Haswell Could Have Compatibility Problems With Older PSUs

[Rune Olsen]

Hmm, that might be the cause. Thanks for the tip mate

 

[ojas]

Since i already see people jumping up and down about needing a new PSU for Haswell, and before people start claiming how AMD is better because it can use older power supplies: Relax.
What i've read is that there will be BIOS implemented switch that detects whether the 0.05mA requirement is met or not, if it isn't then C6, C7 will be disabled.
Obviously, this only really matters to mobile device users, who don't re-use their old laptop/tablet or phone's PSU anyway, so i really don't think it's an issue.

 

[CrArC]

Most cheap PSUs already have internal loads (resistors) to guarantee the minimum load they need to remain reasonably stable.
Where Haswell may be more of a challenge to cheap PSUs is response time to large load swings when the CPU keeps bouncing between C7 and full-power: too little feedback bandwidth means the voltage rails will sag when the CPU wakes up which may trip under-voltage lock-outs or surge when the CPU goes to sleep and trip over-voltage lock-outs. Depending on how fast Haswell can switch from full-power to sleep, the PSU's feedback delay could potentially encounter situations where the CPU is already going back to sleep while the PSU is still ramping up from the previous wake-up or vice-versa.

The bandwidth of the system between PSU and motherboard should be enough compared to the transient power draw of the processor, surely? Bear in mind that transitioning from low C-states is a LOT slower than transitioning through varying levels of CPU load.

Moreover, isn't this condition already handled at the motherboard's VRM stage? the vreg setpoints are controlled by the CPU based on its knowledge of core temps, freq, workload, anticipated workload, and calculated short-term limits on maximum power states (i.e. how fast the chip can transition between states and what states it can transition to). Besides, through normal VRM operation, with their local energy store in the caps on the board, the degree of transient load appears (somewhat) reduced to the PSU. If this wasn't the case already, wouldn't our systems have trouble moving between active and sleep modes?

 

[nhat11]

Yea glad intel is looking to powersavings now instead of power each year since the market is heading into mobile devices.

 

[beayn]

Why is there no quote button?
@maui67 They will still be able to make cheap PSUs that support the new processors. It will just take a couple extra capacitors and/or resistors to make the PSU supply stable voltage at low amps. The article says quality PSUs work because they take all that stuff into account when building them.

 

[InvalidError]

Bear in mind that transitioning from low C-states is a LOT slower than transitioning through varying levels of CPU load.

One of the biggest reasons for Haswell's power-efficiency gains is its ability to switch power states about 10X faster than Sandy and Ivy did so it can spend ~10X as much time in deep-sleep than SB/IB do when under light loads.

 

[markdj]

Stable voltage at 0 watts....ummm interesting science.

That's just normal science. You don't need current flow to have voltage. You need both voltage and current for for watts. Voltage is just a measure of potential. 5 volts, 0 amps will give 0 watts

 

[Pinhedd]

Stable voltage at 0 watts....ummm interesting science.

That's just normal science. You don't need current flow to have voltage. You need both voltage and current for for watts. Voltage is just a measure of potential. 5 volts, 0 amps will give 0 watts

switched mode power supplies have a hard time supplying stable voltage at low loads.

 

[danwat1234]

I doubt this will be an issue. The motherboards will probably have capacitors and filtering to cushion weird voltage swings. More than likely any old ATX power supply of sufficient load ratings can run it fine.

 

[InvalidError]

switched mode power supplies have a hard time supplying stable voltage at low loads.

Most cheap PSUs I have dismantled have bleeder resistors on 3.3/5/12V rails drawing 2-3W to guarantee minimum loading. Not pretty for power-efficiency but it gets the job done.

More efficient PSUs put more effort in minimizing leakage inductance and use shunt regulators instead of plain resistors.

 

[skux]

Stable voltage at 0 watts....ummm interesting science.

Nothing wrong with this. Do you not understand that a voltage source can have a well defined voltage regardless of current draw?
For example, a car lead-acid battery would have a pretty stable voltage when ramping the load down from 1W to 0W.

Why is it the top voted comment?

 

[Pinhedd]

Stable voltage at 0 watts....ummm interesting science.

Nothing wrong with this. Do you not understand that a voltage source can have a well defined voltage regardless of current draw?
For example, a car lead-acid battery would have a pretty stable voltage when ramping the load down from 1W to 0W.

Why is it the top voted comment?

Because it's correct. If PC power supplies were linearly regulated like they were in the 1980s a 0 watt load would not be an issue but they are not linearly regulated, they use a switching feedback loop for regulation.

 

[InvalidError]

Because it's correct. If PC power supplies were linearly regulated like they were in the 1980s a 0 watt load would not be an issue but they are not linearly regulated, they use a switching feedback loop for regulation.

It is perfectly possible to create a switching power supply that can provide reasonably stable voltage at ~0W output.

All you need is to design a switcher that can operate in discontinuous mode with low enough minimum energy transfer per cycle to avoid overshooting the output voltage target when output caps need to get topped off while operating at near-zero load.

A PSU (any type) never operates at truly zero load since output capacitors and power semiconductors have leakage currents.

Even a linear regulator can drift above its nominal output voltage if there is no internal shunt or some form of load on the output to prevent leakage current though the series power device(s) from floating the output above its nominal value. In most linear regulators, the output divider/reference network usually provides sufficient load to offset leakage from power devices so this rarely comes up as a design consideration.

Switching PSUs often do a very similar trick by feeding the PWM circuitry from a regulated output after initial bootstrapping so the PWM's operation contributes to guaranteeing baseline load on the main output during normal operation.