Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [pcgames] Zen engineering samples in the wild. Units up to the 32 core are being tested.
New Posts  All Forums:Forum Nav:

[pcgames] Zen engineering samples in the wild. Units up to the 32 core are being tested. - Page 14

post #131 of 212
There are no "real" cores and hyper threads.An intel Hyperthreaded core can spawn two logical threads which are treated similarly by Windows. Both threads have access to the core's hardware (logic). What the OS does is to make sure all hardware resources are managed properly, eg not forcing an already busy core to run two virtual threads when another idle core is available.
Mastodon Ryzen
(12 items)
 
HP Z220
(8 items)
 
 
CPUMotherboardGraphicsRAM
R7 1800X Asus Crosshair VI Hero Sapphire RX Vega 64 reference Gskill TridentZ 
Hard DriveOptical DriveCoolingOS
Pny SSD 240GB Crucial MX100 CM Nepton 280L Win 10 
MonitorPowerCaseMouse
Acer Predator XG270HU Freesync XFX 750W Pro HAF XM Logitech G502 
CPUMotherboardGraphicsCooling
i7 3770 HP Quadro K2000 HP 
OSPowerCaseMouse
Win 7  HP 400W HP CMT RAT 7 
  hide details  
Reply
Mastodon Ryzen
(12 items)
 
HP Z220
(8 items)
 
 
CPUMotherboardGraphicsRAM
R7 1800X Asus Crosshair VI Hero Sapphire RX Vega 64 reference Gskill TridentZ 
Hard DriveOptical DriveCoolingOS
Pny SSD 240GB Crucial MX100 CM Nepton 280L Win 10 
MonitorPowerCaseMouse
Acer Predator XG270HU Freesync XFX 750W Pro HAF XM Logitech G502 
CPUMotherboardGraphicsCooling
i7 3770 HP Quadro K2000 HP 
OSPowerCaseMouse
Win 7  HP 400W HP CMT RAT 7 
  hide details  
Reply
post #132 of 212
Quote:
Originally Posted by epic1337 View Post

where did i say hyper-threading creates a smaller core?

thats why its more beneficial to simply add more cores, or even a bunch of smaller cores as they're an independent resource that can benefit the other threads, instead of simply using hyper-threading.

Its beneficial to add more homogeneous cores, with hyper-threading, so the software developer can utilize them as he sees fit.
post #133 of 212
Quote:
Originally Posted by Lee Patekar View Post

Its beneficial to add more homogeneous cores, with hyper-threading, so the software developer can utilize them as he sees fit.
applications and programs already defaults to using core0~3, its how microsoft programmed their scheduler.
they simply go for the sequence "fill up 0 to 3", this is why core0 almost always have more load than the rest of the cores.

which means, you only need to assign the big cores to 0~3, and the little cores to 4~7, the applications and programs will automatically fill them up.
post #134 of 212
Quote:
Originally Posted by epic1337 View Post

applications and programs already defaults to using core0~3, its how Windows programmed them, they simply go for the sequence "fill up 0 to 3", this is why core0 almost always have more load than the rest of the cores.

which means, you only need to assign the big cores to 0~3, and the little cores to 4~7, the applications and programs will automatically fill them up.

... do you even code?

Core 0 and Core 1 are the same physical core. You can query the core mapping to see what logical core is connected to which physical core and on which package. You can then assign threads to them yourself if you choose.

You don't change the CPU to accommodate the software, you write software optimized for your target hardware. And in this case, target hardware has homogeneous cores, sometimes with hyper-threading. You set your affinities accordingly.

Or you do the lazy thing and let windows throw the threads wherever.. in which case it will stuff the highest priority thread on the first physical core and balance the rest according to some algorithm... and knowing Microsoft its most likely sub-optimal.
Edited by Lee Patekar - 7/23/16 at 11:37am
post #135 of 212
Quote:
Originally Posted by Lee Patekar View Post

... do you even code?

Core 0 and Core 1 are the same physical core. You can query the core mapping to see what logical core is connected to which physical core and on which package. You can then assign threads to them yourself if you choose.

You don't change the CPU to accommodate the software, you write software optimized for your target hardware. And in this case, target hardware has homogeneous cores, sometimes with hyper-threading. You set your affinities accordingly.

Or you do the lazy thing and let windows throw the threads wherever.. in which case it will stuff the highest priority thread on the first physical core and balance the rest according to some algorithm... and knowing Microsoft its most likely sub-optimal.

isn't that what i just said?
post #136 of 212
Quote:
Originally Posted by epic1337 View Post

isn't that what i just said?

Following what you say is difficult when you refer to hyper-threading as one main core with a little one.

Anyway I'm out. Good luck pushing your 4+2 idea.
post #137 of 212
Quote:
Originally Posted by Lee Patekar View Post

Following what you say is difficult when you refer to hyper-threading as one main core with a little one.

Anyway I'm out. Good luck pushing your 4+2 idea.

do you even read?
Quote:
Originally Posted by epic1337 View Post

the purpose of hyper threading is to utilize what remains of the resource in the main core, making each core more efficient in managing resources.
which means its also much slower than the main thread, otherwise the threads would forcibly clash with each other for resource.

my point in bringing up hyper threading is that, its not even a core, nor does these extra threads even perform as fast as a real core.
your point was to make a CPU with many cores which are all fast cores, adding hyper-threading simply contradicts this as half of the CPU threads aren't as fast as real cores.
Edited by epic1337 - 7/23/16 at 11:56am
post #138 of 212
Quote:
Originally Posted by TheBloodEagle View Post

No, not for mobile. But I mean the first 4 cores would have no HT but be high clock (5Ghz) and then rest of the cores, say the 4 others in an 8 core would be 3 or 4 way hyperthreaded (similar to Power8 but less) at a lower clock.

Ohhhh, I see what you mean. Sure, sure. I'm out of the loop here and, to be perfectly honest, I'm not sure I care enough to learn. frown.gif That sounds terrible, haha. But, I'll admit it: I'm only on OCN to make my games run faster. I am, in the end, just a benchmark interpreter. If it runs faster and is within my budget, I'll buy it--whatever black magic they need to do, I'm not super concerned about (sans ethical concerns, hahahaha, but that's a different topic).
The 2013 Build
(14 items)
 
  
CPUMotherboardGraphicsRAM
--- SOLD --- --- SOLD ---  --- SOLD --- --- SOLD --- 
Hard DriveOptical DriveCoolingOS
Samsung 850 EVO 500GB ASUS DRW-24B3ST (24X DVD burner) Noctua NH-U14S w/ NH-T1 TIM Windows 10 Pro (x64) 
MonitorKeyboardPowerCase
ASUS VG248QE w/ Lightboost @ 120Hz Ducky Shine III (MX Brown w/ Red backlights) Kingwin LZP-1000 (1000W) PSU Corsair Carbide Air 540 
Mouse Pad
Razer Goliathus (Precision Control model) - Sta... 
  hide details  
Reply
The 2013 Build
(14 items)
 
  
CPUMotherboardGraphicsRAM
--- SOLD --- --- SOLD ---  --- SOLD --- --- SOLD --- 
Hard DriveOptical DriveCoolingOS
Samsung 850 EVO 500GB ASUS DRW-24B3ST (24X DVD burner) Noctua NH-U14S w/ NH-T1 TIM Windows 10 Pro (x64) 
MonitorKeyboardPowerCase
ASUS VG248QE w/ Lightboost @ 120Hz Ducky Shine III (MX Brown w/ Red backlights) Kingwin LZP-1000 (1000W) PSU Corsair Carbide Air 540 
Mouse Pad
Razer Goliathus (Precision Control model) - Sta... 
  hide details  
Reply
post #139 of 212
Quote:
Originally Posted by epic1337 View Post

the fact that hyper-threading is being applied agrees with my conjecture, they're fitting more threads even if they're slower than the main cores.

This is exactly the opposite of what SMT/hyperthreading is.

Moving to SMT rather than CMT allows fewer, stronger, cores to achieve the same level of multi-threaded performance as more weaker ones. SMT has no performance penalty to anything unless the additional logical cores are actually used.
Quote:
Originally Posted by epic1337 View Post

which means its also much slower than the main thread, otherwise the threads would forcibly clash with each other for resource.

There is no distinction between the logical CPUs presented for each core. There are finite execution and cache resources, and heavily utilizing one logical core will reduce the performance the other can provide, but a core with with SMT enabled is no slower than the same core without SMT enabled, if only one logical core is being utilized.
Quote:
Originally Posted by Slomo4shO View Post

What exactly have you seen?

All you need to come to his conclusion is that Zen has SMT and Zen has symmetric cores.
Quote:
Originally Posted by Lee Patekar View Post

Most software developers don't apply affinity masks, they just set thread priorities and let windows (or unix or QNX) handle the scheduling across all cores.

This is true, but most modern OSes are very HT aware and tend to schedule threads pretty well.
Quote:
Originally Posted by epic1337 View Post

applications and programs already defaults to using core0~3, its how microsoft programmed their scheduler.

This is not what I've seen..

By and large, unless the program has an affinity mask to tell them otherwise, Windows will keep demanding threads from sharing the same cores as long as possible.
Edited by Blameless - 7/23/16 at 1:12pm
Primary
(15 items)
 
Secondary
(13 items)
 
In progress
(10 items)
 
CPUMotherboardGraphicsRAM
5820K @ 4.2/3.5GHz core/uncore, 1.175/1.15v Gigabyte X99 SOC Champion (F22n) Gigabyte AORUS GTX 1080 Ti (F3P) @ 2025/1485, 1... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.34v 
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.06) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Noctua NH-D15 
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x 
CaseMouseAudio
Fractal Design Define R4 Logitech G402 Realtek ALC1150 + M-Audio AV40 
CPUMotherboardGraphicsRAM
X5670 @ 4.4/3.2GHz core/uncore, 1.36 vcore, 1.2... Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Sapphire Fury Nitro OC+ @ 1053/500, 1.225vGPU/1... 2x Samsung MV-3V4G3D/US @ 2000, 10-11-11-30-T1,... 
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 2000, 10-11-11-3... OCZ (Toshiba) Trion 150 120GB Hyundai Sapphire 120GB 3x Hitachi Deskstar 7k1000.C 1TB 
CoolingOSPowerCase
Noctua NH-D14 Windows 7 Pro x64 SP1 Antec TP-750 Fractal Design R5 
Audio
ASUS Xonar DS 
CPUMotherboardGraphicsRAM
i7-6800K @ 4.3/3.5GHz core/uncore, 1.36/1.2v ASRock X99 OC Formula (P3.10) GTX 780 (temporary) 4x4GiB Crucial DDR4-2400 @ 11-13-12-28-T2, 1.33v 
Hard DriveHard DriveCoolingOS
Intel 600p 256GB NVMe 2x HGST Travelstar 7k1000 1TB Corsair H55 (temporary) Windows Server 2016 Datacenter 
PowerCase
Seasonic SS-860XP2 Corsair Carbide Air 540 
  hide details  
Reply
Primary
(15 items)
 
Secondary
(13 items)
 
In progress
(10 items)
 
CPUMotherboardGraphicsRAM
5820K @ 4.2/3.5GHz core/uncore, 1.175/1.15v Gigabyte X99 SOC Champion (F22n) Gigabyte AORUS GTX 1080 Ti (F3P) @ 2025/1485, 1... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.34v 
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.06) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Noctua NH-D15 
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x 
CaseMouseAudio
Fractal Design Define R4 Logitech G402 Realtek ALC1150 + M-Audio AV40 
CPUMotherboardGraphicsRAM
X5670 @ 4.4/3.2GHz core/uncore, 1.36 vcore, 1.2... Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Sapphire Fury Nitro OC+ @ 1053/500, 1.225vGPU/1... 2x Samsung MV-3V4G3D/US @ 2000, 10-11-11-30-T1,... 
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 2000, 10-11-11-3... OCZ (Toshiba) Trion 150 120GB Hyundai Sapphire 120GB 3x Hitachi Deskstar 7k1000.C 1TB 
CoolingOSPowerCase
Noctua NH-D14 Windows 7 Pro x64 SP1 Antec TP-750 Fractal Design R5 
Audio
ASUS Xonar DS 
CPUMotherboardGraphicsRAM
i7-6800K @ 4.3/3.5GHz core/uncore, 1.36/1.2v ASRock X99 OC Formula (P3.10) GTX 780 (temporary) 4x4GiB Crucial DDR4-2400 @ 11-13-12-28-T2, 1.33v 
Hard DriveHard DriveCoolingOS
Intel 600p 256GB NVMe 2x HGST Travelstar 7k1000 1TB Corsair H55 (temporary) Windows Server 2016 Datacenter 
PowerCase
Seasonic SS-860XP2 Corsair Carbide Air 540 
  hide details  
Reply
post #140 of 212
Quote:
Originally Posted by Blameless View Post

This is exactly the opposite of what SMT/hyperthreading is.
Moving to SMT rather than CMT allows fewer, stronger, cores to achieve the same level of multi-threaded performance as more weaker ones. SMT has no performance penalty to anything unless the additional logical cores are actually used.


There is no distinction between the logical CPUs presented for each core. There are finite execution and cache resources, and heavily utilizing one logical core will reduce the performance the other can provide, but a core with with SMT enabled is no slower than the same core without SMT enabled, if only one logical core is being utilized.


This is not what I've seen..
By and large, unless the program has an affinity mask to tell them otherwise, Windows will keep demanding threads from sharing the same cores as long as possible.

except when both threads requires too much resource on the cores, this rarely happens now a days, but does occur when HT-off performs better than HT-on.
plus dedicated resource is better than shared resource, at least on the performance stand-point.


yes i realize that, though there is a distinction on the scheduler level, at least windows could sort of discern when one logical core is loaded and the other shared logical core can't be fully used.
it could be interpreted as a "ratio" of occupancy i suppose, when the core is lightly loaded on both logical cores the ratio is 50-50, meaning they perform the same.
but if the core is heavily loaded on one logical core the ratio falls, for example 67:33, meaning the other thread will only perform at half the performance compared to the other.
theres also an issue with what threads you can put on the same core, presumably HT can perform better when both threads use the same cached data.


i think i poorly worded it, i mean it like "if 0 is full, next is 1, then 2, etc.", unless otherwise the application requests for more cores.
Edited by epic1337 - 7/23/16 at 1:46pm
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Rumors and Unconfirmed Articles
Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [pcgames] Zen engineering samples in the wild. Units up to the 32 core are being tested.