AMD Announces Ryzen 4000 Mobile Processors: 4800U and 4800H

[medi01]

This doesn't make sense. Why do they have different power draws, according to you, if performance between the 4800U and 4800H is equal? Why design two separate tiers at all

I don't get the confusion, frankly.
Bar the "performance is the same" (it doesn't have to be), is it really surprising that chips from the same process could be split into better/worse power consumption groups? I.e. "harvested"?

Take better ones and sell them at a premium, not so efficient ones, for smaller sum, instead of having people win/lose lotteries.

 

[Valantar]

I don't get the confusion, frankly.
Bar the "performance is the same" (it doesn't have to be), is it really surprising that chips from the same process could be split into better/worse power consumption groups? I.e. "harvested"?

Take better ones and sell them at a premium, not so efficient ones, for smaller sum, instead of having people win/lose lotteries.

But "performance is the same", as is being claimed by @ratirt here, is exactly the problem - because it isn't the same. Period.

Binning for efficiency is most definitely a reality, but it can't work miracles such as lowering power draw by 66% for the same performance. Which is why low-power chips are good for bursty workloads (as they are allowed to (often dramatically) exceed their power budgets for limited time spans), but slower than their higher power counterparts in anything continuous.

As I mentioned before, there's plenty of documentation that Intel's 15W low-power chips consume 45-55W (or even more!) in their short turbo windows before scaling down to much lower clock speeds for a sustained 15W power consumption. That's how modern CPUs work, how they have such high boost specifications, and why understanding the difference between boost clock and base clock is crucial to understanding the differences between these products.

The binning at price points you are describing also happens, but mostly within a performance category (i.e. a 15W i7 generally costs the OEM more than a 15W i5 or i3), but chip prices across categories aren't easily comparable. Why? Because you can't outright say that one is "better" than the other. A 45W Ryzen 7 and a 15W Ryzen 7 are both high-bin chips, just different high bins (the 15W is likely binned for overall low voltage, low leakage current and hitting a good point on its DVFS (Dynamic Voltage Frequency Scaling) curve around 15W of package power, while the 45W is likely binned for hitting a good point on its DVFS curve at 45W, better efficiency at high clocks, and better clock scaling overall. Both are high bins, but different high bins, and as such both can realistically command a premium price.

 

[medi01]

But "performance is the same", as is being claimed by @ratirt here, is exactly the problem - because it isn't the same. Period.

It likely isn't the same (because power consumption gap is huge), but your point doesn't stand.
You CAN have different TDP chips from the same manufacturer, built using the same architecture and process with the same perf, but different power consumption.

 

[Valantar]

It likely isn't the same (because power consumption gap is huge), but your point doesn't stand.
You CAN have different TDP chips from the same manufacturer, built using the same architecture and process with the same perf, but different power consumption.

Of course you can. You just can't have that at a respective 15W and 45W outside of extreme and unrealistic binning with such small yields that they have no relevance to real products. 15W is close enough to the realistic power floor of a modern multi-core x86 architecture that pretty much any chip at that wattage will be power limited, and every chip would be able to sustain higher performance if given more power. So while an extreme top-bin 15W chip in theory indeed can be faster in sustained performance than a bottom-bin, near-discarded 45W chip, neither of those bins will actually exist or be made into mass-market products. Which is what we are discussing here, after all. Or did this at some point turn from a discussion of why actual products - the 4800U and 4800H - have the characteristics they have into an abstract theoretical discussion about the absolute possibilities of binning? Because as far as I know, that has never been what this discussion had been about.

We got into this part of the discussion because @ratirt started insisting that the 4800H wouldn't outperform the 4800U because they have the same max boost spec, which again sprang from the failure to accept that saving on iGPU power in a 45W chip would allow for higher CPU power and thus performance. This has been a long and winding road from the seemingly simple task of trying to explain why there isn't a fully enabled iGPU on a 45W APU.


Edit: damn autocorrect.