Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [In] Why AMD Failed, Another Ex-Employee Confession
New Posts  All Forums:Forum Nav:

[In] Why AMD Failed, Another Ex-Employee Confession - Page 8

post #71 of 177
Quote:
Originally Posted by DuckieHo;15354517 
Just want to point at that all 250M transistors (not exactly sure where this number comes from) would not have to be placed by hand. The modules themselves are identical. Some components like cache are millions of interations with matured designs.


I agree with the rest though. biggrin.gif

NVIDIA went through this with the GF100 -> GF110 transition. One of the big changes was that they identified which transistors could be swapped for slower/less leaking ones. They ended up placing transistors into a fast/leak, slow/less leaking, and in-between classes to optimize power consumption without a major impact to performance.

Bulldozer is 2 billion transistors. Each core has 213 million transistors (wikipedia). That leaves only 37 million transistors for the non-cache uncore parts of the processor (edit: this assumes that there is no need for tweaking individual modules or cores in any way). If anything, my estimate probably low by 100-300 million transistors.
Edited by hajile - 10/18/11 at 9:19am
post #72 of 177
Quote:
Originally Posted by jlells01;15351881 
rolleyes.gif

http://jonpeddie.com/press-releases/details/graphics-add-in-board-shipments-decline-from-last-quarter/

This Quarter Market Share:

AMD - 41%
Nvidia - 59%

oh boy, how dear u shatter the dreamworlds of amd/ati fanbois tongue.gif
i7 - 550D
(15 items)
 
  
CPUMotherboardGraphicsRAM
i7 2600k ASUS P8P67 PRO EVGA Geforce Titan 4x4GB G.Skillz Ripjaw Z 
Hard DriveOptical DriveCoolingOS
128GB M4 & 1TB F3 ASUS 24x DVD burner Swiftech H220 Windows 8 Pro 
MonitorKeyboardPowerCase
ASUS VG248QE Das Keyboard Silent Corsair AX850 Corsair Obsidian 550D 
MouseMouse PadAudio
Razer Deathadder 2013 Qck+ X-Fi Titanium HD / Creative T40 II / Senn 598 /... 
  hide details  
Reply
i7 - 550D
(15 items)
 
  
CPUMotherboardGraphicsRAM
i7 2600k ASUS P8P67 PRO EVGA Geforce Titan 4x4GB G.Skillz Ripjaw Z 
Hard DriveOptical DriveCoolingOS
128GB M4 & 1TB F3 ASUS 24x DVD burner Swiftech H220 Windows 8 Pro 
MonitorKeyboardPowerCase
ASUS VG248QE Das Keyboard Silent Corsair AX850 Corsair Obsidian 550D 
MouseMouse PadAudio
Razer Deathadder 2013 Qck+ X-Fi Titanium HD / Creative T40 II / Senn 598 /... 
  hide details  
Reply
post #73 of 177
Quote:
Originally Posted by Vagrant Storm;15354597 
Some times I wonder if the reason AMD didn't do so well with this release is because of all these "Former Employees"

Bottom line, it comes down to a single former employee. Dirk Meyer.
post #74 of 177
Quote:
Originally Posted by hajile;15354756 
Bulldozer is 2 billion transistors. Each core has 213 million transistors (wikipedia). That leaves only 37 million transistors for the non-cache uncore parts of the processor (edit: this assumes that there is no need for tweaking individual modules or cores in any way). If anything, my estimate probably low by 100-300 million transistors.

Even within the core, there are duplicated components that are not cache.

I do agree we assume no tweaking of the duplicated components. However, the lowest-hanging-fruit would be tweaking the the "standard" components first. From the sound of it, AMD did not even do this?
AMD_Bulldozer_blockdiagram_2.jpg
Once again...
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 920 [4.28GHz, HT] Asus P6T + Broadcom NetXtreme II VisionTek HD5850 [900/1200] + Galaxy GT240 2x4GB G.Skill Ripjaw X [1632 MHz] 
Hard DriveOSMonitorKeyboard
Intel X25-M 160GB + 3xRAID0 500GB 7200.12 Window 7 Pro 64 Acer H243H + Samsung 226BW XARMOR-U9BL  
PowerCaseMouseMouse Pad
Antec Truepower New 750W Li Lian PC-V2100 [10x120mm fans] Logitech G9 X-Trac Pro 
  hide details  
Reply
Once again...
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 920 [4.28GHz, HT] Asus P6T + Broadcom NetXtreme II VisionTek HD5850 [900/1200] + Galaxy GT240 2x4GB G.Skill Ripjaw X [1632 MHz] 
Hard DriveOSMonitorKeyboard
Intel X25-M 160GB + 3xRAID0 500GB 7200.12 Window 7 Pro 64 Acer H243H + Samsung 226BW XARMOR-U9BL  
PowerCaseMouseMouse Pad
Antec Truepower New 750W Li Lian PC-V2100 [10x120mm fans] Logitech G9 X-Trac Pro 
  hide details  
Reply
post #75 of 177
Quote:
Originally Posted by Vagrant Storm;15354597 
Some times I wonder if the reason AMD didn't do so well with this release is because of all these "Former Employees"
You think former employees my have sabotage bulldozer because they didn't like Dirk Meyer and the way the company was going?
Predator
(15 items)
 
  
CPUMotherboardGraphicsRAM
Intel 5820k MSI X99  Gigabyte 1080 Ti Turbo HyperX Fury 24GB  
Hard DriveOptical DriveCoolingOS
OCZ Vector 150 Series 240GB Pioneer Blu-ray Burner Noctua NH-D14 Windows 10 x64 
MonitorKeyboardPowerCase
Samsung KU7000 GAMDIAS Ares Gaming Keyboard In Win Commander 1200W NZXT: White Phantom Case 
MouseMouse PadAudio
Ourea Gaming Optical Mouse Ajazz First Blood Professional Gaming Mouse Pad Hyper CloudX 
  hide details  
Reply
Predator
(15 items)
 
  
CPUMotherboardGraphicsRAM
Intel 5820k MSI X99  Gigabyte 1080 Ti Turbo HyperX Fury 24GB  
Hard DriveOptical DriveCoolingOS
OCZ Vector 150 Series 240GB Pioneer Blu-ray Burner Noctua NH-D14 Windows 10 x64 
MonitorKeyboardPowerCase
Samsung KU7000 GAMDIAS Ares Gaming Keyboard In Win Commander 1200W NZXT: White Phantom Case 
MouseMouse PadAudio
Ourea Gaming Optical Mouse Ajazz First Blood Professional Gaming Mouse Pad Hyper CloudX 
  hide details  
Reply
post #76 of 177
This is, at best, a rumor. None of it can be confirmed since the ex-employee remains anonymous, and I guaranteed is bound by multiple NDA's/confidentiality agreements. No one can say any of this is untrue, yet that doesn't prove it is true.

Besides, their new source links to a forum rumors thread regarding Apple and whether they wanted to use AMD processors--which is 29 pages long, and I didn't bother to read through, but regardless, I'm not sure what is and is not accurate.

That being said, companies change--not always for the best, and it's a sad truth that there's always the interest in driving down costs. AMD is no different. And to be quite honest, there may be genuine legit reasons AMD chose the direction that they did--engineers are not high-level managers with a business sense for an entire company, and are not privy to the many goings-on in an entire company. I'm sure that this ex-engineer--if the story is true--had no idea of the business decisions management made, and why they made them. Sometimes we hear of a decision by management that is trickled down and we go, "why in the world would they do that? It makes no sense" When we also don't have all of the info as to why, only the new decision, and perhaps after knowing all of the info, it may make complete sense.

Anyway, I would treat all of this with skepticism. I'm not saying that any of it is untrue, but it's the rare case that an ex-employee speaks out in favor of a company. Their little blurb about how their time at AMD helped pay for a house and porsche is a moot point, and has nothing to do with the business practices of the company and personal feelings about it. I've worked for places that helped me achieve financial gain, but I also absolutely hated working there--financial benefit of working someplace doesn't equal a non-disgruntled attitude toward it.
    
CPUMotherboardGraphicsRAM
i5-2500K Biostar TP67B+ XFX HD5750 1GB 2x4GB DDR3 Corsair 1600 
Hard DriveOSMonitorPower
60GB OCZ SSD, 2x160GB HDD RAID0, 500GB+500GB+1.5TB Windows 7 Ultimate 64-bit Samsung SyncMaster 930B Antec SmartPower 450w 
Case
Antec 900 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
i5-2500K Biostar TP67B+ XFX HD5750 1GB 2x4GB DDR3 Corsair 1600 
Hard DriveOSMonitorPower
60GB OCZ SSD, 2x160GB HDD RAID0, 500GB+500GB+1.5TB Windows 7 Ultimate 64-bit Samsung SyncMaster 930B Antec SmartPower 450w 
Case
Antec 900 
  hide details  
Reply
post #77 of 177
And we take the word of an ex-girlfriend lulz
The Little Guy
(20 items)
 
I come in peace
(14 items)
 
 
CPUMotherboardGraphicsRAM
i7 5820k Asus X99-A MSI R9-290X 2 x Corsair Vengeance LPX 16GB (4 x 4GB) DDR4 2... 
Hard DriveCoolingCoolingCooling
2 x SAMSUNG 840 Pro Series MZ-7PD512BW 2.5" 512... Black Ice GT Stealth 360 Radiator - Blue EK Supremacy CPU Liquid Cooling Block - Nickel ... EK Radeon R9-290X VGA Liquid Cooling Block - Ac... 
CoolingCoolingCoolingOS
Monsoon Series Two Premium D5 / MCP655 Dual Bay... Swiftech MCP655-PWM-DRIVE 12v Water Pump Module... 3 x Noctual NF-F12 120 x 25mm PWM fan Windows 10 Professional 
MonitorMonitorKeyboardPower
2 x Dell UltraSharp U2312HM 23" IPS Asus MG279Q 27" 144 Hz IPS Freesync 35-90Hz Range Razer Black Widow Ultimate Corsair AX-860i 
CaseMouseMouse PadAudio
Cooler Master Cosmos II Razer Naga 2014 Left-Hand edition Razer Vespula Creative Labs Titanium HD 
CPUMotherboardGraphicsRAM
AMD FX-8120 ASUS Crosshair V XFX R7-260X-CNF4 Core Edition Radeon R7 260X 2G... Crucial Ballistix Sport 16GB (2 x 8GB) 240-Pin ... 
Hard DriveHard DriveCoolingOS
4 x Seagate NAS HDD ST2000VN000 2TB 64MB Cache ... 2 x OCZ Vertex 3 240 GB Corsair H-80 Windows Server 2012 Essentials 
PowerCaseOtherOther
Corsair HX-850 Lian Li PC-X500B LSI 9300 MegaRAID SAS 9361-8i (LSI00417) PCI-Ex... LSI LSICVM02 (LSI00418) CacheVault Accessory Ki... 
Other
Intel Ethernet Server Adapter i210 
Other
Surface Book 
  hide details  
Reply
The Little Guy
(20 items)
 
I come in peace
(14 items)
 
 
CPUMotherboardGraphicsRAM
i7 5820k Asus X99-A MSI R9-290X 2 x Corsair Vengeance LPX 16GB (4 x 4GB) DDR4 2... 
Hard DriveCoolingCoolingCooling
2 x SAMSUNG 840 Pro Series MZ-7PD512BW 2.5" 512... Black Ice GT Stealth 360 Radiator - Blue EK Supremacy CPU Liquid Cooling Block - Nickel ... EK Radeon R9-290X VGA Liquid Cooling Block - Ac... 
CoolingCoolingCoolingOS
Monsoon Series Two Premium D5 / MCP655 Dual Bay... Swiftech MCP655-PWM-DRIVE 12v Water Pump Module... 3 x Noctual NF-F12 120 x 25mm PWM fan Windows 10 Professional 
MonitorMonitorKeyboardPower
2 x Dell UltraSharp U2312HM 23" IPS Asus MG279Q 27" 144 Hz IPS Freesync 35-90Hz Range Razer Black Widow Ultimate Corsair AX-860i 
CaseMouseMouse PadAudio
Cooler Master Cosmos II Razer Naga 2014 Left-Hand edition Razer Vespula Creative Labs Titanium HD 
CPUMotherboardGraphicsRAM
AMD FX-8120 ASUS Crosshair V XFX R7-260X-CNF4 Core Edition Radeon R7 260X 2G... Crucial Ballistix Sport 16GB (2 x 8GB) 240-Pin ... 
Hard DriveHard DriveCoolingOS
4 x Seagate NAS HDD ST2000VN000 2TB 64MB Cache ... 2 x OCZ Vertex 3 240 GB Corsair H-80 Windows Server 2012 Essentials 
PowerCaseOtherOther
Corsair HX-850 Lian Li PC-X500B LSI 9300 MegaRAID SAS 9361-8i (LSI00417) PCI-Ex... LSI LSICVM02 (LSI00418) CacheVault Accessory Ki... 
Other
Intel Ethernet Server Adapter i210 
Other
Surface Book 
  hide details  
Reply
post #78 of 177
Quote:
Originally Posted by DuckieHo;15354952 
Even within the core, there are duplicated components that are not cache.

I do agree we assume no tweaking of the duplicated components. However, the lowest-hanging-fruit would be tweaking the the "standard" components first. From the sound of it, AMD did not even do this?
AMD_Bulldozer_blockdiagram_2.jpg

die.jpg
FXCPU_Die.jpg

Even if we assume that the integer cores are duplicated, AMD stated that the second integer core only adds 10% to the die size. To push the argument further (note: I'm not being too serious here, just having a little fun with a (hopefully) light-hearted debate) let's assume that both sides of the FPU are identical, assume that both integer modules are identical and that within an integer unit the ALU's are identical, assume that memory controllers are identical, HT's are identical, etc

Let's assume that when this is all finished, we have just 50 million transistors to lay out by hand. With a 50 man team (49 as he claims), each person would have to lay out 1,000,000 transistors by hand. Let's further assume that a new design takes five years and that the engineers work 4000 hours per year (that's 16hr/workday) and 20,000 work hours total. Assuming that anyone working that many hours would actually be able to work the entire time and that the work would be of the best quality possible, that still leaves 50 transistors laid out per hour or slightly less than one per minute.

All of this still assumes that the engineer is capable of laying out each circuit in the optimal manner on the first try and that the engineer doesn't have any time to have another engineer glance at the circuits to double check nor does the engineer have the time to error test.

If the numbers become more real (fewer work hours, more circuits, need of more time to design and error-test circuits, etc) the necessary amount of people quickly grows to thousands.

In the famous words of Billy Mays, "But wait, there's more." As the design team grows, efficiency drops and strange errors begin to appear as the different parts of the design come together leading to the need for even more designers and testers and time.
post #79 of 177
Any huge endeavor like this has things that go wrong. Even intel has things that went wrong in all cpus up to this point. Nobody is perfect.

The problem is that when things fail (like bulldozer and the Boston Red Sox), all the bad things surface.
Silent But Deadly
(15 items)
 
  
CPUMotherboardGraphicsRAM
3570k @4.4 Gigabyte Z77X-UD3H Powercolor AMD 470 8GB Samsung 
Hard DriveCoolingCoolingOS
Cricual M4 64 / WD6400AAKS Silver Arrow SB-E Arctic Cooling 7970 7 64 Ultimate 
MonitorPowerCaseMouse
BenQ XL2410T Corsair AX650 Fractal Define R4 Logitech G5 
  hide details  
Reply
Silent But Deadly
(15 items)
 
  
CPUMotherboardGraphicsRAM
3570k @4.4 Gigabyte Z77X-UD3H Powercolor AMD 470 8GB Samsung 
Hard DriveCoolingCoolingOS
Cricual M4 64 / WD6400AAKS Silver Arrow SB-E Arctic Cooling 7970 7 64 Ultimate 
MonitorPowerCaseMouse
BenQ XL2410T Corsair AX650 Fractal Define R4 Logitech G5 
  hide details  
Reply
post #80 of 177
Quote:
Originally Posted by hajile;15355255 
Even if we assume that the integer cores are duplicated, AMD stated that the second integer core only adds 10% to the die size. To push the argument further (note: I'm not being too serious here, just having a little fun with a (hopefully) light-hearted debate) let's assume that both sides of the FPU are identical, assume that both integer modules are identical and that within an integer unit the ALU's are identical, assume that memory controllers are identical, HT's are identical, etc

Let's assume that when this is all finished, we have just 50 million transistors to lay out by hand. With a 50 man team (49 as he claims), each person would have to lay out 1,000,000 transistors by hand. Let's further assume that a new design takes five years and that the engineers work 4000 hours per year (that's 16hr/workday) and 20,000 work hours total. Assuming that anyone working that many hours would actually be able to work the entire time and that the work would be of the best quality possible, that still leaves 50 transistors laid out per hour or slightly less than one per minute.

All of this still assumes that the engineer is capable of laying out each circuit in the optimal manner on the first try and that the engineer doesn't have any time to have another engineer glance at the circuits to double check nor does the engineer have the time to error test.

If the numbers become more real (fewer work hours, more circuits, need of more time to design and error-test circuits, etc) the necessary amount of people quickly grows to thousands.

In the famous words of Billy Mays, "But wait, there's more." As the design team grows, efficiency drops and strange errors begin to appear as the different parts of the design come together leading to the need for even more designers and testers and time.
However, engineers aren't working at a transistor at a time. They are looking a sub circuits comprised of a group of transistors. In addition, not every single transistor requires to be laid out by hand. There should be enough knowledge transfer and experience from previous generations where some circuits are already well-developed and optimized. This knowledge would also allow engineers to target circuits that would provide the most benefit as well. I assume that they are not starting from scratch.


It absolutely is a lot more work.... and it looks like AMD did not have the time, money, skills, and/or will to do it. Not everything has to be optimized... just the things that suck the worst.
Quote:
Originally Posted by homestyle;15355264 
Any huge endeavor like this has things that go wrong. Even intel has things that went wrong in all cpus up to this point. Nobody is perfect.

The problem is that when things fail (like bulldozer and the Boston Red Sox), all the bad things surface.
However, this is not just a little failure.... this is the failure of their flagship.

Intel really dropped the ball on the multi-billion Larrabee project. However, that was a side-project that did not impact their core x86 business in the near-term and Intel has the budget to absorb that failure.
Once again...
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 920 [4.28GHz, HT] Asus P6T + Broadcom NetXtreme II VisionTek HD5850 [900/1200] + Galaxy GT240 2x4GB G.Skill Ripjaw X [1632 MHz] 
Hard DriveOSMonitorKeyboard
Intel X25-M 160GB + 3xRAID0 500GB 7200.12 Window 7 Pro 64 Acer H243H + Samsung 226BW XARMOR-U9BL  
PowerCaseMouseMouse Pad
Antec Truepower New 750W Li Lian PC-V2100 [10x120mm fans] Logitech G9 X-Trac Pro 
  hide details  
Reply
Once again...
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 920 [4.28GHz, HT] Asus P6T + Broadcom NetXtreme II VisionTek HD5850 [900/1200] + Galaxy GT240 2x4GB G.Skill Ripjaw X [1632 MHz] 
Hard DriveOSMonitorKeyboard
Intel X25-M 160GB + 3xRAID0 500GB 7200.12 Window 7 Pro 64 Acer H243H + Samsung 226BW XARMOR-U9BL  
PowerCaseMouseMouse Pad
Antec Truepower New 750W Li Lian PC-V2100 [10x120mm fans] Logitech G9 X-Trac Pro 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Rumors and Unconfirmed Articles
Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [In] Why AMD Failed, Another Ex-Employee Confession