

 

[thecoolnessrune]

Weren't they making some 16-core 64-threaded server CPU?

The POWER8 CPU architecture was released a year ago, and is 8-way multi-threaded. It's top chip contains 12 cores and 96 threads.

You can get a mid-range scale out server containing 2 such beasts for the low starting price of $37,000 from IBM

 

[Cogman]

The clock speed wars are over. It's all about parallelism now (task dependent obviously) and smaller processes means the ability to pack more on a chip.

So, the thing is, we aren't getting much from packing on more. Already 90% of most processors transistor count is dedicated to the various cache layers. Adding more cache really don't buy you more performance, it is more of a "Well, shoot, we don't know what to do with all of these extra transistors. Might as well build out the cache".

Ditching clock speed for parallelism was basically the processor manufactures saying "Crap, we don't want to just increase cache sizes, what can we do? MOER CORES!". For 2, 4, and even 8 cores that is an appealing use of the extra transistors that smaller nodes gain. However, most applications struggle to effectively use those extra cores. More only buys the average consumer so much. This is why we aren't seeing the 128 core cpu's we were promised back in 2005.

We also ditched clock speed because we simply couldn't grab those higher clock speeds. We start running into several physical limitations of silicon transistors, heat production, etc. that abruptly ended the ghz war.

Now, the focus isn't on cores or clock speed. It is on energy efficiency. Performance tends to go up because energy efficient CPUs tend to have a little more headroom when it comes to clock speed (a 1ghz cpu running at 15W can probably run just fine at 2ghz and 45W) however it isn't the focus. This is part of the reason why we are seeing such small performance gains with each new generation of intel CPUs.

Yay hitting physical limitations! It will be interesting to see how the industry will evolve now that we are hitting some very real barriers to general performance increases.

 

[ultimatebob]

come on stock price...

It hasn't gone up in years. I should have dumped it all when I left.

 

[AMCRambler]

The industry has been working on EUV lithography technology for a long time. After 20nm processes are tapped out, EUV becomes necessary to achieve the next process. Very interesting stuff.

My brother works for ASML as a technician on their current Twin Scan immersion systems. It's fun talking to him about what's next in lithography tech. I guess ASML has been working on developing commercial production EUV machines since 2009.

The way the EUV light is created is pretty cool science too. They hit molten tin particles with a beam from a CO2 laser and that tin particle emits EUV wavelength light. The challenge becomes directing that light where you want it to go. In addition I guess the interior of the machine needs to have a complete vacuum as even particles of air will block or absorb the EUV light. Inside the machines the light is collected by ultrasmooth mirrors that can have no defects that would deform the light.

On top of all that science this article talks about what needs to go into the fabs that will be built to run this technology. Even cleaner clean rooms, vibration dampening for the buildings and equipment, the list goes on. It sounds enormously cost prohibitive. Getting this technology into production facilities is going to be a huge feat of engineering. All so we can get new gadgets that are even thinner, faster and with a longer battery life.

 

[n7]

Bah, probably can't even taste them at that size.

KT

:awe:

 

[Kaido]

Weren't they making some 16-core 64-threaded server CPU?

The 15-core is already out. Free shipping too! :awe:

http://www.amazon.com/IBM-E7-4890-Pentadeca-core-Processor-Upgrade/dp/B00ONT0KCM/

 

[techs]

The 15-core is already out. Free shipping too! :awe:

http://www.amazon.com/IBM-E7-4890-Pentadeca-core-Processor-Upgrade/dp/B00ONT0KCM/

GaGH. I'll wait till thy are under 8 grand.
Or just steal them from the Federation.