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 18

post #171 of 177
Quote:
Originally Posted by Ceadderman;15365573 
Especially from a guy with a whopping conversational total of 181 posts. rolleyessmileyanim.gif

Post count means crap all, I have more posts than you, does that mean I know more than you?

Duckie has the most posts out of anyone in this forum that I know of, does that mean he knows more than all of us combined? (He probably does, but that's beside the point.)
Quote:
Originally Posted by AtomicFrost;15366001 
They claim that Fusion is the future, but the number of people actually buying an AMD Fusion platform are quite low. AMD just doesn't have the brand recognition to pull it off something like this in the retail market.

I've actually seen quite a few Bobcat netbooks around where I live.
Quote:
Originally Posted by hajile;15366672 
The Athlon64 (K8) core is slower per clock than Bulldozer (in IPC, K10.5 > Bulldozer > K8). In addition, I don't believe (I have no definite proof) that the K10.5 core could be used. It has more ALU's and would take up much more die area. Further, the largest difference is not the integer cores, but is instead the earlier stages of the pipeline.

As to P4 (first shipped November 20, 2000) being good until A64 (original Athlon shipped June 23, 1999). This is not quite accurate. (note at this point: The design team lead on the A7 AND A8 was Dirk Meyer, the same design team lead on Bulldozer.) The performance of the original Athlons was better than P4 and they launched earlier. Later iterations (technically A8) such as Sledgehammer added 64-bit support later in 2003 (and 15% IPC iirc), but Athlon beat the P4 even before Sledgehammer (beat it in price, performance, and power consumption).

The problem was that P4 was a speed demon without good IPC (20-25% slower IPC than Athlon) and the side affects (non-linear scaling, heat, power consumption, branch complexity, etc) of high-speed chips was either not completely understood or unknown at the time. The architecture had flaws from its first inception (even if these flaws weren't understood). Here's an article from 2000 done by Anandtech describing the netburst architecture. He was fairly excited, but even then, the clockspeed problem was apparent (the problem became even more apparent when the first chips launched (another anand article) at 1.5 Ghz.

The takeaway, the initial launch of the P4 was terrible and there was no reason to buy because it was expensive and offered no advantages.

Just like the P4 was the future that wasn't (some of the tech used, but the project was scrapped), I believe that there's a huge possibility that CMT doesn't work as well in reality as it does on paper. Once again, only time will tell (unless you have means to design and test CMT processors in your spare time).

You're forgetting something though, the original Willamette P4s (1.4 and 1.5Ghz on Socket 428) were indeed slower than the Athlon XPs and Pentium IIIs, but the Northwoods were the fastest things you could buy before the Athlon64 and weren't that bad...The really bad P4s were the Prescotts, a massive rework of the core but put out too much heat, most of the P4s heat problem came from Prescott and the 90nm process being leaky.

You repeated pretty much what I'm saying, the original P4 launched at far too low of a clock speed to really be competitive, the Northwood P4s were at the right clock speeds to generally beat the AthlonXPs, but then the Athlon64 came out and Prescott turned out to be a dud.

And the K10.5 core (Tweaked K8) is generally faster than BD clock for clock because of the cache and front-end issues in BD, the L1 cache is too small (Especially since one part of it is shared) and the L2 and L3 caches are too slow (Nearly double the latency of a K10.5 cache) with the front-end decoder being too slow for two threads at once (It ends up doing one less op a clock than K10.5 when you have both pipelines working)

And yes, it's very true CMT may not work out as well as it was theoretically meant to, but at least unlike the P4 it wasn't designed for marketing from the start.
    
CPUMotherboardGraphicsRAM
Intel Core i5 3570k @ 4.5Ghz ASRock Z77 Pro3 Powercolor Radeon HD7950 3GB @ 1150/1350 4x4GB G.Skill Ares 2000Mhz CL9 
Hard DriveHard DriveHard DriveHard Drive
Samsung 840 250GB Western Digital Black 1TB WD1002FAEX Seagate Barracuda 3TB ST3000DM001 Samsung Spinpoint EcoGreen 2TB 
Optical DriveCoolingCoolingCooling
Pioneer DVR-220LBKS Noctua NH-D14 Scythe Gentle Typhoon 1850rpm Corsair AF140 Quiet Edition 
CoolingOSMonitorMonitor
Arcitc Cooling Acclero Twin Turbo II Arch Linux x86-64, amdgpu BenQ G2220HD BenQ G2020HD 
KeyboardPowerCaseMouse
Ducky Shine III Year of the Snake, Cherry Blue Silverstone Strider Plus 600w CoolerMaster CM690 II Black and White SteelSeries Sensei Professional 
Mouse PadAudioOther
Artisan Hien Mid Japan Black Large ASUS Xonar DX NZXT Sentry Mesh 30w Fan Controller 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Intel Core i5 3570k @ 4.5Ghz ASRock Z77 Pro3 Powercolor Radeon HD7950 3GB @ 1150/1350 4x4GB G.Skill Ares 2000Mhz CL9 
Hard DriveHard DriveHard DriveHard Drive
Samsung 840 250GB Western Digital Black 1TB WD1002FAEX Seagate Barracuda 3TB ST3000DM001 Samsung Spinpoint EcoGreen 2TB 
Optical DriveCoolingCoolingCooling
Pioneer DVR-220LBKS Noctua NH-D14 Scythe Gentle Typhoon 1850rpm Corsair AF140 Quiet Edition 
CoolingOSMonitorMonitor
Arcitc Cooling Acclero Twin Turbo II Arch Linux x86-64, amdgpu BenQ G2220HD BenQ G2020HD 
KeyboardPowerCaseMouse
Ducky Shine III Year of the Snake, Cherry Blue Silverstone Strider Plus 600w CoolerMaster CM690 II Black and White SteelSeries Sensei Professional 
Mouse PadAudioOther
Artisan Hien Mid Japan Black Large ASUS Xonar DX NZXT Sentry Mesh 30w Fan Controller 
  hide details  
Reply
post #172 of 177
Quote:
Originally Posted by hajile;15366672 
The quotes in the article are taken from a macrumors thread posted in April 2010 from a guy claiming to be a former AMD engineer turned IP lawyer.

Here's a link to the original thread (29 pages)

From the posters profile on macrumors.

.

I was hoping for a linkedin profile or a company profile page on a website. This so called ex-employee doesn't seem to be credible. Even some of the things he said, such as saying hand placing cells is more efficient than machines seems a bit weird. The rest of what he said can be applied to every corporation - out with the old highly paid workers and in with the new unskilled low paid employees.
Edited by PoopaScoopa - 10/19/11 at 4:07pm
post #173 of 177
I just hope AMD get their act together and deliver something better next time around. I've been a fan since Socket A, and they were still the obvious choice for my budget when I got this rig (the alternatives were LGA775 or LGA1366, ie EOL or too expensive), but there's just no way I'd buy something like BD.
Edited by HothBase - 10/19/11 at 4:09pm
 
Portable Desktop
(12 items)
 
Console Cave
(11 items)
 
CPUMotherboardGraphicsRAM
Intel Core i5 2500K ASRock P67 Extreme4 ASUS GTX 670 DCII Kingston HyperX Fury 
Hard DriveHard DriveHard DriveHard Drive
Seagate 600 Samsung Spinpoint F4 Seagate Barracuda Green Seagate Desktop 
Hard DriveHard DriveHard DriveHard Drive
Seagate FreeAgent GoFlex Seagate Expansion Desktop Maxtor Basics Seagate Backup Plus Portable 
Optical DriveCoolingCoolingCooling
Samsung SE-208AB Thermalright Archon (Rev. A) 2x Phanteks PH-F140HP_BL (CPU) 2x Noctua NF-A14 PWM (Intake) 
CoolingCoolingOSMonitor
Thermalright TY-150 (Exhaust) Noctua NF-S12A PWM (Exhaust) Windows 10 Pro DELL UltraSharp U2711 
KeyboardPowerCaseMouse
QPAD MK-80 Pro SeaSonic X-560 Lian Li PC-P50WB Func MS-3 R2 
Mouse PadAudioAudioAudio
SteelSeries QcK Creative Sound Blaster Z Swans M10 Creative Aurvana Live! 
Other
Lian Li LED50 
CPUMotherboardGraphicsGraphics
Intel Pentium G3258 ASUS H81I-PLUS Intel HD Graphics ASUS GTX 670 DirectCU II 
RAMHard DriveCoolingOS
Corsair Vengeance LP OCZ Vector Scythe Big Shuriken 2 rev. B Windows 10 Pro N 
MonitorMonitorPowerCase
LG PA70G BenQ G2420HDBL Cooler Master B500 ver.2 Cooler Master Elite 130 
MonitorAudioAudioOther
LG PA70G Yamaha HTR-3065 Andersson SRP 2.5 PlayStation 
OtherOtherOtherOther
PlayStation 2 Gamecube Dreamcast Xbox 
OtherOtherOther
Xbox 360 PlayStation 3 Wii 
  hide details  
Reply
 
Portable Desktop
(12 items)
 
Console Cave
(11 items)
 
CPUMotherboardGraphicsRAM
Intel Core i5 2500K ASRock P67 Extreme4 ASUS GTX 670 DCII Kingston HyperX Fury 
Hard DriveHard DriveHard DriveHard Drive
Seagate 600 Samsung Spinpoint F4 Seagate Barracuda Green Seagate Desktop 
Hard DriveHard DriveHard DriveHard Drive
Seagate FreeAgent GoFlex Seagate Expansion Desktop Maxtor Basics Seagate Backup Plus Portable 
Optical DriveCoolingCoolingCooling
Samsung SE-208AB Thermalright Archon (Rev. A) 2x Phanteks PH-F140HP_BL (CPU) 2x Noctua NF-A14 PWM (Intake) 
CoolingCoolingOSMonitor
Thermalright TY-150 (Exhaust) Noctua NF-S12A PWM (Exhaust) Windows 10 Pro DELL UltraSharp U2711 
KeyboardPowerCaseMouse
QPAD MK-80 Pro SeaSonic X-560 Lian Li PC-P50WB Func MS-3 R2 
Mouse PadAudioAudioAudio
SteelSeries QcK Creative Sound Blaster Z Swans M10 Creative Aurvana Live! 
Other
Lian Li LED50 
CPUMotherboardGraphicsGraphics
Intel Pentium G3258 ASUS H81I-PLUS Intel HD Graphics ASUS GTX 670 DirectCU II 
RAMHard DriveCoolingOS
Corsair Vengeance LP OCZ Vector Scythe Big Shuriken 2 rev. B Windows 10 Pro N 
MonitorMonitorPowerCase
LG PA70G BenQ G2420HDBL Cooler Master B500 ver.2 Cooler Master Elite 130 
MonitorAudioAudioOther
LG PA70G Yamaha HTR-3065 Andersson SRP 2.5 PlayStation 
OtherOtherOtherOther
PlayStation 2 Gamecube Dreamcast Xbox 
OtherOtherOther
Xbox 360 PlayStation 3 Wii 
  hide details  
Reply
post #174 of 177
Quote:
Originally Posted by Brutuz;15372978 
Post count means crap all, I have more posts than you, does that mean I know more than you?

Duckie has the most posts out of anyone in this forum that I know of, does that mean he knows more than all of us combined? (He probably does, but that's beside the point.)

I've actually seen quite a few Bobcat netbooks around where I live.

You're forgetting something though, the original Willamette P4s (1.4 and 1.5Ghz on Socket 428) were indeed slower than the Athlon XPs and Pentium IIIs, but the Northwoods were the fastest things you could buy before the Athlon64 and weren't that bad...The really bad P4s were the Prescotts, a massive rework of the core but put out too much heat, most of the P4s heat problem came from Prescott and the 90nm process being leaky.

You repeated pretty much what I'm saying, the original P4 launched at far too low of a clock speed to really be competitive, the Northwood P4s were at the right clock speeds to generally beat the AthlonXPs, but then the Athlon64 came out and Prescott turned out to be a dud.

And the K10.5 core (Tweaked K8) is generally faster than BD clock for clock because of the cache and front-end issues in BD, the L1 cache is too small (Especially since one part of it is shared) and the L2 and L3 caches are too slow (Nearly double the latency of a K10.5 cache) with the front-end decoder being too slow for two threads at once (It ends up doing one less op a clock than K10.5 when you have both pipelines working)

And yes, it's very true CMT may not work out as well as it was theoretically meant to, but at least unlike the P4 it wasn't designed for marketing from the start.

I think that the actual execution units are probably faster than the execution units in K10.5 per clock. Also, it seems that there is not a huge need for a larger L1 cache; the size of L2 and L3 compensate provided the latencies are as they should be (I agree with you that the latencies are too high, but I think that they are partially tied to yield issues). There is, to my knowledge, no solid evidence that the front-end is too narrow (though I suspect that it is in some circumstances).

The most likely culprit is still that terrible branch prediction. If Bulldozer is being forced to flush its (lengthy) pipeline 30% more often than Sandybridge because its branching misses 30% more, then all other issues become secondary. If the cache holds and transfers the instructions fast enough and the decoder decodes fast enough and the branch prediction "guesses" wrong, then performance will still suffer.
Quote:
Originally Posted by PoopaScoopa;15373095 
I was hoping for a linkedin profile or a company profile page on a website. This so called ex-employee doesn't seem to be credible. Even some of the things he said, such as saying hand placing cells is more efficient than machines seems a bit weird. The rest of what he said can be applied to every corporation - out with the old highly paid workers and in with the new unskilled low paid employees.
I have no further proof of his current or previous employment. As implied before, I somewhat doubt his credentials and even if they are true, his statements still do not withstand a careful cross examination (lawyer reference intended).
post #175 of 177
Quote:
Originally Posted by hajile;15375770 
I think that the actual execution units are probably faster than the execution units in K10.5 per clock. Also, it seems that there is not a huge need for a larger L1 cache; the size of L2 and L3 compensate provided the latencies are as they should be (I agree with you that the latencies are too high, but I think that they are partially tied to yield issues). There is, to my knowledge, no solid evidence that the front-end is too narrow (though I suspect that it is in some circumstances).

The most likely culprit is still that terrible branch prediction. If Bulldozer is being forced to flush its (lengthy) pipeline 30% more often than Sandybridge because its branching misses 30% more, then all other issues become secondary. If the cache holds and transfers the instructions fast enough and the decoder decodes fast enough and the branch prediction "guesses" wrong, then performance will still suffer.

The execution does one more op per clock than the K10.5 ones, but the K10.5 one services one core, whereas this services two, so while it's faster in single-threaded scenarios assuming nothing uses the other shared core, it's slower overall when the two shared cores are both pegged at 100%.

The cache would also be and issue, as would the branch prediction, if AMD gets even just the branch prediction and the front-end fixed that'll fix the low IPC and scaling of multiple cores. (eg. Just under 1 for single threaded Cinebench, 6 for 8 cores? That's nearing HT speeds)
    
CPUMotherboardGraphicsRAM
Intel Core i5 3570k @ 4.5Ghz ASRock Z77 Pro3 Powercolor Radeon HD7950 3GB @ 1150/1350 4x4GB G.Skill Ares 2000Mhz CL9 
Hard DriveHard DriveHard DriveHard Drive
Samsung 840 250GB Western Digital Black 1TB WD1002FAEX Seagate Barracuda 3TB ST3000DM001 Samsung Spinpoint EcoGreen 2TB 
Optical DriveCoolingCoolingCooling
Pioneer DVR-220LBKS Noctua NH-D14 Scythe Gentle Typhoon 1850rpm Corsair AF140 Quiet Edition 
CoolingOSMonitorMonitor
Arcitc Cooling Acclero Twin Turbo II Arch Linux x86-64, amdgpu BenQ G2220HD BenQ G2020HD 
KeyboardPowerCaseMouse
Ducky Shine III Year of the Snake, Cherry Blue Silverstone Strider Plus 600w CoolerMaster CM690 II Black and White SteelSeries Sensei Professional 
Mouse PadAudioOther
Artisan Hien Mid Japan Black Large ASUS Xonar DX NZXT Sentry Mesh 30w Fan Controller 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Intel Core i5 3570k @ 4.5Ghz ASRock Z77 Pro3 Powercolor Radeon HD7950 3GB @ 1150/1350 4x4GB G.Skill Ares 2000Mhz CL9 
Hard DriveHard DriveHard DriveHard Drive
Samsung 840 250GB Western Digital Black 1TB WD1002FAEX Seagate Barracuda 3TB ST3000DM001 Samsung Spinpoint EcoGreen 2TB 
Optical DriveCoolingCoolingCooling
Pioneer DVR-220LBKS Noctua NH-D14 Scythe Gentle Typhoon 1850rpm Corsair AF140 Quiet Edition 
CoolingOSMonitorMonitor
Arcitc Cooling Acclero Twin Turbo II Arch Linux x86-64, amdgpu BenQ G2220HD BenQ G2020HD 
KeyboardPowerCaseMouse
Ducky Shine III Year of the Snake, Cherry Blue Silverstone Strider Plus 600w CoolerMaster CM690 II Black and White SteelSeries Sensei Professional 
Mouse PadAudioOther
Artisan Hien Mid Japan Black Large ASUS Xonar DX NZXT Sentry Mesh 30w Fan Controller 
  hide details  
Reply
post #176 of 177
You've got to take this kind of stuff with a grain of salt. While definitely interesting to read everyone has their own take on things that may or may not completely reflect reality. That is assuming this guy really used to work at AMD as an engineer.

Still, an interesting read. Lets hops that AMD pulls it together, whatever the problem with Bulldozer was, and puts out some better product.
SBD:
(18 items)
 
  
CPUMotherboardGraphicsRAM
i7-3770k Gigabyte ga-z77x-up4 tb EVGA GTX 980 SC 32GB G.SKILL Trident X F3-1600C7Q-32GTX  
Hard DriveHard DriveOptical DriveCooling
Samsung 840 Pro 256GB SSD Western Digtal 2TB RE4 Plextor 24x DL Burner ThermalTake Water 2.0 Extreme 
OSMonitorMonitorKeyboard
Windows 7 Pro 64-bit Acer XB270HU 2560x1440, IPS-type panel, 144hz, ... EIZO FG2421 1920x1080 VA 120Hz QuickFire Cherry Blue 
PowerCaseMouseMouse Pad
Lepa G1000 1kw Corsair 550D Logitech G400 Roccat Taito 
AudioAudio
Xonar Essence ST Niles SI-275 Amplifier 
  hide details  
Reply
SBD:
(18 items)
 
  
CPUMotherboardGraphicsRAM
i7-3770k Gigabyte ga-z77x-up4 tb EVGA GTX 980 SC 32GB G.SKILL Trident X F3-1600C7Q-32GTX  
Hard DriveHard DriveOptical DriveCooling
Samsung 840 Pro 256GB SSD Western Digtal 2TB RE4 Plextor 24x DL Burner ThermalTake Water 2.0 Extreme 
OSMonitorMonitorKeyboard
Windows 7 Pro 64-bit Acer XB270HU 2560x1440, IPS-type panel, 144hz, ... EIZO FG2421 1920x1080 VA 120Hz QuickFire Cherry Blue 
PowerCaseMouseMouse Pad
Lepa G1000 1kw Corsair 550D Logitech G400 Roccat Taito 
AudioAudio
Xonar Essence ST Niles SI-275 Amplifier 
  hide details  
Reply
post #177 of 177
[ame]http://www.youtube.com/watch?v=tfGJ8bVKFPc[/ame]

Just wanted to post the video sorry for not being on topic
AMD FX ~Seronx
(16 items)
 
  
CPUMotherboardGraphicsRAM
FX-9800P Acer Wasp R7 M440 SK Hynix HMA41GS6AFR8N-TF 
Hard DriveHard DriveOptical DriveCooling
KINGSTON RBU-SNS8152S3128GG2 TOSHIBA MQ01ABD100 HL-DT-ST DVDRAM GUE1N Stock 
OSMonitorKeyboardPower
Microsoft Windows 10 Home Build 14393 Viewsonic XG2401 24 Hz-144 Hz Ducky Channel Shine 3 Stock 65W 
CaseMouseMouse PadAudio
Acer Exoskeleton Steelseries Rival 300 Razer Megasoma AMD-Realtek ALC255 
  hide details  
Reply
AMD FX ~Seronx
(16 items)
 
  
CPUMotherboardGraphicsRAM
FX-9800P Acer Wasp R7 M440 SK Hynix HMA41GS6AFR8N-TF 
Hard DriveHard DriveOptical DriveCooling
KINGSTON RBU-SNS8152S3128GG2 TOSHIBA MQ01ABD100 HL-DT-ST DVDRAM GUE1N Stock 
OSMonitorKeyboardPower
Microsoft Windows 10 Home Build 14393 Viewsonic XG2401 24 Hz-144 Hz Ducky Channel Shine 3 Stock 65W 
CaseMouseMouse PadAudio
Acer Exoskeleton Steelseries Rival 300 Razer Megasoma AMD-Realtek ALC255 
  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