Intel's Future Chips: News, Rumours & Reviews

Page 147 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
Don't worry, it won't be that long before we have organic computing with neural networks. Then all of this linear versus quantum stuff can go right out the window and our computers can be just as messed up as we are, with full neurosis, neurotic behaviors and other mental conditions. Some days your system will just tell you "I don't feel like gaming today. I haz sad." LOL.
 
  • Like
Reactions: RodroX

InvalidError

Titan
Moderator
Don't worry, it won't be that long before we have organic computing with neural networks.
The problems with bio-computers are very slow propagation time (the human brain has a "tick rate" estimated in the neighborhood of 100Hz based on how activity goes around the brain when a stimumus triggers activity) and less-than-exact accuracy, so there is no danger of organic computing replacing conventional computing in most applications. I would expect them to be particularly well-suited for coping with the analog world in ways digital computers cannot.

Quantum computers have a few challenges of their own, starting with manufacturing and programming. The biggest of which regarding hypothetical use as a substitute for conventional computing is that quantum machines don't naturally do sequential operations.
 
Something will have to give. Silicon as a base material is about used up. If Intel is having these kind of issues now, just trying to reach 7nm, wait until they try 5nm. They'll have to figure a way for cpus to run at 1.0v or less, just to stop physical burnout and bleeding. Which won't happen anytime soon. So when game devs want to push BF10 on 16k monitors at 600Hz, it's going to take cores, not speed, to get the job done. Short code strings, fast IPC, multiple threads.

Or go backwards and take half the stuff back out of the cpu, like the memory controller, igpu etc and free up i/o. Modular cpus, multiple sockets for different performance levels and applications. Daughter boards. Infinity Fabric on a motherboard scale, not a cpu scale.

MSI already built a modular motherboard, don't see why Intel couldn't do similar with a cpu, but I'm betting that's already gone across the minds of some amd folks.

Intel started 5nm research a while ago and have been working on Silicon alternatives for a very long time. Problem is finding one as good as silicon and as cost effective. A lot of the new ideas, like carbon nanotubes, are not as cost effective.

I would think the entire idea behind Forevos is to be able to be more modular. Being able to stitch together any components required for the job is pretty much a modular idea. The problem with taking from the CPU is it increases the overall space required. The reason so much has gone onto the CPU is to shrink overall size. Have you ever looked at the motherboard for an ultrabook? Its about 1/4 of the size of the thing. Then think of cell phones.

The future is bright and I am sure the biog companies have plenty of ideas. We will just have to wait and see which ones are actually viable solutions.
 

Karadjgne

Titan
Ambassador
I don't think so, not really. For years Intel did these little performance bumps, the performance difference between Ivy-Bridge and kabylake is almost negligible. Intel really had no need for a reality check. But everybody uses the competition as a sounding board. Coaches don't just run plays to perfect them, they study the other teams plays, figure out the percentages of use, how to adapt their defense to maximize defensibility and offense to maximize the other guys weaknesses. Can ya blame Intel for doing the same?

Nobody liked WWII, but without it, we'd not have even the simplest things we take for granted, like the Philips head screw.

AMD has finally gotten the party started with the Ryzen cpus, gonna be interesting to see who gets drunk, pukes and passes out first.
 

Karadjgne

Titan
Ambassador
Prolly not as many as you think. A dgpu isn't much more than a physically extended igpu, Intel has been building mobo's for years, has optane and other ram experience, VRM's, so really it's just a cooperative thing between several different departments. And unlike both amd and nvidia, they've got 2 options for competition to compare with.
 
  • Like
Reactions: Isaiah4110

InvalidError

Titan
Moderator
Competition will bring down Nvidia's outrageous prices.
But Intel GPUs might have many bugs at launch as its their first dGPU ever.
AMD's GPU pricing is pretty bad too since they try to closely match Nvidia's $/performance.

This isn't Intel's first attempt at discrete GPUs, that (dis)honor would go to the 740 series from 25 years ago. As Kara wrote though, a dGPU is nothing more than an extended IGP, so nothing fundamentally new there, nearly all of Intel's experience over the past 20+ years of chipset (G-series chipsets prior to IMC) and CPU IGPs (Nehalem to present) is directly transferable. The two biggest difference are the availability of dedicated local VRAM instead of having to share the IMC with the CPU and external devices, and presence of the PCIe bus between CPU and GPU.

Intel's CPUs are used in massive NUMA supercomputers, so I doubt managing multiple memory pools of any sort for a GPU poses a meaningful challenge. Similarly, Intel wrote the PCIe spec which is used extensively in aforementioned massive supercomputers, so that shouldn't be anything resembling an obstacle either.
 

InvalidError

Titan
Moderator
Well, unless Intel made the decision to book TSMC 7nm+ almost a year ago, Intel would be months behind everyone else on getting those 7nm+ wafer starts unless it bought wafer starts from other TSMC clients. Wouldn't make too much sense to waste that much extra money on an experimental product multiple generations behind on building interest and potentially becoming economically viable.
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995
Continuing with RUMORS (notice emphasis)
It seems Intel is going through a big internal restructuring involving several aspects.

1) Intel is planing employee layouts soon. that is actually not a rumor, I've seen it confirmed somewhere

2) Intel is looking to get rid of its memory business. This one I heard in multiple places, and makes sense. Memory is a high volume low margin product. Intel has never been interested in low margin products (it's not the first time intel abandons memory). In addition Intel is capacity constrained, better use that capacity for processors.

3) And ... this is the big one ... Intel is considering to go fabless and sell its fabs to some pure-play foundry.
IF (big if here) this is true, there are only 2 companies with the pockets to do so, Samsung and TSMC. I'll bet on the later.
Some will say this is impossible, but i say this is gonna happen someday, though i don't think it will be soon. After all, over the years, all companies (DEC, HP, AMD, IBM) have been abandoning FABS and gone fabless, Intel is the last one standing.
 
Continuing with RUMORS (notice emphasis)
It seems Intel is going through a big internal restructuring involving several aspects.

1) Intel is planing employee layouts soon. that is actually not a rumor, I've seen it confirmed somewhere

2) Intel is looking to get rid of its memory business. This one I heard in multiple places, and makes sense. Memory is a high volume low margin product. Intel has never been interested in low margin products (it's not the first time intel abandons memory). In addition Intel is capacity constrained, better use that capacity for processors.

3) And ... this is the big one ... Intel is considering to go fabless and sell its fabs to some pure-play foundry.
IF (big if here) this is true, there are only 2 companies with the pockets to do so, Samsung and TSMC. I'll bet on the later.
Some will say this is impossible, but i say this is gonna happen someday, though i don't think it will be soon. After all, over the years, all companies (DEC, HP, AMD, IBM) have been abandoning FABS and gone fabless, Intel is the last one standing.

1. TH got a response but not confirming anything.

2. Maybe but considering they are still pusing Optane I almost doubt it. Its a big advantage over AMD even if its not cheap right now.

3. Highly doubt. Lets use AMD. The biggest difference is AMD did not and still does not have the funds to put into the kind of R&D for process fabrication. Just to make an example, Intel spent $7 billion on upgrading FAB 42 to 7nm EUV. AMDs 2019 revenue was roughly $6 billion. They couldn't afford to build a FAB that could handle a more advanced process.

I could be wrong on all counts. Intel may eventually go FABless but I would assume it to be a point in time where the cost is well outside what they could afford.
 

InvalidError

Titan
Moderator
3) And ... this is the big one ... Intel is considering to go fabless and sell its fabs to some pure-play foundry.
LOL, no. Intel is the biggest semiconductor manufacturer in the world, its fabs aren't going anywhere. The reason why AMD and other smaller players have gone fab-less is because they don't ship enough product to consistently keep entire fabs running at max capacity 24/7. As processes become more expensive, any production stop due to overstock becomes increasingly costly and unacceptable. Intel has the opposite issue of struggling to keep up with on-going demand to the point of having to contract out lower priority devices and having to back-port some devices to older processes where possible, no issues keeping fabs busy 24/7 there.

If Intel finds itself with excess fab capacity, it is more likely to compete with TSMC and Samsung with fabs-for-hire services similar to what it was doing for Altera for the couple of years before it bought Altera out.

I could be wrong on all counts. Intel may eventually go FABless but I would assume it to be a point in time where the cost is well outside what they could afford.
If you can afford to book 100% of someone else's multiple fabs for several years at a time, it is cheaper to build your own since you save the third-party fab's margins and have other efficiency gains from vertical integration. Of course, it also means you directly take the hit when your fabs fall behind targets on process migration.
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995
1. TH got a response but not confirming anything.

2. Maybe but considering they are still pusing Optane I almost doubt it. Its a big advantage over AMD even if its not cheap right now.

3. Highly doubt. Lets use AMD. The biggest difference is AMD did not and still does not have the funds to put into the kind of R&D for process fabrication. Just to make an example, Intel spent $7 billion on upgrading FAB 42 to 7nm EUV. AMDs 2019 revenue was roughly $6 billion. They couldn't afford to build a FAB that could handle a more advanced process.

I could be wrong on all counts. Intel may eventually go FABless but I would assume it to be a point in time where the cost is well outside what they could afford.
1.
https://www.oregonlive.com/silicon-forest/2020/01/intel-plans-layoffs-in-data-center-group.html
  1. Intel Non-Volatile Memory Solutions Group (NSG) keeps losing money. NSG had an operating loss of US$499 million just las quarter. Maybe they just want to get rid off NAND flash, but I bet they will sell the whole group if a buyer is found.
  2. Yeah I don't see it short term, But who knows beyond 7 nm
And of course, 2 & 3 are just rumors.
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995
Intel is the biggest semiconductor manufacturer in the world
In terms of market cap yes (although Samsung was the biggest for like a year or so). But that will not last long:
bb393356-7070-4153-9554-b116ef580307.png

In terms of sales it's second after Samsung
Samsung $75.7 billion
Intel $69.8 billion
TSMC $34.2 billion
However Intel has to design the chip and then FAB it, and finally sell the product. Semiconductor production is just a fraction of its sales. On the other TSMC sales are 100% silicon waffers. The design and sales of the products is done by its customers (AMD, Apple, ...). So in terms of pure silicon wafflers production TSMC is way bigger.

If Intel finds itself with excess fab capacity, it is more likely to compete with TSMC and Samsung with fabs-for-hire services similar to what it was doing for Altera for the couple of years before it bought Altera out.
Intel already tried to enter the foundry busines (search for Intel custom foundry), with little success so far. The 10nm failure had a significant impact. However the strict rules of a vertically integrated company compared with TSMC-samsung-GF played an even bigger role in keeping customers away from Intel.
 
Status
Not open for further replies.