Here in this thread I'm going to go into little more detail about some misconceptions and things that aren't understood clearly about Intel's 2012 CPU code-named Ivy Bridge
Namely, 1) Benefits of a new process technology, and 2) Configurable TDP.
1. 22nm Tri-Gate process technology:
Unfortunately, there's lot of misunderstandings about the benefits that process technology generation gives. It's important to understand that its something that gives CPU architects to play around. Whether the potential is realized is another story.
Take a look at this curve graph for 22nm:![]()
It's important to note the voltage points. Remember, there always exist trade-offs. Because the 17W Ivy Bridge chips going into Macbook Airs and Ultrabooks will be noticeably faster than the predecessor, they won't be able to scale down the voltage, and save power.
What about the thing that says "37% faster"? Well, that's at very low operating voltages. It seems its at about 0.7V. Sandy Bridge requires well over 1V to reach HFM(High Frequency Mode), which is the highest Base frequency, and the same is true with Turbo Mode. So the gain will be closer to 20%. The operating frequency that runs around 0.7V is LFM(Low Frequency Mode).
To wrap the two points up:
A. The gains will be nominal, nothing out of the ordinary
B. You won't see much lower TDP, if at all.
But it does not mean the "37% faster" won't be taken advantage of. That's where Configurable TDP comes in. Remember how I said process technologies enable CPU architects to do "play around"?
2. Configurable TDP
On certain Ivy Bridge mobile chips, there are 3 TDP points, named cTDP down, Nominal TDP, and cTDP up. With 17W SKUs the 3 TDP points are 14W/17W/25W. Each TDP points have their own base clocks.
What Configurable TDP allows is for system designers to "configure" the chip around the system they want to build rather than doing it the other way, which is building a chassis just for that SKU.
You may ask What's the difference from Throttling and Turbo?
Throttling: It's only when the chip reaches an unsafe temperature, say 90C
Turbo Mode: Preset timers, power consumption, and upon reaching temperature that's not unsafe for the CPU, say 60C
Something good to know about Turbo Mode: You probably know its not guaranteed frequency. It can easily ramp down. Base clocks are the guaranteed frequency. Both throttling and Turbo Mode plays around that.
Another important thing to know about Turbo Mode: It's really about response times. To make the system feel "snappier". Oh, and your system WILL be faster about 95% of the time since the duration you run your system at full versus the duration you don't run it at full is much less.
Here's a graph showing how cTDP works:![]()
You can see that the chip with cTDP down loses its Turbo(compared to Nominal) on workloads that allow the chip to reach max TDP, but retain similar performance on workloads that are light enough to be much below max TDP.
Relation between cTDP and LFM: The LFM frequency in Nominal TDP is the base frequency for cTDP down. The 22nm transistor benefits would mean much higher frequencies are possible at the cTDP down(compared to if it didn't have 22nm process).
Configurability of cTDP can be set by the manufacturer. The trigger can be:
-Environmental: Meaning if you switch from say a Tablet mode, to a Notebook mode. Notebook mode might have better cooling, so it can go from cTDP down to Nominal, or Nominal to cTDP up
-Manual: So it works like a Turbo button a user presses.
-Fixed: The manufacturer has set the CPU at certain TDP just for the convenience of designing at that set frequency
Configurable TDP summary: Systems are designed around worst case. By allowing lower TDP figures using cTDP down, the manufacturer can lower that worst case and make a thinner, sleeker system. cTDP up allows faster operation when more cooling is available, like when docked with better cooling. Turbo clock speeds are kept the same to benefit responsiveness, regardless of the form factor/TDP.
If a Chassis allowing 14W had a Nominal 17W TDP chip in it, the chip would throttle, which results in:
A: Lower performance
B: Higher chassis temperature
Throttling on desktop processors happen at around 80-85C. Because the CPU will constantly try to reach maximum frequency AGAINST the throttling mechanism, the temperature will stay at those temperatures until the load is reduced. Which will in turn raise the chassis temperatures beyond if it wasn't throttling. Lower performance is due to the loss of Turbo Mode.
(Of course, you can put the same chip in an overspecced system, so it doesn't need it, but for absolute thin and light systems, lower TDP is necessary. One can imagine a "smaller than 11" MBA" MBA in the future )
Namely, 1) Benefits of a new process technology, and 2) Configurable TDP.
1. 22nm Tri-Gate process technology:
Unfortunately, there's lot of misunderstandings about the benefits that process technology generation gives. It's important to understand that its something that gives CPU architects to play around. Whether the potential is realized is another story.
Take a look at this curve graph for 22nm:
It's important to note the voltage points. Remember, there always exist trade-offs. Because the 17W Ivy Bridge chips going into Macbook Airs and Ultrabooks will be noticeably faster than the predecessor, they won't be able to scale down the voltage, and save power.
What about the thing that says "37% faster"? Well, that's at very low operating voltages. It seems its at about 0.7V. Sandy Bridge requires well over 1V to reach HFM(High Frequency Mode), which is the highest Base frequency, and the same is true with Turbo Mode. So the gain will be closer to 20%. The operating frequency that runs around 0.7V is LFM(Low Frequency Mode).
To wrap the two points up:
A. The gains will be nominal, nothing out of the ordinary
B. You won't see much lower TDP, if at all.
But it does not mean the "37% faster" won't be taken advantage of. That's where Configurable TDP comes in. Remember how I said process technologies enable CPU architects to do "play around"?
2. Configurable TDP
On certain Ivy Bridge mobile chips, there are 3 TDP points, named cTDP down, Nominal TDP, and cTDP up. With 17W SKUs the 3 TDP points are 14W/17W/25W. Each TDP points have their own base clocks.
What Configurable TDP allows is for system designers to "configure" the chip around the system they want to build rather than doing it the other way, which is building a chassis just for that SKU.
You may ask What's the difference from Throttling and Turbo?
Throttling: It's only when the chip reaches an unsafe temperature, say 90C
Turbo Mode: Preset timers, power consumption, and upon reaching temperature that's not unsafe for the CPU, say 60C
Something good to know about Turbo Mode: You probably know its not guaranteed frequency. It can easily ramp down. Base clocks are the guaranteed frequency. Both throttling and Turbo Mode plays around that.
Another important thing to know about Turbo Mode: It's really about response times. To make the system feel "snappier". Oh, and your system WILL be faster about 95% of the time since the duration you run your system at full versus the duration you don't run it at full is much less.
Here's a graph showing how cTDP works:
You can see that the chip with cTDP down loses its Turbo(compared to Nominal) on workloads that allow the chip to reach max TDP, but retain similar performance on workloads that are light enough to be much below max TDP.
Relation between cTDP and LFM: The LFM frequency in Nominal TDP is the base frequency for cTDP down. The 22nm transistor benefits would mean much higher frequencies are possible at the cTDP down(compared to if it didn't have 22nm process).
Configurability of cTDP can be set by the manufacturer. The trigger can be:
-Environmental: Meaning if you switch from say a Tablet mode, to a Notebook mode. Notebook mode might have better cooling, so it can go from cTDP down to Nominal, or Nominal to cTDP up
-Manual: So it works like a Turbo button a user presses.
-Fixed: The manufacturer has set the CPU at certain TDP just for the convenience of designing at that set frequency
Configurable TDP summary: Systems are designed around worst case. By allowing lower TDP figures using cTDP down, the manufacturer can lower that worst case and make a thinner, sleeker system. cTDP up allows faster operation when more cooling is available, like when docked with better cooling. Turbo clock speeds are kept the same to benefit responsiveness, regardless of the form factor/TDP.
If a Chassis allowing 14W had a Nominal 17W TDP chip in it, the chip would throttle, which results in:
A: Lower performance
B: Higher chassis temperature
Throttling on desktop processors happen at around 80-85C. Because the CPU will constantly try to reach maximum frequency AGAINST the throttling mechanism, the temperature will stay at those temperatures until the load is reduced. Which will in turn raise the chassis temperatures beyond if it wasn't throttling. Lower performance is due to the loss of Turbo Mode.
(Of course, you can put the same chip in an overspecced system, so it doesn't need it, but for absolute thin and light systems, lower TDP is necessary. One can imagine a "smaller than 11" MBA" MBA in the future
)
Last edited: