Overclock.net › Forums › Industry News › Hardware News › [AT] AMD Zen Microarchitecture: Dual Schedulers, Micro-Op Cache and Memory Hierarchy Revealed
New Posts  All Forums:Forum Nav:

[AT] AMD Zen Microarchitecture: Dual Schedulers, Micro-Op Cache and Memory Hierarchy Revealed - Page 23

post #221 of 233

I tend to not include eastern Europe in such comparisons because it is, in terms of anything that has to do with acquisitive power and basic goods/utilities, completely different to western Europe.

 

For the price of a 4 bedroom flat in Poland you can't buy a shack in the middle of nowhere in this half of the continent. And the average wage in Hungary is less than the minimum legal wage in Spain and four times less than the average wage in Britain.


Edited by Artikbot - 9/26/16 at 2:40am
   
AGP bencher
(14 items)
 
CPUMotherboardGraphicsRAM
Ryzen R7 1700 Gigabyte GA-AX370-Gaming 5 Sapphire HD 6950 2GiB 2x8GB KFA2 HOF DDR4-3600 
Hard DriveHard DriveHard DriveHard Drive
Crucial MX100 256GB Seagate 600 Series 240GB Seagate 7200.14 2TB Samsung F3 1TB 
CoolingCoolingCoolingCooling
EKWB Supreme HF XSPC Rasa GPU EK XT360 EK 4.0 
OSMonitorMonitorKeyboard
W10 Pro LG IPS235 LG E2250V KUL ES-87 
PowerCaseMouseAudio
SF Leadex II 650W Lian Li PC-A05NB Logitech G9 Xonar DX 
AudioAudio
SMSL SA-S3+Technics CB-250 Sennheiser HD555 
CPUMotherboardRAMHard Drive
AMD A10-5700 Gigabyte F2A75M-HD2 G.SKILL Ares 2133 CL9 Hitachi 5K750 
Hard DriveCoolingOSMonitor
Momentus .7 200GB Noctua NH-L9a Server 2012 R2 Standard AUO B156HW01 
PowerCaseOther
PicoPSU-80-WI-25V AIO Aluminium Handmade TP-Link Archer Something Something Wi-Fi AC 
CPUCPUCPUMotherboard
Core2Duo E6400 Core2Quad Q6600 Pentium Dual Core E5200 AsRock 4COREDUAL-SATA2 R2.0 
GraphicsRAMHard DriveOptical Drive
A dumpload of ancient AGP cards Kingston Value DDR2-667 CL4 2T @CL3 1T Seagate 160GB 7200.10 LG IDE DVD-ROM 
CoolingCoolingOSMonitor
Ghettomade CPU waterblock 49cc 2stroke engine copper radiator WinXP SP2 32bit ProView 17" 
PowerCase
Tacens Radix V 550W Ghetto aluminium bench 
  hide details  
Reply
   
AGP bencher
(14 items)
 
CPUMotherboardGraphicsRAM
Ryzen R7 1700 Gigabyte GA-AX370-Gaming 5 Sapphire HD 6950 2GiB 2x8GB KFA2 HOF DDR4-3600 
Hard DriveHard DriveHard DriveHard Drive
Crucial MX100 256GB Seagate 600 Series 240GB Seagate 7200.14 2TB Samsung F3 1TB 
CoolingCoolingCoolingCooling
EKWB Supreme HF XSPC Rasa GPU EK XT360 EK 4.0 
OSMonitorMonitorKeyboard
W10 Pro LG IPS235 LG E2250V KUL ES-87 
PowerCaseMouseAudio
SF Leadex II 650W Lian Li PC-A05NB Logitech G9 Xonar DX 
AudioAudio
SMSL SA-S3+Technics CB-250 Sennheiser HD555 
CPUMotherboardRAMHard Drive
AMD A10-5700 Gigabyte F2A75M-HD2 G.SKILL Ares 2133 CL9 Hitachi 5K750 
Hard DriveCoolingOSMonitor
Momentus .7 200GB Noctua NH-L9a Server 2012 R2 Standard AUO B156HW01 
PowerCaseOther
PicoPSU-80-WI-25V AIO Aluminium Handmade TP-Link Archer Something Something Wi-Fi AC 
CPUCPUCPUMotherboard
Core2Duo E6400 Core2Quad Q6600 Pentium Dual Core E5200 AsRock 4COREDUAL-SATA2 R2.0 
GraphicsRAMHard DriveOptical Drive
A dumpload of ancient AGP cards Kingston Value DDR2-667 CL4 2T @CL3 1T Seagate 160GB 7200.10 LG IDE DVD-ROM 
CoolingCoolingOSMonitor
Ghettomade CPU waterblock 49cc 2stroke engine copper radiator WinXP SP2 32bit ProView 17" 
PowerCase
Tacens Radix V 550W Ghetto aluminium bench 
  hide details  
Reply
post #222 of 233
Quote:
Originally Posted by KyadCK View Post

What's funny about that is Minecraft hates core count. Unless you're going Spigot and Bungee, you'll almost never see more than one or two cores actually be used.

Then again, Minecraft runs at 20tick, so it can be smooth even with 30fps. *shrug*

independent forge SERVER, not the client... completely different beast I commonly had 10 cores active out of the 16 actual.

It turns out that the i7-3960x backup gaming computer we have does better at dealing with lag spikes but the twin-G34 does better at just dealing with the game and produces like 1/10th the total heat.
post #223 of 233
Quote:
Originally Posted by Blameless View Post

If I can get an 8c/16 thread Zen, that can do at least 90% of the clock speed of an average 6900K and roughly 90% of the work per clock of a 6900K, amounting to more than 80% of the total performance, for ~75% of the price of a 6900K, then that I what will be in my next primary system...unless I catch a really good deal on a 6900K or 6950X in the meantime and can't afford a Zen in addition to it, when it actually shows up..

Isn't that impossible, even on paper?

Broadwell-E 8C @ 3.0GHz = 384GFLOP/s
Zeppelin 8C @ 3.0GHz = 192GFLOP/s

512-bit combined FPU width vs. 256-bit combined FPU width per core.
Edited by The Stilt - 9/26/16 at 3:22am
post #224 of 233
Quote:
Originally Posted by KyadCK View Post

Power litterally doesn't matter.

2x 140w vs 2x 90w = 100w difference. Power in the midwest is 11c/kw, so the amd setup would cost 1.1 cents more per hour. If performance is equal, and you render 24/7/365, the amd rig cost $96.36 more per year. Depends how much you're paying for each side, but in the end the power cost is insignificant compared to the price of components unless you live in a place that charges 30c/kw or something equally insane.

Say that to datacenters
Power Bac G5
(6 items)
 
  
CPUGraphicsRAMHard Drive
Intel i7 6950X EVGA GTX 1080 Ti GSkill 32GB 4133Mhz Samsung 960 Pro 1TB 
CoolingMonitor
Corsair H110i ASUS PG278Q 
  hide details  
Reply
Power Bac G5
(6 items)
 
  
CPUGraphicsRAMHard Drive
Intel i7 6950X EVGA GTX 1080 Ti GSkill 32GB 4133Mhz Samsung 960 Pro 1TB 
CoolingMonitor
Corsair H110i ASUS PG278Q 
  hide details  
Reply
post #225 of 233
Quote:
Originally Posted by TheLAWNOOB View Post

Quote:
Originally Posted by KyadCK View Post

Power litterally doesn't matter.

2x 140w vs 2x 90w = 100w difference. Power in the midwest is 11c/kw, so the amd setup would cost 1.1 cents more per hour. If performance is equal, and you render 24/7/365, the amd rig cost $96.36 more per year. Depends how much you're paying for each side, but in the end the power cost is insignificant compared to the price of components unless you live in a place that charges 30c/kw or something equally insane.

Say that to datacenters

Datacenters do not pay 11c/KW, but they do similar calculations, or they would buy nothing but T-series chips or similar, which they do not.

Nice mathematical rebuttal though.
Quote:
Originally Posted by prjindigo View Post

Quote:
Originally Posted by KyadCK View Post

What's funny about that is Minecraft hates core count. Unless you're going Spigot and Bungee, you'll almost never see more than one or two cores actually be used.

Then again, Minecraft runs at 20tick, so it can be smooth even with 30fps. *shrug*

independent forge SERVER, not the client... completely different beast I commonly had 10 cores active out of the 16 actual.

It turns out that the i7-3960x backup gaming computer we have does better at dealing with lag spikes but the twin-G34 does better at just dealing with the game and produces like 1/10th the total heat.

Not different at all. Internally they're exactly the same, the server just doesn't run the renderer. That's why OpenToLAN even works. When running single player you are running the Forge server, complete with perms, and connecting to it, inside one Java instance.

10 cores are not truly active for the main thread (TPS), and certainly not for forge. What you see in Task Manager or HTOP is not actual core load, as the Schd spread it around the cores during wait states. Some things can be multi-threaded like saving, loading new chunks form memory, etc, but the actual processing is all done on the main thread for our enjoyable pain.

The lovely thing about MC is that it actually is 20TPS, so you can do 40ms of work and still not feel it. If you aren't too picky you can even deal with 15TPS (67ms) Unfortunately, when doing, say, 210 active mod forge like my current mess around server, it does not take a lot to bring that to it's knees with absurd entity and active chunk count. The reason why I specified Spigot and Bungee is using the Cluster design, you can give every single Dim it's own java instance at the cost of a little overhead. Forge doesn't do that, and nor does stock MC, they love to load all their chunks in one instance for reasons unknown.

Which is exactly why the 3960X does better under actual load. And why my 5960X at 4.5 in my desktop does better than my 8350 stock in my quick-and-dirty server. The 8350 is still "enough" (20TPS) for medium load, but it'll be going to the E5-1650v3 server when I get the time.
Forge
(17 items)
 
Forge-LT
(7 items)
 
 
CPUMotherboardGraphicsGraphics
Intel i7-5960X (4.625Ghz) ASUS X99-DELUXE/U3.1 EVGA 1080ti SC2 Hybrid EVGA 1080ti SC2 Hybrid 
RAMHard DriveCoolingOS
64GB Corsair Dominator Platinum (3000Mhz 8x8GB) Samsung 950 Pro NVMe 512GB EK Predator 240 Windows 10 Enterprise x64 
MonitorKeyboardPowerCase
2x Acer XR341CK Corsair Vengeance K70 RGB Corsair AX1200 Corsair Graphite 780T 
MouseAudioAudioAudio
Corsair Vengeance M65 RGB Sennheiser HD700 Sound Blaster AE-5 Audio Technica AT4040 
Audio
30ART Mic Tube Amp 
CPUMotherboardGraphicsRAM
i7-4720HQ UX501JW-UB71T GTX 960m 16GB 1600 9-9-9-27 
Hard DriveOSMonitor
512GB PCI-e SSD Windows 10 Pro 4k IPS 
  hide details  
Reply
Forge
(17 items)
 
Forge-LT
(7 items)
 
 
CPUMotherboardGraphicsGraphics
Intel i7-5960X (4.625Ghz) ASUS X99-DELUXE/U3.1 EVGA 1080ti SC2 Hybrid EVGA 1080ti SC2 Hybrid 
RAMHard DriveCoolingOS
64GB Corsair Dominator Platinum (3000Mhz 8x8GB) Samsung 950 Pro NVMe 512GB EK Predator 240 Windows 10 Enterprise x64 
MonitorKeyboardPowerCase
2x Acer XR341CK Corsair Vengeance K70 RGB Corsair AX1200 Corsair Graphite 780T 
MouseAudioAudioAudio
Corsair Vengeance M65 RGB Sennheiser HD700 Sound Blaster AE-5 Audio Technica AT4040 
Audio
30ART Mic Tube Amp 
CPUMotherboardGraphicsRAM
i7-4720HQ UX501JW-UB71T GTX 960m 16GB 1600 9-9-9-27 
Hard DriveOSMonitor
512GB PCI-e SSD Windows 10 Pro 4k IPS 
  hide details  
Reply
post #226 of 233
Quote:
Originally Posted by KyadCK View Post

Quote:
Originally Posted by TheLAWNOOB View Post

Quote:
Originally Posted by KyadCK View Post

Power litterally doesn't matter.

2x 140w vs 2x 90w = 100w difference. Power in the midwest is 11c/kw, so the amd setup would cost 1.1 cents more per hour. If performance is equal, and you render 24/7/365, the amd rig cost $96.36 more per year. Depends how much you're paying for each side, but in the end the power cost is insignificant compared to the price of components unless you live in a place that charges 30c/kw or something equally insane.

Say that to datacenters

Datacenters do not pay 11c/KW, but they do similar calculations, or they would buy nothing but T-series chips or similar, which they do not.

Nice mathematical rebuttal though.
Quote:
Originally Posted by prjindigo View Post

Quote:
Originally Posted by KyadCK View Post

What's funny about that is Minecraft hates core count. Unless you're going Spigot and Bungee, you'll almost never see more than one or two cores actually be used.

Then again, Minecraft runs at 20tick, so it can be smooth even with 30fps. *shrug*

independent forge SERVER, not the client... completely different beast I commonly had 10 cores active out of the 16 actual.

It turns out that the i7-3960x backup gaming computer we have does better at dealing with lag spikes but the twin-G34 does better at just dealing with the game and produces like 1/10th the total heat.

Not different at all. Internally they're exactly the same, the server just doesn't run the renderer. That's why OpenToLAN even works. When running single player you are running the Forge server, complete with perms, and connecting to it, inside one Java instance.

10 cores are not truly active for the main thread (TPS), and certainly not for forge. What you see in Task Manager or HTOP is not actual core load, as the Schd spread it around the cores during wait states. Some things can be multi-threaded like saving, loading new chunks form memory, etc, but the actual processing is all done on the main thread for our enjoyable pain.

The lovely thing about MC is that it actually is 20TPS, so you can do 40ms of work and still not feel it. If you aren't too picky you can even deal with 15TPS (67ms) Unfortunately, when doing, say, 210 active mod forge like my current mess around server, it does not take a lot to bring that to it's knees with absurd entity and active chunk count. The reason why I specified Spigot and Bungee is using the Cluster design, you can give every single Dim it's own java instance at the cost of a little overhead. Forge doesn't do that, and nor does stock MC, they love to load all their chunks in one instance for reasons unknown.

Which is exactly why the 3960X does better under actual load. And why my 5960X at 4.5 in my desktop does better than my 8350 stock in my quick-and-dirty server. The 8350 is still "enough" (20TPS) for medium load, but it'll be going to the E5-1650v3 server when I get the time.

Power consumption does not matter at all. Ok good argument.
Power Bac G5
(6 items)
 
  
CPUGraphicsRAMHard Drive
Intel i7 6950X EVGA GTX 1080 Ti GSkill 32GB 4133Mhz Samsung 960 Pro 1TB 
CoolingMonitor
Corsair H110i ASUS PG278Q 
  hide details  
Reply
Power Bac G5
(6 items)
 
  
CPUGraphicsRAMHard Drive
Intel i7 6950X EVGA GTX 1080 Ti GSkill 32GB 4133Mhz Samsung 960 Pro 1TB 
CoolingMonitor
Corsair H110i ASUS PG278Q 
  hide details  
Reply
post #227 of 233
Quote:
Originally Posted by KyadCK View Post

Datacenters do not pay 11c/KW, but they do similar calculations, or they would buy nothing but T-series chips or similar, which they do not.

Nice mathematical rebuttal though.
Not different at all. Internally they're exactly the same, the server just doesn't run the renderer. That's why OpenToLAN even works. When running single player you are running the Forge server, complete with perms, and connecting to it, inside one Java instance.

10 cores are not truly active for the main thread (TPS), and certainly not for forge. What you see in Task Manager or HTOP is not actual core load, as the Schd spread it around the cores during wait states. Some things can be multi-threaded like saving, loading new chunks form memory, etc, but the actual processing is all done on the main thread for our enjoyable pain.

The lovely thing about MC is that it actually is 20TPS, so you can do 40ms of work and still not feel it. If you aren't too picky you can even deal with 15TPS (67ms) Unfortunately, when doing, say, 210 active mod forge like my current mess around server, it does not take a lot to bring that to it's knees with absurd entity and active chunk count. The reason why I specified Spigot and Bungee is using the Cluster design, you can give every single Dim it's own java instance at the cost of a little overhead. Forge doesn't do that, and nor does stock MC, they love to load all their chunks in one instance for reasons unknown.

Which is exactly why the 3960X does better under actual load. And why my 5960X at 4.5 in my desktop does better than my 8350 stock in my quick-and-dirty server. The 8350 is still "enough" (20TPS) for medium load, but it'll be going to the E5-1650v3 server when I get the time.

Oh well, 8350 is quite old, so it is expected to be outperformed by the 5960x, which has 8c/16t.
Desktop PC
(22 items)
 
ASUS ROG G72GX
(6 items)
 
 
CPUMotherboardGraphicsGraphics
AMD FX-8320 ASRock Fatal1ty 990FX Killer Sapphire Nitro+ RX480 Sapphire R9 290 Tri-X 
GraphicsRAMHard DriveHard Drive
XFX RX470 Singlefan Mushkin Redline 996996 2x4GB 2133Mhz Maxtor 6Y080L0 80GB 7200 RPM 8MB Western Digital 160GB 7200RPM 8MB 
Hard DriveHard DriveHard DriveOptical Drive
Maxtor 250GB 7200RPM 8MB Corsair Force LS WesternDigital Blue 500GB 7200RPM 16MB ASUS DVD-RW 
CoolingOSOSOS
ThermalTake Frio Silent 14 Windows 10 Enterprise Linux Mint 17.3 Rosa OphCrack 
MonitorKeyboardPowerCase
ASUS VS228HR Logitech K120 Corsair VS650 ThermalTake View 27 
MouseAudio
Bloody V5 Corsair HS30 Raptor 
CPUMotherboardGraphicsRAM
Intel Mobile Core 2 Duo P8700 G72GX NVIDIA GeForce GTX 260M  Hyundai Electronics  
RAMRAM
Hyundai Electronics  Hyundai Electronics  
  hide details  
Reply
Desktop PC
(22 items)
 
ASUS ROG G72GX
(6 items)
 
 
CPUMotherboardGraphicsGraphics
AMD FX-8320 ASRock Fatal1ty 990FX Killer Sapphire Nitro+ RX480 Sapphire R9 290 Tri-X 
GraphicsRAMHard DriveHard Drive
XFX RX470 Singlefan Mushkin Redline 996996 2x4GB 2133Mhz Maxtor 6Y080L0 80GB 7200 RPM 8MB Western Digital 160GB 7200RPM 8MB 
Hard DriveHard DriveHard DriveOptical Drive
Maxtor 250GB 7200RPM 8MB Corsair Force LS WesternDigital Blue 500GB 7200RPM 16MB ASUS DVD-RW 
CoolingOSOSOS
ThermalTake Frio Silent 14 Windows 10 Enterprise Linux Mint 17.3 Rosa OphCrack 
MonitorKeyboardPowerCase
ASUS VS228HR Logitech K120 Corsair VS650 ThermalTake View 27 
MouseAudio
Bloody V5 Corsair HS30 Raptor 
CPUMotherboardGraphicsRAM
Intel Mobile Core 2 Duo P8700 G72GX NVIDIA GeForce GTX 260M  Hyundai Electronics  
RAMRAM
Hyundai Electronics  Hyundai Electronics  
  hide details  
Reply
post #228 of 233
Quote:
Originally Posted by KyadCK View Post

2x 140w vs 2x 90w = 100w difference.

Power consumption matters quite a bit. However, TDP is only loosely related to overall power consumption. Most parts aren't run at maximum load 24/7 and and faster parts complete peak load tasks in less time, returning to lower loads sooner.

TDP figures are most useful for estimating peak power/cooling requirements, not operational costs.
Quote:
Originally Posted by The Stilt View Post

Isn't that impossible, even on paper?

Only if we are talking about peak AVX throughput.

Actual average application performance, even for those that make fairly heavy use of AVX, isn't going to have anywhere near a 100% IPC advantage on Broadwell-E...not unless Zen is worse than even the most pessimistic estimates.
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 #229 of 233
Quote:
Originally Posted by ku4eto View Post

Quote:
Originally Posted by KyadCK View Post

Datacenters do not pay 11c/KW, but they do similar calculations, or they would buy nothing but T-series chips or similar, which they do not.

Nice mathematical rebuttal though.
Not different at all. Internally they're exactly the same, the server just doesn't run the renderer. That's why OpenToLAN even works. When running single player you are running the Forge server, complete with perms, and connecting to it, inside one Java instance.

10 cores are not truly active for the main thread (TPS), and certainly not for forge. What you see in Task Manager or HTOP is not actual core load, as the Schd spread it around the cores during wait states. Some things can be multi-threaded like saving, loading new chunks form memory, etc, but the actual processing is all done on the main thread for our enjoyable pain.

The lovely thing about MC is that it actually is 20TPS, so you can do 40ms of work and still not feel it. If you aren't too picky you can even deal with 15TPS (67ms) Unfortunately, when doing, say, 210 active mod forge like my current mess around server, it does not take a lot to bring that to it's knees with absurd entity and active chunk count. The reason why I specified Spigot and Bungee is using the Cluster design, you can give every single Dim it's own java instance at the cost of a little overhead. Forge doesn't do that, and nor does stock MC, they love to load all their chunks in one instance for reasons unknown.

Which is exactly why the 3960X does better under actual load. And why my 5960X at 4.5 in my desktop does better than my 8350 stock in my quick-and-dirty server. The 8350 is still "enough" (20TPS) for medium load, but it'll be going to the E5-1650v3 server when I get the time.

Oh well, 8350 is quite old, so it is expected to be outperformed by the 5960x, which has 8c/16t.

8c/16t doesn't matter, if you actually read what I typed. 4.5Ghz Haswell vs 4Ghz Piledriver single core is the only comparison you need to make.

But yes, he's putting low-clocked PD Opterons against an Ivy-E, his situation is even worse. The difference is if load is light enough it doesn't actually matter... Until you get those lag spikes he was talking about.
Quote:
Originally Posted by Blameless View Post

Quote:
Originally Posted by KyadCK View Post

2x 140w vs 2x 90w = 100w difference.

Power consumption matters quite a bit. However, TDP is only loosely related to overall power consumption. Most parts aren't run at maximum load 24/7 and and faster parts complete peak load tasks in less time, returning to lower loads sooner.

TDP figures are most useful for estimating peak power/cooling requirements, not operational costs.
Quote:
Originally Posted by The Stilt View Post

Isn't that impossible, even on paper?

Only if we are talking about peak AVX throughput.

Actual average application performance, even for those that make fairly heavy use of AVX, isn't going to have anywhere near a 100% IPC advantage on Broadwell-E...not unless Zen is worse than even the most pessimistic estimates.

True, though I did specify "if equal performance" and "peak usage 24/7/365", which meets your quota. It is absolute worst case scenario for power usage.

Any other load (while remaining equal perf) and the difference will be smaller, hence being insignificant with the cost of power in the US. It's a percentage game, not an integer one.
Forge
(17 items)
 
Forge-LT
(7 items)
 
 
CPUMotherboardGraphicsGraphics
Intel i7-5960X (4.625Ghz) ASUS X99-DELUXE/U3.1 EVGA 1080ti SC2 Hybrid EVGA 1080ti SC2 Hybrid 
RAMHard DriveCoolingOS
64GB Corsair Dominator Platinum (3000Mhz 8x8GB) Samsung 950 Pro NVMe 512GB EK Predator 240 Windows 10 Enterprise x64 
MonitorKeyboardPowerCase
2x Acer XR341CK Corsair Vengeance K70 RGB Corsair AX1200 Corsair Graphite 780T 
MouseAudioAudioAudio
Corsair Vengeance M65 RGB Sennheiser HD700 Sound Blaster AE-5 Audio Technica AT4040 
Audio
30ART Mic Tube Amp 
CPUMotherboardGraphicsRAM
i7-4720HQ UX501JW-UB71T GTX 960m 16GB 1600 9-9-9-27 
Hard DriveOSMonitor
512GB PCI-e SSD Windows 10 Pro 4k IPS 
  hide details  
Reply
Forge
(17 items)
 
Forge-LT
(7 items)
 
 
CPUMotherboardGraphicsGraphics
Intel i7-5960X (4.625Ghz) ASUS X99-DELUXE/U3.1 EVGA 1080ti SC2 Hybrid EVGA 1080ti SC2 Hybrid 
RAMHard DriveCoolingOS
64GB Corsair Dominator Platinum (3000Mhz 8x8GB) Samsung 950 Pro NVMe 512GB EK Predator 240 Windows 10 Enterprise x64 
MonitorKeyboardPowerCase
2x Acer XR341CK Corsair Vengeance K70 RGB Corsair AX1200 Corsair Graphite 780T 
MouseAudioAudioAudio
Corsair Vengeance M65 RGB Sennheiser HD700 Sound Blaster AE-5 Audio Technica AT4040 
Audio
30ART Mic Tube Amp 
CPUMotherboardGraphicsRAM
i7-4720HQ UX501JW-UB71T GTX 960m 16GB 1600 9-9-9-27 
Hard DriveOSMonitor
512GB PCI-e SSD Windows 10 Pro 4k IPS 
  hide details  
Reply
post #230 of 233
Quote:
Originally Posted by TheLAWNOOB View Post

Say that to datacenters


"That to datacenters."

Datacenters use all kinds of chips if they're run by intelligent people who aren't intel's for a kickback. High power instant capacity processors do have a place.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [AT] AMD Zen Microarchitecture: Dual Schedulers, Micro-Op Cache and Memory Hierarchy Revealed