Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [Seeking Alpha] AMD: More Confirmation Of Zen Troubles
New Posts  All Forums:Forum Nav:

[Seeking Alpha] AMD: More Confirmation Of Zen Troubles - Page 6

post #51 of 114
Quote:
Originally Posted by looncraz View Post

Orochi is almost certainly 8100 clocked to 3Ghz, and the Zen is also at 3GHz.

For a performance comparison with your CPU, clock it to 3GHz and realize that we are talking about multithreaded performance on eight core CPUs.

Actually AMD is talking about a per-core stance. Zen 2SMT core is 2x an 8100's 2CMT core's performance at same clock.

And zen has 2x the cores as well.

Not surprising how many of the trolls missed this bit.

Remember, Zen was designed by THE best CPU designer of the modern era and the reason Bulldozer sucked is they removed the SMT scheduler from the prototypes in a risky move that didn't pay off. I don't wanna get too technical but the basic is that Zen has CMT-under-SMT, it has twice the math power per core as an 8350 at a matching clock and the SMT scheduler on top of that to make it all work slick. On top of that the density of 14nm allows it to have twice the processors to begin with... so basically a Zen 8 core will be about 3.5x more powerful than a Vishera pulled to the same clock speed and the SMT function will reduce a great deal of the latency that Poopdoozie chips suffered... and they've strapped DDR4 support onto it too.

Vishera = 2,3,4 core chips with twin CMT sub-cores
Zen = 4,8 core SMT chips with quad CMT sub-cores

CMT is Concurrent Multi-Thread... meaning that the cpu using both math cores can ONLY do the maths and has to be fed the maths by the OS. When you're only doing the kind of maths these are designed for it works pretty good. my 2.3GHz twin G34 Opteron (two 8/16 7360s) eats Folding@Home like candy but can't do jack else without slowing way down. It stomps out the CPU work twice as fast as my 3.5GHz i7-4930 running 12 threads. That's 73.6 vs 42 when you do simple clock*core yet the Opteron board easily produces a little over twice as fast because the folding@home work is CPU aware and leverages those little CMTs.

SMT is Simultaneous Multi-Thread. This is what intel used to offer on i5's until they discovered people were AOK paying twice as much for a processor.

What SMT excels at over CMT is dealing with interruptions. Things like the OS, Antivirus, Skype, streaming and everything else. So when you have a SMT processor you can get away with having tons of useless tasks loitering around in memory since they won't really slow down your main 3 or 4 threads. This is why the i5's work so well. You tune em up and get your ducks all passive and they make great little gaming boxes... unless you're used to i7 SMT. Leveraging the SMT as a functional utility on the operating system makes a huge difference too. It isn't necessary in gaming. Hell, if Zen came out without SMT it'd still be a monster of a chip. With SMT 4 core Zen APU's that have 16 CMT math processing and a fairly robust 14nm R400 APU core set on the side they'll really rock.

Basically AMD may have caught up with intel in the last 12 months. intel has been almost completely idle for more than two years just bringing lab tech to consumer use.

... and you Know that AMD is gonna price Zen sweetly.
Edited by prjindigo - 9/4/16 at 3:54am
post #52 of 114
fx has 8 integer unit and 8 floating point that can if i understand well become 4 under specific task. rolleyes.gif
Gaming PC
(18 items)
 
HTPC v2
(11 items)
 
 
CPUMotherboardGraphicsRAM
Amd Fx 8300 95w  GIGABYTE GA-78LMT V 6.0 Asus ENGTS250 Dark Knight 1gb.... it display 10... 2x2gb + 2x4gb 1333 cl9 12gb 
Hard DriveCoolingOSMonitor
128gb ssd + 500gb 5400rpm Arctic Cooling Freezer i32 Plus Windows 10 64bit Pro Wasabi Mango UHD400 4k 60hz 
PowerCaseAudio
SeaSonic S12II 620 DEEPCOOL DUKASE WHV2 White SMSL SA-50 to Fluance SX6 Bookshelf 
  hide details  
Reply
Gaming PC
(18 items)
 
HTPC v2
(11 items)
 
 
CPUMotherboardGraphicsRAM
Amd Fx 8300 95w  GIGABYTE GA-78LMT V 6.0 Asus ENGTS250 Dark Knight 1gb.... it display 10... 2x2gb + 2x4gb 1333 cl9 12gb 
Hard DriveCoolingOSMonitor
128gb ssd + 500gb 5400rpm Arctic Cooling Freezer i32 Plus Windows 10 64bit Pro Wasabi Mango UHD400 4k 60hz 
PowerCaseAudio
SeaSonic S12II 620 DEEPCOOL DUKASE WHV2 White SMSL SA-50 to Fluance SX6 Bookshelf 
  hide details  
Reply
post #53 of 114
@prjindigo

Orochi = 2, 3, 4 / Dual-core Multi[Quad]-Module Processor.
Summit Ridge = 4, 8 / Quad-core Multi[Dual]-Complex Processor.

Clustered Multithreading for AMD is two cores/two threads with parallel resources, a shared front-end, and a shared floating point unit.
[Two retires, two schedulers, two sets of execution datapaths, two load/store units, two level 1 data caches = parallel @ lower complexity]

Simultaneous Multithreading for AMD is a single core/two threads with concurrent resources, a shared front-end, and a shared floating point unit.
[One retire, one scheduler, one set of execution datapaths, one load/store unit, one level 1 data cache = concurrent @ higher complexity]

[For Bulldozer:]
A shared front-end and floating point unit has no impact on performance. The cause of the lower performance is instead the L1 instruction cache and L2 cache. Which does not do reads and writes to both threads in parallel. To increase the port width of both, would increase the complexity, which requires more area in particular for added control logic and buffers. The increase in port width would be cheaper, than adding an extra L1 instruction cache and extra L2 cache.

The purpose of the module is two cores/threads operating like a GPU when single program, multiple data is executed. Why fetch twice, when the architecture can fetch once? Writes to memory can also be merged if they are lower half and upper half of the same data points. The above problem with caches show the issue when multiple programs, multiple data is ran on the module.

The two cores means there isn't much an impact as hyperthreading, which can go down >10% in processor performance when heavy SPMD/light MPMD. The SMT performance drop is fixed in Clustered Simultaneous Multithreading. Which fuses the two above ideas into one and adds both thread contexts between the cores. So, two cores can operate on a single thread[high instruction throughput] or a single core on two threads[low instruction throughput], etc.
Edited by Seronx - 9/4/16 at 6:08am
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
post #54 of 114
Quote:
Originally Posted by prjindigo View Post

Actually AMD is talking about a per-core stance.

Orochi is, specifically, an eight-core Bulldozer die. So they're comparing octo-core CPUs.
Quote:
Originally Posted by prjindigo View Post

Zen 2SMT core is 2x an 8100's 2CMT core's performance at same clock.

And zen has 2x the cores as well.

I think this is a mischaracterization of what CMT is. There's no such thing as a "2CMT core." CMT is a physical form of multiple cores, not one core. A module consists of two tightly interwoven cores... sharing one highly flexible FPU which can concurrently execute two threads (basically, the FPU uses an SMT design). Bulldozer's problem is that it only has two ALUs without the supporting infrastructure to enable 100% utilization. The fact that AMD managed as much as they did with just two ALUs is actually quite impressive... AMD started adding capabilities to the other pipelines to make up for their deficit in ALU resources right away.
Quote:
Originally Posted by prjindigo View Post

Remember, Zen was designed by THE best CPU designer of the modern era and the reason Bulldozer sucked is they removed the SMT scheduler from the prototypes in a risky move that didn't pay off. I don't wanna get too technical but the basic is that Zen has CMT-under-SMT, it has twice the math power per core as an 8350 at a matching clock and the SMT scheduler on top of that to make it all work slick. On top of that the density of 14nm allows it to have twice the processors to begin with... so basically a Zen 8 core will be about 3.5x more powerful than a Vishera pulled to the same clock speed and the SMT function will reduce a great deal of the latency that Poopdoozie chips suffered... and they've strapped DDR4 support onto it too.

SMT takes much more than just a scheduler to be successful. Intel began with little more than a scheduler for Hyper-Threading and we know that the first version was quite terrible, often losing total performance... but the system would feel smoother because Windows XP was able to schedule two hardware threads, allowing its broken I/O system with its horrid stalls to be bypassed more often.
Quote:
Originally Posted by prjindigo View Post

CMT is Concurrent Multi-Thread... meaning that the cpu using both math cores can ONLY do the maths and has to be fed the maths by the OS. When you're only doing the kind of maths these are designed for it works pretty good.

That is always the case, 100% of the time. SMT relies on the OS to schedule for each thread just like CMT relies on the OS to schedule for each core/thread. Even with one core and one thread the OS is responsible for scheduling tasks on the CPU... and it becomes even more critical how well it can do so.
Quote:
Originally Posted by prjindigo View Post


SMT is Simultaneous Multi-Thread. This is what intel used to offer on i5's until they discovered people were AOK paying twice as much for a processor.

You do know that Intel calls SMT "Hyper-Threading," right? And they only offer it on i3, i7 or Xeon. AFAIK, they never had an i5 with HT. Pentium 4 HT was an SMT enabled processor, then Intel canceled SMT because it did so little and they switched to a high IPC design which did not support it. They brought it back because it allowed them to create a higher level of market segmentation. Higher IPC CPUs are better with SMT than low IPC CPUs (Bulldozer would be horrible with it.. hard to share just two ALUs...).
Quote:
Originally Posted by prjindigo View Post

What SMT excels at over CMT is dealing with interruptions. Things like the OS, Antivirus, Skype, streaming and everything else. So when you have a SMT processor you can get away with having tons of useless tasks loitering around in memory since they won't really slow down your main 3 or 4 threads. This is why the i5's work so well. You tune em up and get your ducks all passive and they make great little gaming boxes... unless you're used to i7 SMT. Leveraging the SMT as a functional utility on the operating system makes a huge difference too. It isn't necessary in gaming. Hell, if Zen came out without SMT it'd still be a monster of a chip. With SMT 4 core Zen APU's that have 16 CMT math processing and a fairly robust 14nm R400 APU core set on the side they'll really rock.

Exactly wrong. SMT adds about 20~25% more, CMT adds 85~90% more. One SMT core versus one CMT module... with equal IPC... CMT is MUCH better.
Quote:
Originally Posted by prjindigo View Post

Basically AMD may have caught up with intel in the last 12 months. intel has been almost completely idle for more than two years just bringing lab tech to consumer use.

... and you Know that AMD is gonna price Zen sweetly.

Intel hasn't been idle, they just haven't been doing anything other than their tick tock strategy - which is a safe way for them to progress when they have no competition other than their existing systems. Intel will still sell all of the CPUs they can manage to make, so they have little to lose from an in-stride pass.

AMD, though, did not have that luxury. Zen is their hail marry pass. Zen+ is a very special move to attempt total parity, or even slight superiority. Next, AMD is trying to skip 10nm and jump to 7nm.. possibly beating Intel to the punch if all goes as planned (doubtful).
post #55 of 114
Quote:
Originally Posted by Seraphic View Post

It seems Zen might limited to around 3.0Ghz. =D
https://www.linkedin.com/pulse/amd-new-zen-cpusapus-least-10-market-share-servers-pcs-luskin?published=u

Zen will be a 4GHz CPU all day long.

8-core 95W will be clocked at or near Intel's 8-core CPUs. Quad cores will be clocked similar to Intel quad cores... Haswell, anyway.

Not sure if 4GHz will be a base clock or a turbo clock, but Zen has all the hallmarks of a high-clocking design. AMD did not strip any of the high clockrate features from Zen that they had with Excavator and 14nm LPP isn't much of a limiting factor (provided they distribute the clock bus and power properly...). The only potential limiter of clock speed is the new low latency cache system.
post #56 of 114
Seeking Alpha biggrin.gif
right
post #57 of 114
Quote:
Originally Posted by looncraz View Post

Zen will be a 4GHz CPU all day long.

8-core 95W will be clocked at or near Intel's 8-core CPUs. Quad cores will be clocked similar to Intel quad cores... Haswell, anyway.

Not sure if 4GHz will be a base clock or a turbo clock, but Zen has all the hallmarks of a high-clocking design. AMD did not strip any of the high clockrate features from Zen that they had with Excavator and 14nm LPP isn't much of a limiting factor (provided they distribute the clock bus and power properly...). The only potential limiter of clock speed is the new low latency cache system.

Not convinced that it'll do ~4GHz base (I'm fully expecting GloFo to be the weak link there, at least for the first iteration), but you're right that Zen definitely looks like a high-clock design. RE the cache, IIRC they put the caches (or at least the L3) on a separate clock domain, so it shouldn't be much of a limiter. High-3Ghz range stock boost clocks and overclocks into the low-mid 4GHz without too much difficulty is definitely possible, I think.
SPEAR
(22 items)
 
Laptop #3
(7 items)
 
 
CPUMotherboardGraphicsRAM
i5 2500k @ 4.2ghz ASRock Z68 Extreme3 Gen3 Gigabyte WF3 Radeon R9 270X x2 G.Skill Ripjaws X 4x4gb @ 1600 
Hard DriveHard DriveHard DriveHard Drive
OCZ Agility 3 120gb (Linux) OCZ Agility 3 120gb (Windows) WD Caviar Black 2tb Hitachi 500gb 
Optical DriveOptical DriveCoolingOS
Sony Optiarc LightScribe LaCie external BluRay writer Antec Kühler H2O 920 + Corsair SP120 P/P Arch Linux + radeonsi + E20 
OSMonitorKeyboardPower
Windows 10 Pro (Insider release) ASUS VS238H-V @ 70hz Corsair K90 Corsair TX750 V2 
CaseMouseAudioAudio
None CM Storm Recon Altec Lansing ATP3 Creative Aurvana Live! 
OtherOther
Masscool Shin-Etsu G751 ASUS PCE-N10 
CPUGraphicsGraphicsRAM
Intel i7 4980HQ @ 2.8/4.0GHz Intel Iris Pro 5200 AMD Radeon R9 M370X 2x8GB DDR3 @ 1600MHz 
Hard DriveOSMonitor
1TB PCIe SSD OS X 10.11.2 El Capitan 15" @ 2880x1800 
  hide details  
Reply
SPEAR
(22 items)
 
Laptop #3
(7 items)
 
 
CPUMotherboardGraphicsRAM
i5 2500k @ 4.2ghz ASRock Z68 Extreme3 Gen3 Gigabyte WF3 Radeon R9 270X x2 G.Skill Ripjaws X 4x4gb @ 1600 
Hard DriveHard DriveHard DriveHard Drive
OCZ Agility 3 120gb (Linux) OCZ Agility 3 120gb (Windows) WD Caviar Black 2tb Hitachi 500gb 
Optical DriveOptical DriveCoolingOS
Sony Optiarc LightScribe LaCie external BluRay writer Antec Kühler H2O 920 + Corsair SP120 P/P Arch Linux + radeonsi + E20 
OSMonitorKeyboardPower
Windows 10 Pro (Insider release) ASUS VS238H-V @ 70hz Corsair K90 Corsair TX750 V2 
CaseMouseAudioAudio
None CM Storm Recon Altec Lansing ATP3 Creative Aurvana Live! 
OtherOther
Masscool Shin-Etsu G751 ASUS PCE-N10 
CPUGraphicsGraphicsRAM
Intel i7 4980HQ @ 2.8/4.0GHz Intel Iris Pro 5200 AMD Radeon R9 M370X 2x8GB DDR3 @ 1600MHz 
Hard DriveOSMonitor
1TB PCIe SSD OS X 10.11.2 El Capitan 15" @ 2880x1800 
  hide details  
Reply
post #58 of 114
Quote:
Originally Posted by looncraz View Post

Zen will be a 4GHz CPU all day long.

8-core 95W will be clocked at or near Intel's 8-core CPUs. Quad cores will be clocked similar to Intel quad cores... Haswell, anyway.

Not sure if 4GHz will be a base clock or a turbo clock, but Zen has all the hallmarks of a high-clocking design. AMD did not strip any of the high clockrate features from Zen that they had with Excavator and 14nm LPP isn't much of a limiting factor (provided they distribute the clock bus and power properly...). The only potential limiter of clock speed is the new low latency cache system.

Probably higher clock are the reason for delay?
Basically I will be buying new CPU in next few months... I really wanna get ZEN. I hope it will be good.
post #59 of 114
The article's methodology of estimating performance is fine. The only problem is that it hinges on one big IF condition. I think AMD was comparing to an 8350 but if they were comparing to an 8100 the signs are there. It's not like anyone preorders CPUs like they do a video game so this only matters to stock investors who want to and already did speculate before release.
post #60 of 114
Seeking what? Seeking what alpha? Seriously these guys are scrubs to stock markets and blatant manipulators.

The performance increase is not from bulldozer but excavator which means Seeking Alpha is full retard.
Power Tower
(22 items)
 
SteamBox
(9 items)
 
Doge Miner
(7 items)
 
CPUMotherboardGraphicsRAM
Ryzen 1700X AX370-Gaming 5 AMD Radeon R9 200 Series G.Skill DDR4-2400 
RAMRAMRAMHard Drive
G.Skill DDR4-2400 G.Skill DDR4-2400 G.Skill DDR4-2400 Samsung 840 Pro 
Hard DriveHard DriveHard DriveHard Drive
CX300 Crucial 480GB Toshiba 4TB Toshbia 4TB Western Digital Black 1TB 
CoolingOSMonitorMonitor
h110i Windows 10 42" LG TV 20" Digitizer ASUS 
KeyboardPowerCaseMouse
Corsair Vengeance Mechanical Keyboard  850watt Vampire Gold Rated NZXT S340 Elite Corsair RGB FPS Mouse 
Mouse PadAudio
Borderlands Mousepad Realtek HD 
  hide details  
Reply
Power Tower
(22 items)
 
SteamBox
(9 items)
 
Doge Miner
(7 items)
 
CPUMotherboardGraphicsRAM
Ryzen 1700X AX370-Gaming 5 AMD Radeon R9 200 Series G.Skill DDR4-2400 
RAMRAMRAMHard Drive
G.Skill DDR4-2400 G.Skill DDR4-2400 G.Skill DDR4-2400 Samsung 840 Pro 
Hard DriveHard DriveHard DriveHard Drive
CX300 Crucial 480GB Toshiba 4TB Toshbia 4TB Western Digital Black 1TB 
CoolingOSMonitorMonitor
h110i Windows 10 42" LG TV 20" Digitizer ASUS 
KeyboardPowerCaseMouse
Corsair Vengeance Mechanical Keyboard  850watt Vampire Gold Rated NZXT S340 Elite Corsair RGB FPS Mouse 
Mouse PadAudio
Borderlands Mousepad Realtek HD 
  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 › [Seeking Alpha] AMD: More Confirmation Of Zen Troubles