Graphics card power usage anomaly

What the f*ck are people doing using that web based shit? I suppose you read your emails on a webpage too? Get a newsreader, they're free!

Well, the RAM chips on the Nano are very tiny.

R9 280X 60% 250W $299 4.3Bil 28nm 2048/128/32 GCN 1.0 GDDR5 288GB/s R9 Nano 100% 175W $649 8.9Bil 28nm 4096/256/64 GCN 3.0 HBM 512GB/s

The Nano uses HBM, which is mounted inside the GPU. There's no ring of chips around the outside of the GPU on the Nano. One of the tricks, is getting the GPU and HBM to the same height, so that the cooler meets them properly. This is easiest to do, if there is only one supplier of HBM. There might have been a problem in a later HBM product, with two suppliers of HBM die, who were not making the chips the same height. There was a fair bit of whinging about how the cooler was supposed to fit on the later stuff. If you're taking an HBM product apart, you'd need to "read the ancient texts" to see if there are any rules to follow during disassembly.

I'm not convinced all the Wiki info I'm seeing is correct. It's possible the process the 28nm is done in, is actually different, and there could be a leakage mitigation method in place on the newer one.

There is too much of a difference between them, to just hand-wave it away.

If you have a newer card, the chip power management is dynamic, and a function of what the card is doing. The cards can go into clock-limited mode (Vcore to max to allow the clocking, power is much less than TDP), or TDP-limited mode (TDP power limiter trips, clock rate is reduced to stay within TDP).

Perfcap: "VREF" = Vcore_Max, power is 1/3rd of allowed TDP

Perfcap: "PWR" = TDP limit at full power, Vcore is reduced to meet limit, Clock drops because of VCore

And stuff like that makes hand-waving more difficult.

Paul

If it's designed more cleverly is could be that on average the transistors spend less time doing work (and more time in a static state).

Why would anyone take the advice of someone who doesn't even know about how digital transistor circuits use power?

You can't tell somebody off because the news server they connect to doesn't happen to carry all the newsgroups you posted to. Idiot.

Many Usenet newsservers let you post to groups that they don't carry.

Apologies if that's not what you meant.

That was what I meant. I have definitely had groups rejected when trying to reply in the past but maybe they were unusual or non standard groups.

Not everybody knows everything.

You would think AMD would have mastered something like that ages ago.

Of course I can, my newsserver carries every single group. Get a decent one.

Or at least one that doesn't cry like a baby when you dare to use a group it doesn't approve of.