• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

What is really needed to fix Intel

Joined
May 25, 2022
Messages
116 (0.13/day)
It took Intel until 2010 to get an integrated memory controller, and on the client parts, it didn't get it fully until Sandy Bridge in 2011. 7 years after the dinky competitor AMD. Do you know why?

Intel is a finance company with engineers on the side #1

Because they used the latest process for CPUs, but N-1 for the MCH, and N-2 for the IO Hub. By moving to an IMC, it would have meant they couldn't utilize older fabs. So since the industry was nowhere near 2.5D interconnects, Intel should have opened up their fabs way back then. Do you understand this? Without this line of thinking they could have had an integrated memory controller maybe 2001, maybe 1999!

It took Intel many years before they took the Atom seriously. Do you know why?

Intel is a finance company with engineers on the side #2

Because they were scared of losing margins and wanted an artificial gap of 10x CPU performance between Atom and Core. They said it publicly. Then Apple in 2013 beat Silvermont to a pulp, and the rest is history. Silvermont had an anemic 4EU HD Graphics controller. Originally they wanted a 2EU version.

Intel had a chance of partaking in Apple's hardware ecosystem. Even their haphazard contra-revenue Atom approach would have worked better, since getting early in would have meant the ecosystem would be x86(thus Intel's favor). But they lost it. Do you know why?

Intel is a finance company with engineers on the side #3

Steve Jobs wanted to include Intel in the iPhone/iPad hardware. Paul Otellini in all his finance wisdom, said "No, I don't believe in your vision and I think your volumes are too low".

Some say it wasn't about using CPUs, but their fabs. So what? That would have been a boon too. It surely would have helped Gelsinger's rush-to-18A strategy too.

Why did Intel rush into increasing vector sizes with AVX, AVX2, and AVX512, leading to decreased clocks and thermal issues and software fragmentation? Every two years they doubled it. Do you know why?

Intel is a finance company with engineers on the side #4

Because at that time their main threat was Nvidia, or they thought. It would come to be eventually, but their line of reasoning was that boosting CPU general purpose FP performance would discourage transition to Nvidia. It's not entirely a stupid idea, until it is. They should have stayed at AVX2, and AVX512 should have been AVX3 - AVX512's instructions without the 512-bit width.

-Why would a company at the forefront of Moore's Law not bring a Celeron until they were forced to? Lower cost is natural.
-Why would a multi-national leading chip company not integrate many things as possible and wait for so long? Again it's natural.
-Why would a chip company with massive hand at every part of a PC ecosystem not look at every single ways to have a laptop with great, great battery life? Smaller, lower power, faster, that is at the HEART of Moore's Law.
 
Joined
Feb 10, 2023
Messages
266 (0.42/day)
Location
Lake Superior
What is really needed to fix Intel?
A working, leading-edge process that can attract foundry customers which in turn creates enough revenue to invest in their fabs.
I don't think 18A is that.
 
Joined
Jan 14, 2019
Messages
12,159 (5.74/day)
Location
Midlands, UK
System Name Nebulon B
Processor AMD Ryzen 7 7800X3D
Motherboard MSi PRO B650M-A WiFi
Cooling be quiet! Dark Rock 4
Memory 2x 24 GB Corsair Vengeance DDR5-4800
Video Card(s) AMD Radeon RX 6750 XT 12 GB
Storage 2 TB Corsair MP600 GS, 2 TB Corsair MP600 R2
Display(s) Dell S3422DWG, 7" Waveshare touchscreen
Case Kolink Citadel Mesh black
Audio Device(s) Logitech Z333 2.1 speakers, AKG Y50 headphones
Power Supply Seasonic Prime GX-750
Mouse Logitech MX Master 2S
Keyboard Logitech G413 SE
Software Windows 10 Pro
I admit, I haven't read your full post, but judging by the length of it, it seems that you have a very elaborate opinion on the matter with not much room for further discussion.
 
Joined
Mar 18, 2008
Messages
5,440 (0.90/day)
Location
Australia
System Name Night Rider | Mini LAN PC | Workhorse
Processor AMD R7 5800X3D | Ryzen 1600X | i7 970
Motherboard MSi AM4 Pro Carbon | GA- | Gigabyte EX58-UD5
Cooling Noctua U9S Twin Fan| Stock Cooler, Copper Core)| Big shairkan B
Memory 2x8GB DDR4 G.Skill Ripjaws 3600MHz| 2x8GB Corsair 3000 | 6x2GB DDR3 1300 Corsair
Video Card(s) MSI AMD 6750XT | 6500XT | MSI RX 580 8GB
Storage 1TB WD Black NVME / 250GB SSD /2TB WD Black | 500GB SSD WD, 2x1TB, 1x750 | WD 500 SSD/Seagate 320
Display(s) LG 27" 1440P| Samsung 20" S20C300L/DELL 15" | 22" DELL/19"DELL
Case LIAN LI PC-18 | Mini ATX Case (custom) | Atrix C4 9001
Audio Device(s) Onboard | Onbaord | Onboard
Power Supply Silverstone 850 | Silverstone Mini 450W | Corsair CX-750
Mouse Coolermaster Pro | Rapoo V900 | Gigabyte 6850X
Keyboard MAX Keyboard Nighthawk X8 | Creative Fatal1ty eluminx | Some POS Logitech
Software Windows 10 Pro 64 | Windows 10 Pro 64 | Windows 7 Pro 64/Windows 10 Home

Ruru

S.T.A.R.S.
Joined
Dec 16, 2012
Messages
12,547 (2.89/day)
Location
Jyväskylä, Finland
System Name 4K-gaming
Processor AMD Ryzen 7 5800X @ PBO +200 -20CO
Motherboard Asus ROG Crosshair VII Hero
Cooling Arctic Freezer 50, EKWB Vector TUF
Memory 32GB Kingston HyperX Fury DDR4-3200
Video Card(s) Asus GeForce RTX 3080 TUF OC 10GB
Storage A pack of SSDs totaling 3.2TB + 3TB HDDs
Display(s) 27" 4K120 IPS + 32" 4K60 IPS + 24" 1080p60
Case Corsair 4000D Airflow White
Audio Device(s) Asus TUF H3 Wireless / Corsair HS35
Power Supply EVGA Supernova G2 750W
Mouse Logitech MX518 + Asus TUF P1 mousepad
Keyboard Roccat Vulcan 121 AIMO
VR HMD Oculus Rift CV1
Software Windows 11 Pro
Benchmark Scores It runs Crysis
Actually the first gen Core i7s on LGA1366 had IMC in 2008. 2009 with the consumer LGA1156.
 

qxp

New Member
Joined
Oct 27, 2024
Messages
2 (0.50/day)
Here are few more thoughts:
  • Historically Intel started as a cutting-edge fab company. The CPU arose as a side project to find something for the fab to do. So there is a lot of attachment to having not just your own fab, but a cutting edge fab.
  • Intel has a history for making weird processor designs - the early one was i860, which most notable use was as a graphics card for NeXT systems. The next famous flop was Pentium IV
  • From the point of view of computer architecture NVidia GPUs are horrible - they incorporate every known mistake and then some. Recent GPUs have improved somewhat, but it should have been easy for Intel to beat NVidia back 10 years ago.
  • Intel did have a fantastic competitor - a Xeon Phi card which could provide around 1 TFlop of compute compared to NVidia 2 TFlop at the time. This should have been the killer because Xeon Phi was just a massively multicore Pentium with AVX512 compute unit and much easier to program than NVidia GPU. This would have resulted in faster and better software, if only Intel released at prices competitive with consumer GPU cards and with a video port.
  • This, of course, did not happen for the obvious reason - the card would have competed with Intel server chips that earned a lot of profit. At the time a lot of server farms would host websites and having a 100+ cheap Pentium server would allow to sell 100+ virtual instances and cut into Intel revenue. So instead they played stupid business tricks of making their own product uncompetitive lest it destroys existing pricing structure.
  • The main problem that Intel is facing now is the collapse in demand for Intel servers, which brought a lot of revenue. In order to fix that they need to release a competitive product, but that needs a lot of money they are short on.
  • So Intel was done in by their desire to protect existing business of selling server chips at large margins.
  • And now that NVidia value is through the roof and *their* chips are in demand they are doing the same stupid tricks. Did you wonder why all the new GPUs have these monstrous heatsinks and small amounts of RAM ? It is so they don't physically fit into regular servers and they are hard to use for AI compared with "professional" chips like H100 which have less compute than 4090. I've heard of a company that have been asked whether they could make a small run of NVidia GPUs with normal blower-style heatsinks for a cluster and they said yes, but then NVidia threatened them and they pulled out.
 
Joined
Aug 19, 2024
Messages
183 (2.47/day)
Location
Texas, USA
System Name Obliterator
Processor Ryzen 7 7700x PBO
Motherboard ASRock x670e Steel Legend
Cooling Noctua NH-D15 G2 LBC
Memory G.skill Trident Z5 Neo 6000@CL30
Video Card(s) ASRock rx7900 GRE Steel Legend
Storage 2 x 2TB Samsung 990 pro nmve ssd 2 X 4TB Samsung 870 evo sata ssd 1 X 18TB WD Gold sata hdd
Display(s) LG 27GN750-B
Case Fractal Torrent
Audio Device(s) Klipsch promedia heritage 2.1
Power Supply FSP Hydro TI 1000w
Mouse SteelSeries Prime+
Keyboard Lenovo SK-8825 (L)
Software Windows 10 Enterprise LTSC 21H2 / Windows 11 Enterprise LTSC 24H2 with multiple flavors of VM
Just my two-cents...

For as much as the upper executives may go on at the mouth, Intel has been surprisingly tight-lipped on the R&D front (for a while, to my reckoning). Others are really hyping their next thing, but not Intel. I wonder if they are working on something big. Something that is taking up large shares of their attention that would normally focus on other areas of their business.

I could be wrong. Maybe they got nothin, and they really are in trouble. They are still a very large company that is a force in the computer world. I'm not counting them out.

My guess is memory/storage of some type. The successor to 3-d xpoint perhaps.

Either way, strange times in the computer biz.
 
Joined
May 25, 2022
Messages
116 (0.13/day)
Here are few more thoughts:
  • From the point of view of computer architecture NVidia GPUs are horrible - they incorporate every known mistake and then some. Recent GPUs have improved somewhat, but it should have been easy for Intel to beat NVidia back 10 years ago.
This part makes a lot of sense. Until Voodoo, graphics were only FP64, so it was expensive to make and used in 3D workstations such as SGI. Then founders of Voodoo realized lowering precision, and estimating and approximating was fine for client graphics. It doesn't need full accuracy. It's doing only what is needed, and nothing more.

(On a little side note, I suspect using even less precision such as FP16 and Int8 is what causes hallucination for AI models - nothing is free)
- So Intel was done in by their desire to protect existing business of selling server chips at large margins.
  • And now that NVidia value is through the roof and *their* chips are in demand they are doing the same stupid tricks. Did you wonder why all the new GPUs have these monstrous heatsinks and small amounts of RAM ? It is so they don't physically fit into regular servers and they are hard to use for AI compared with "professional" chips like H100 which have less compute than 4090. I've heard of a company that have been asked whether they could make a small run of NVidia GPUs with normal blower-style heatsinks for a cluster and they said yes, but then NVidia threatened them and they pulled out.
Yup. Did that with Xeons during Itanium days. Did that again with Atom during their peak times(Sandy Bridge to Skylake). A finance company that hires engineers.
What is really needed to fix Intel?
A working, leading-edge process that can attract foundry customers which in turn creates enough revenue to invest in their fabs.
I don't think 18A is that.
You are focusing on the tech side.* 18A is defeatured, and is essentially a rebranded 20A. It was 20A with 15% perf over Intel 3 and then 18A got 10% on top of that. 18A on recent slide said 15% over Intel 3. The density is only 30% over Intel 3. It takes until 18A-P a year later to get the 10% originally promised.

Mind you Intel 3 promised improved density for HP libraries, and it ended up being a mere 10% density, if you used the HD library. At least it kept the performance.

*The key is loss of personnel and brain drain is what's causing all the problems. It's culture, it's tradition, all rearing it's ugly head. They had this problem for decades, now that they lost the only advantage they really had(process), everything is surfacing.
 
Joined
Jan 29, 2023
Messages
1,521 (2.37/day)
Location
France
System Name KLM
Processor 7800X3D
Motherboard B-650E-E Strix
Cooling Arctic Cooling III 280
Memory 16x2 Fury Renegade 6000-32
Video Card(s) 4070-ti PNY
Storage 500+512+8+8+2+1+1+2+256+8+512+2
Display(s) VA 32" 4K@60 - OLED 27" 2K@240
Case 4000D Airflow
Audio Device(s) Edifier 1280Ts
Power Supply Shift 1000
Mouse 502 Hero
Keyboard K68
Software EMDB
Benchmark Scores 0>1000
Intel fix would be to be merged with AMD & NVidia, no more competion, all brains working/going in the same way.

Fixed.
 
Joined
Jan 11, 2022
Messages
823 (0.80/day)
What intel needs is what the world needs, an economic revival.
so all that ai bubble and risk averse tech stock money can flow in to real projects
which in turn make people and businesses buy new computers
 
Joined
Jan 29, 2023
Messages
1,521 (2.37/day)
Location
France
System Name KLM
Processor 7800X3D
Motherboard B-650E-E Strix
Cooling Arctic Cooling III 280
Memory 16x2 Fury Renegade 6000-32
Video Card(s) 4070-ti PNY
Storage 500+512+8+8+2+1+1+2+256+8+512+2
Display(s) VA 32" 4K@60 - OLED 27" 2K@240
Case 4000D Airflow
Audio Device(s) Edifier 1280Ts
Power Supply Shift 1000
Mouse 502 Hero
Keyboard K68
Software EMDB
Benchmark Scores 0>1000
Yes, brains are corruptible too, if not all one is enough to be too risky to take the risk...

... but imagine the final result of such an enterprise :p

No more bugs, no more traversing stutter, no more BSOD (or GSOD for Insiders).....................................
 
Top