Intel CEO Pat Gelsinger: I hope to build chips for Lisa Su and AMD

[starsmine]

9 hours ago, porina said:

That applies to all fabs. The numbers have not represented a physical feature size for a long time.

But they mean something explicit still. They mean a specified set of parameters, which a large number of them are physical size, just not the length anymore. 

A transistor has length, width, tox thickness, with finfet, fin height, and fin pitch.
just because none of them match specifically with the number, the overall size does shrink to make the overall density kind of equivalent to if it was a planar transistor at that gate length.

And with GAAFET, even more parameters. 
so yes, 7nm, 5nm, 4nm are all marketing names. but like Henry said, converting 20A to 2nm is just confusing, sure they equal each other, but they dont mean the same thing in this context. 

Angstrom is a made up non SI unit to begin with. And intel, (as well as TSMC) dropped the nm naming schemas

[porina]

Maybe I'm missing something here. I don't get the confusion unless some are just overcomplicating things.

 

For example, if I were to say 3nm in this context, we could take it to mean any of TSMC N3, or Intel 3, or likely more. Moving to Angstrom makes a kind of sense once we get to low nm, and the currently accepted usage of 10A = 1nm isn't exactly difficult to comprehend.

[leadeater]

15 hours ago, starsmine said:

And with GAAFET, even more parameters. 
so yes, 7nm, 5nm, 4nm are all marketing names. but like Henry said, converting 20A to 2nm is just confusing, sure they equal each other, but they dont mean the same thing in this context. 

How is this any different to before? TSMC N7 isn't the same as Samsung 7LPP or even TSMC N7P etc etc. Some '7nm" processes use DUV and some use EUV etc

 

Intel 10nm became Intel 7 because it was most similar to what other industry Fabs were calling their "7nm" processes.

 

Intel 18A will be similar to whatever TSMC calls theirs which will be whatever they decide it is. The reason to moving to different naming is N1.8 or 1.8LPP etc just sounds and looks bad. It's "just a name" as was everything since like 45nm.

 

15 hours ago, starsmine said:

kind of equivalent to if it was a planar transistor at that gate length.

Not so much anymore or for a while, that's been diverging away from being comparable to the "nm number" for a while now. TSMC N3 may be nothing at all equivalent to actual 3nm if that were possible.

[HenrySalayne]

5 hours ago, leadeater said:

How is this any different to before? TSMC N7 isn't the same as Samsung 7LPP or even TSMC N7P etc etc

 

Intel 10nm became Intel 7 because it was most similar to what other industry Fabs were calling their "7nm" processes.

 

Intel 18A will be similar to whatever TSMC calls theirs which will be whatever they decide it is. The reason to moving to different naming is N1.8 or 1.8LPP etc just sounds and looks bad. It's "just a name" as was everything since like 45nm.

The very same argument can be made for calling RAM speed Megahertz instead of Megatransfers. A widely accepted wrong use of terms doesn't make it any better.

 

And Intel is in my opinion especially to blame here. Renaming all their nodes to "surpass" the competition by name is adding even more malicious confusion. TSMC has claimed that N3 derivates are competing with Intel 18A. This has to be seen, but I honestly don't think we should feed the marketing departments by calling these nodes 3 nm and 1.8 nm respectively.

Tom's Hardware shows everything that's atrocious with this particular terminology in this article:

https://www.tomshardware.com/news/tsmc-our-3nm-node-comparable-to-intels-18nm-tech

 

"3 nm class" and "1.8 nm class"? What the hell is a "X nm class" if it's all arbitrary made up names? Intel 18A is in a better "class" but has worse technology, because they are beaten by TSMC's lower class 3 nm tech?

This is a good example of unnecessary and harmful segmentation and categorisation. It adds nothing for the reader. On the contrary, it breathes life into outdated concepts. Why the author felt the need to add this is beyond me. It doesn't serve the article, it doesn't educate readers, it just adds another layer of confusion preventing readers from actually diving into the intricacies.

[Godlygamer23]

1 hour ago, HenrySalayne said:

The very same argument can be made for calling RAM speed Megahertz instead of Megatransfers. A widely accepted wrong use of terms doesn't make it any better.

Using frequency isn't incorrect though. A certain number of cycles are completed per second, with the rising and falling part of the clocks being utilized for data transfer. 

[HenrySalayne]

1 hour ago, Godlygamer23 said:

Using frequency isn't incorrect though. A certain number of cycles are completed per second, with the rising and falling part of the clocks being utilized for data transfer. 

It is incorrect! The definition of frequency is pretty certain about periodicity. One clock cycle is one period. A coin doesn't magically have twice the value because it has two sides...

DDR4-3200 has 3200 MT/s but a frequency of 1600 MHz. With future technologies (we already see in GPUs) the discrepancy might even be a factor of four or higher.

[leadeater]

3 hours ago, HenrySalayne said:

And Intel is in my opinion especially to blame here. Renaming all their nodes to "surpass" the competition by name is adding even more malicious confusion. TSMC has claimed that N3 derivates are competing with Intel 18A. This has to be seen, but I honestly don't think we should feed the marketing departments by calling these nodes 3 nm and 1.8 nm respectively.

Tom's Hardware shows everything that's atrocious with this particular terminology in this article:

https://www.tomshardware.com/news/tsmc-our-3nm-node-comparable-to-intels-18nm-tech

 

"3 nm class" and "1.8 nm class"? What the hell is a "X nm class" if it's all arbitrary made up names? Intel 18A is in a better "class" but has worse technology, because they are beaten by TSMC's lower class 3 nm tech?

This is a good example of unnecessary and harmful segmentation and categorisation. It adds nothing for the reader. On the contrary, it breathes life into outdated concepts. Why the author felt the need to add this is beyond me. It doesn't serve the article, it doesn't educate readers, it just adds another layer of confusion preventing readers from actually diving into the intricacies.

As I said in my post they have always been widely different across all Fabs and even from the same fab, there is no and has never been "7nm" class, "5nm" class, "3nm" class. They are ALL lies and aren't like some impossible to achieve theoretical process technology if it were done the old way etc.

 

Intel absolutely is not to blame here at all. TSMC, GloFo, Samsung all relaxed their "standards" on what is an equivalent process node to some hypothetical node size while Intel did not so there you have 3 companies more to blame than Intel. Intel very rightly and justly as a business aligned with this now entirely accepted even more foundation of lies and renamed their Intel 10nm node to Intel 7 aligning their lies with everyone else's lies.

 

The reality is Intel is perfectly fine here changing their node names, again. Pushing down a naming path that is already broken and dumb resulting in not all that nice to read and present is not a good idea. Those that have some idea about fabrication and nodes won't have any problem looking between companies, comparing and discussing them.

 

The whole situation was already completely irrelevant in just purely looking at "nm" since a 7nm fab process that used EUV is vastly better than one that used DUV and a 5nm DUV node might not be better in certain aspects than a 7nm EUV node. If you are actually truly interested in fabrication nodes, assessing and discussing them then nothing at all has changed or changed before.

 

Just be thankful the NAND manufacturers didn't persist with equivalency naming like the silicon fabricators did, one giant mess avoided.

 

Also TSMC can claim whatever they like, doesn't actually mean anything and they will always say their stuff is better for whatever reason they think, and they should. You think TSMC is going to say "Yea Intel's XYZ is going to be better than our current and future XYZ"? Never going to happen.

[starsmine]

1 hour ago, Godlygamer23 said:

Using frequency isn't incorrect though. A certain number of cycles are completed per second, with the rising and falling part of the clocks being utilized for data transfer. 

I love that you disproved yourself.
exactly a certain number of cycles is completed per second. 
frequency is cycles per second.

Transfers happen twice per cycle. 

AKA 3000Mhz at double data is 6000MT/s

[Godlygamer23]

3 hours ago, HenrySalayne said:

It is incorrect! The definition of frequency is pretty certain about periodicity. One clock cycle is one period. A coin doesn't magically have twice the value because it has two sides...

DDR4-3200 has 3200 MT/s but a frequency of 1600 MHz. With future technologies (we already see in GPUs) the discrepancy might even be a factor of four or higher.

3 hours ago, starsmine said:

I love that you disproved yourself.
exactly a certain number of cycles is completed per second. 
frequency is cycles per second.

Transfers happen twice per cycle. 

AKA 3000Mhz at double data is 6000MT/s

"RAM speed" is pretty generic, so it could've gone in either direction.

 

If you're looking at how memory manufacturers advertise their products, Newegg advertises it as a unitless figure, but it is interpreted as megatransfers/second. If you go directly to the websites for G.Skill and Corsair(as examples), they both define the transfer rate as MT/s, not frequency.

 

https://www.gskill.com/specification/165/390/1662622062/F5-5600J2834F16GX2-TZ5NR-Specification

https://www.corsair.com/us/en/p/memory/cmh32gx5m2b5200c40/vengeance-rgb-32gb-2x16gb-ddr5-dram-5200mhz-c40-memory-kit-black-cmh32gx5m2b5200c40

 

Any misuse of the terms in the case of memory speed is down to users, not manufacturers. 

[starsmine]

1 hour ago, Godlygamer23 said:

"RAM speed" is pretty generic, so it could've gone in either direction.

 

If you're looking at how memory manufacturers advertise their products, Newegg advertises it as a unitless figure, but it is interpreted as megatransfers/second. If you go directly to the websites for G.Skill and Corsair(as examples), they both define the transfer rate as MT/s, not frequency.

 

https://www.gskill.com/specification/165/390/1662622062/F5-5600J2834F16GX2-TZ5NR-Specification

https://www.corsair.com/us/en/p/memory/cmh32gx5m2b5200c40/vengeance-rgb-32gb-2x16gb-ddr5-dram-5200mhz-c40-memory-kit-black-cmh32gx5m2b5200c40

 

Any misuse of the terms in the case of memory speed is down to users, not manufacturers. 

Im dead

But im not sure what you are arguing here either. Yes, its the users (and often media/retailers) the screw it up and make things confusing. 

5200Mhz ddr5 is possible and is coming before the end of its life cycle. here we have the benifite of the doubt to understand that its a typo.
But what about 1600Mhz DDR4? 

[Godlygamer23]

2 minutes ago, starsmine said:

Im dead

I'm curious how many people really pay attention to links. I read the product pages. I wasn't peeping for issues within the link, in this case.

 

So exactly what is the problem? With how manufacturers treat memory, or how users treat it? 

 

Since you decided to edit post as I posted mine...

 

The issue is that in the case of lithography, manufacturers are absolutely at fault, at least on the marketing side. I don't know how engineers view it. In the case of memory, which is what the comparison was...it's caused by users. The lithographic manufacturers need to change how they represent their products, but there's nothing wrong with memory is being represented by manufacturers. It is a user problem.

[starsmine]

7 minutes ago, Godlygamer23 said:

I'm curious how many people really pay attention to links. I read the product pages. I wasn't peeping for issues within the link, in this case.

 

So exactly what is the problem? With how manufacturers treat memory, or how users treat it? 

1600Mhz DDR4 are you talking about 3200MT/s or 1600MT/s?

Micron plans to release 12800 MT/s DDR5 in 2026, so anyone who says 6400MHz when they mean MT/s, aka DDR5 ram at 5200MHz will exist to. 

that's why users and retailers and media using MHz is confusing, we know they are often using it wrong. 

same way anyone who WRONGLY uses 2k to talk about 1440p screens when 2k has always meant 2048x1080p. aka, 1080p, aka NOT 1440p. 

[Godlygamer23]

4 minutes ago, starsmine said:

1600Mhz DDR4 are you talking about 3200MT/s or 1600MT/s?

Micron plans to release 12800 MT/s DDR5 in 2026, so anyone who says 6400MHz when they mean MT/s

thats why users and retailers and media usign MHz is confusing. 

From my small search on Newegg, it appears that Newegg has everything down, and advertises it the way it's supposed to be. However, I just did a quick search on Amazon, and went to a kit of DDR5 memory from Corsair, and found that the listing is incorrect.

 

Within the item name, I see the speed advertised, but instead of it being a unitless figure, it has "MHz" on the end, and they state it again as you scroll down. If you scroll down further, they list the speed once again, but as a unitless figure. That is totally unacceptable, and it's worse because it's Corsair's page, and they should be advertising things correctly, and if the reasoning is the typical "bigger number is better", that's even worse because it's deliberate misrepresentation of the product.

 

So Amazon definitely has issues. Not sure about other sites though, such as Walmart or Best Buy.

 

Corsair kit for reference: https://www.amazon.com/CORSAIR-VENGEANCE-5600MHz-Compatible-Computer/dp/B09NCNF2ZQ/ref=sr_1_2?crid=19BIWQYMUQNO1&dib=eyJ2IjoiMSJ9.VWoWKTcGSa9x7pE9sEg3ZBlNjsRYQaSjLecCE3lWDJbFlpEKdaHqNZNy35nLWRFR6zcc5GhxpJAMPmp23KOTFvdM-71yg6ZM6Z2G-ZwCR4lhDmt_sZRoGh8_DaZUwrbIMMpg85GOhbKtRJOYitxez7SD6ih71gdgxUJAKu7V2cG6mQofDwzwdkxNzOTumLKO6n-Md3xwSThM1_Z8bvYRvE0kWNJxub20okXXowbjXJA.WuThr-7LRMxm_hMSSQ53Q7cNtyApweVLSWuTaHNtkqI&dib_tag=se&keywords=corsair%2Bvengeance%2Bddr5&qid=1709080825&sprefix=corsair%2Caps%2C79&sr=8-2&th=1