Overclock.net › Forums › Industry News › Hardware News › [Anandtech] Apple's Cyclone Microarchitecture Detailed
New Posts  All Forums:Forum Nav:

[Anandtech] Apple's Cyclone Microarchitecture Detailed - Page 5

post #41 of 42
Quote:
Originally Posted by DuckieHo View Post

Again.... your statement is wrong.... A 1.3GHz quad-core can power-gate two cores and throttle exactly 300MHz. What KIND of load matters! You can't compare a dual-core to a quad-core without holding the workload constant.

It's just physics.... many tablet SoCs do have the thermal and power availability to allow these chips to run faster.


It's not that simple.... not all threads require the same resources and most threads in games still have to deal synchronization. The game example varies but a few faster cores will generally beat out more slower cores in gaming (have to worry about caches as well).

However, if you have an embarrassingly parallel workload.... a 4x500MHz cores should perform close to a 2x1GHz cores. It's very close but not quite the same generally due to some process overhead.

Your 4x500MHz is probably faster because of context switching. The OS is going to be constantly interrupting at least one core to perform system tasks. These interrupts also make performance less deterministic. This is why the current generation of consoles dedicate some cores for system use only and the idea behind Nvidia's 5th core model (where the majority of interrupts are handled by this other core).

Another issue with the power theoretics being discussed here is linear power scaling with frequency. This holds true in theory, but is far from true in practice due to real transistors being less than ideal (and also because extra cores share the same uncore features which cost the same either way). In this case, we're most likely talking about that 1.0GHz becoming 0.7-0.8GHz at the same power level.

A final consideration is processors like Qualcomms which have per-core power planes. In this case, you can dedicate one core to the OS and downclock it to a couple hundred MHz and run the other cores at almost full speed (or perhaps at full speed if there are things like IO blocks that allow intermittent downclocking -- something likely on IO starved mobile devices).
post #42 of 42
Quote:
Originally Posted by hajile View Post

Your 4x500MHz is probably faster because of context switching. The OS is going to be constantly interrupting at least one core to perform system tasks. These interrupts also make performance less deterministic. This is why the current generation of consoles dedicate some cores for system use only and the idea behind Nvidia's 5th core model (where the majority of interrupts are handled by this other core).
Correct, but I was more toward theoretical workloads disregarding other devices and OS. The point was that complexity goes up quite rapidly with multithreading and the answer is not always "just use more threads!" (as some people believe).

Quote:
Originally Posted by hajile View Post

A final consideration is processors like Qualcomms which have per-core power planes. In this case, you can dedicate one core to the OS and downclock it to a couple hundred MHz and run the other cores at almost full speed (or perhaps at full speed if there are things like IO blocks that allow intermittent downclocking -- something likely on IO starved mobile devices).

We had discussed this with power gating and clock gating. There is also dynamic voltage that is currently implemented as well.

If interested, there is also Near-Threshold Voltage as an even more advanced technique to reduce voltage and power consumption: http://www.realworldtech.com/near-threshold-voltage/
Edited by DuckieHo - 4/3/14 at 7:04am
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: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [Anandtech] Apple's Cyclone Microarchitecture Detailed