News AMD Corrects Socket AM5 for Ryzen 7000 Power Specs: 230W Peak Power, 170W TDP

[InvalidError]

Under-volting can potentially increase the headroom but it's a huge hassle to find the correct spot that won't cause instability when running some rare case.

Linus did an undervolting experiment last week where he found out that the default voltage curve was 200+mV above what his test CPU needed. Seems like the default curve is grossly conservative/pessimistic. Having to hand-tune voltages/offsets may be a pain in the ass but it can save 30+% on the amount of power necessary for a given amount of performance.

 

[martinch]

Under-volting and under-clocking are always an option if you favor power-efficiency over maximum performance.

Yeah, that's true. I'm just lazy and don't want to have to go fiddling (obviously, ideally components would be both power-efficient and performant )

 

[g-unit1111]

The sad fact is that there is no rules/ regulations that limits how much power, or set a efficiency target, for PC hardware. As result, we see companies trying to outdo each other by gradually relaxing power limits in irresponsible ways.

I feel like there should be, and things should be getting more energy efficient, not less. Especially given current climate / energy restriction challenges. But that’s an entirely different topic of discussion altogether.

 

[InvalidError]

I feel like there should be, and things should be getting more energy efficient, not less. Especially given current climate / energy restriction challenges. But that’s an entirely different topic of discussion altogether.

That is kind of how the GHz-at-any-cost race ended the first time around. Power and cooling are major costs for datacenters, which put intense pressure on performance per watt. It isn't as much of an issue for home PCs where households have 0-2 high-end systems with low duty cycles.

 

[TerryLaze]

I feel like there should be, and things should be getting more energy efficient, not less. Especially given current climate / energy restriction challenges. But that’s an entirely different topic of discussion altogether.

Maximum power draw increasing doesn't tell you anything about efficiency, if the performance increase is higher than the power increase then it's still more efficient than a system drawing less power.

The major problem we are having is reviews not benching for efficiency, they show you power draw as max power draw with a power virus (which is useless) and they show you performance with a variety of programs without showing you the power draw.

 

[digitalgriffin]

Style wise the heat spreaders look good.

Like they are trying to restrain the beast beneath.

The exposed power caps are a bit risky. Not all paste will be compatible. For example artic silver 5 would be a NO GO because it's highly capacitive in nature.

I wonder if they will have to reformulate the plastic. If overclocks push beyond the standard TDP, it would hasten embrittlement and warping.

 

[g-unit1111]

Maximum power draw increasing doesn't tell you anything about efficiency, if the performance increase is higher than the power increase then it's still more efficient than a system drawing less power.

The major problem we are having is reviews not benching for efficiency, they show you power draw as max power draw with a power virus (which is useless) and they show you performance with a variety of programs without showing you the power draw.

Yeah it will be interesting to see what the benchmarks say when AM5 is finally released.

 

[alceryes]

Linus did an undervolting experiment last week where he found out that the default voltage curve was 200+mV above what his test CPU needed. Seems like the default curve is grossly conservative/pessimistic. Having to hand-tune voltages/offsets may be a pain in the ass but it can save 30+% on the amount of power necessary for a given amount of performance.

This is similar to what I have dialed in on my 9900k. I took the time to test, tweak, test, tweak, etc.
End result is an undervolted CPU with a locked 4.8/4.6 that regularly matches the performance numbers of ones allowed to boost to 5GHz. That's how I'm able to run this beast on a relatively small H80i v2 cooler. It's getting old though. I can't wait to build out my AM5 rig.

 

[InvalidError]

The exposed power caps are a bit risky. Not all paste will be compatible. For example artic silver 5 would be a NO GO because it's highly capacitive in nature.

The clumps of on-package MLCCs are almost certainly all power supply bypass caps, each clump likely belonging to the same power rail. Extra bypass capacitance between power and ground is practically never a bad thing unless you go grossly overboard with it. If AMD is worried about SMD contamination as I know I would be if the exposed under-IHS area is nearly impossible to clean open space , it can always encapsulate them in conformal coating or underfill.