Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

diamond.g

macrumors G4
Mar 20, 2007
11,158
2,467
OBX
what homes? average homes, support up to 6000W, even more, depending on the country/city/house. Maybe i miss understood you
If only 1500W is supported...you turn 1 high end PC +1 1200BTU AC and you are done
But still a PC that draw almost over 1200W is insane for the times that we are living
I think they are talking from an American perspective. The average American single phase circuit is 15 or 20 amps with 120V which would be either 1800W or 2400W depending on what the breaker says for the circuit. We have 2 phase power, which would double the wattage at the same amperage, but in the US we typically use that for higher load devices (dryer, A/C, Oven, etc).
 
  • Like
Reactions: venom600

deconstruct60

macrumors G5
Mar 10, 2009
12,318
3,907
There are AMX units in both P-core and E-core complexes. The units in E-cores are smaller and have lower throughput.


OK so it is smoother. But the E cores are 4x as slow. If lean on the E core version for an extended period of time that should get most stuff eventually escalated to a P core. (at least if not an Apple watch SoC). [ Is it really even using any 'vectors' or just load two numbers at a time? Saving instruction loading but the data seems to be moving about a slow as just using regular instructions. ] But there is no high overhead 'trap and move' . Definitely doesn't waste as much die space as Intel is doing though.


And I'm not sure there is a decent way to implement a really slow, but far more compact AVX-512 for Intel's 'problem'. It is slightly different work space (it isn't technically restricted to just matrix.)
 

deconstruct60

macrumors G5
Mar 10, 2009
12,318
3,907
what homes? average homes, support up to 6000W, even more, depending on the country/city/house. Maybe i miss understood you

The power to a home going into a fuse box and split out into different circuits with individual fuse control on each circuit. In USA, normal, mundane plugs are on a 2400 W ( 120V 20 amp) circuit with other stuff. Limiting a single plug usage to less than 63% of that is a reasonable safety rule of thumb. So if are in a built in 1950-1970's house the multiple sockets in a room (and overhead lights) may all be on the same single circuit. A circuit may span multiple rooms where sizable electrical appliances are not the norm. (e.g. a repurposed bedroom into a 'working' den. ). Depending upon routing needed to other rooms one room may have one set of sockets on one circuit and a farther wall on a circuit shared with another room. Depends upon the layout. An electric stove ( a 220V feed) , electric drier , furnace , etc. would be on their own individual circuits with little to no circuit sharing.. All of that aggregates up to 6000W , but that doesn't mean you can point all of the house feed into some random socket in a random room in a dwelling.

If know want to plug a single highly abnormal power draw to a single socket in a specific room and can modify the fuse box then can run a single line out to a single socket.



If only 1500W is supported...you turn 1 high end PC +1 1200BTU AC and you are done

15 100W light bulbs would not be needed to light up a classic norm sized room (or two). (pre-era of the open floor plan mania and mega size houses. ) . It is a highly abnormal amount of power for a normal room (or set of smallish rooms) .
 

venom600

macrumors 65816
Mar 23, 2003
1,301
1,101
Los Angeles, CA
If only 1500W is supported...you turn 1 high end PC +1 1200BTU AC and you are done
But still a PC that draw almost over 1200W is insane for the times that we are living

I just moved into a new apartment in Seattle this month, a big city. I plugged in my 12000BTU portable AC (since they still don't build apartments with central AC here) and as soon as I turned it on, it overloaded the circuit for the living room and tripped the breaker because the overhead lights (efficient LEDs) and the AC were pulling too much power. The scenario you present is already reality.
 

leman

macrumors Core
Oct 14, 2008
19,308
19,298
OK so it is smoother. But the E cores are 4x as slow. If lean on the E core version for an extended period of time that should get most stuff eventually escalated to a P core. (at least if not an Apple watch SoC). [ Is it really even using any 'vectors' or just load two numbers at a time? Saving instruction loading but the data seems to be moving about a slow as just using regular instructions. ] But there is no high overhead 'trap and move' . Definitely doesn't waste as much die space as Intel is doing though.

And I'm not sure there is a decent way to implement a really slow, but far more compact AVX-512 for Intel's 'problem'. It is slightly different work space (it isn't technically restricted to just matrix.)

I would also think that AMX is much simpler to implement — it's a very straightforward device with a simple ISA. I'm wondering how it will be with SVE/SVE2 (but then again ARM has it in A510...)

But the E cores are 4x as slow. If lean on the E core version for an extended period of time that should get most stuff eventually escalated to a P core.

Oh, of course. It's just that E-cores offer some additional throughput — 20-25% is nothing to sneeze about.

Seems silly to complain about a CPU using the equivalent of three incandescent light bulbs for actual computer use tasks, when people are burning a a million times that amount ‘mining’ for useless crypto like bitcoin.

One thing does not exclude the other. Just because crypto mining is a large environmental problem does not justify increasing energy cost of running consumer PC hardware. Personally, I don't much care about the enthusiast-class hardware, that's niche enough, but the increased power consumption also does affect middle-class and these things add up.
 

diamond.g

macrumors G4
Mar 20, 2007
11,158
2,467
OBX
I would also think that AMX is much simpler to implement — it's a very straightforward device with a simple ISA. I'm wondering how it will be with SVE/SVE2 (but then again ARM has it in A510...)



Oh, of course. It's just that E-cores offer some additional throughput — 20-25% is nothing to sneeze about.



One thing does not exclude the other. Just because crypto mining is a large environmental problem does not justify increasing energy cost of running consumer PC hardware. Personally, I don't much care about the enthusiast-class hardware, that's niche enough, but the increased power consumption also does affect middle-class and these things add up.
What would you consider middle class hardware?
 

johannnn

macrumors 68020
Nov 20, 2009
2,221
2,341
Sweden
Seems silly to complain about a CPU using the equivalent of three incandescent light bulbs for actual computer use tasks, when people are burning a a million times that amount ‘mining’ for useless crypto like bitcoin.
Trust me I resent everyone that mines useless crypto as well. I really do. We as humans can do better than that.
 
  • Like
Reactions: pdoherty

deconstruct60

macrumors G5
Mar 10, 2009
12,318
3,907
I would also think that AMX is much simpler to implement — it's a very straightforward device with a simple ISA. I'm wondering how it will be with SVE/SVE2 (but then again ARM has it in A510...)

I would expect there is bigger overlap between NEON and SVE2 than with AMX. SVE2 is a superset of NEON And similar 'tap dancing' required if trying to use NEON (or a dense cluster of multi/div ops ) and AMX very close together.

SVE2 should push AMX usage into a smaller, tighter niche. And SVE2 would get a bigger transistor budget. But it won't be surprising in the short term that AMX delays SVE2 from showing up on Apple's implementations. Mostly because it will be a budget battle than small fab process would give lots of relief to.


Oh, of course. It's just that E-cores offer some additional throughput — 20-25% is nothing to sneeze about.

In the watch, homepod, and phone SoC space, yes. In the desktop processor space, we'll see when Ryzen 7000 gets some optimizations rolled out for their new SIMD subsystem. For entry - midrange Intel Gen 13 competition it is probably helpful.
 

diamond.g

macrumors G4
Mar 20, 2007
11,158
2,467
OBX
The i3-i5 crop of CPUs, stuff you find in most cheap desktop PCs towers.
I see. If anything 12th Gen has made the power requirements make more sense (all core turbo power was always able to go over the base power requirements). The base needs haven't changed though, so yay?

GPU's have upped power requirements, not sure how they will be able to reign it in to be honest.
 

diamond.g

macrumors G4
Mar 20, 2007
11,158
2,467
OBX
I see. If anything 12th Gen has made the power requirements make more sense (all core turbo power was always able to go over the base power requirements). The base needs haven't changed though, so yay?

GPU's have upped power requirements, not sure how they will be able to reign it in to be honest.
Ehhh, looks like (Stock) power requirements for GPUs hasn't gone up as much as was initially reported.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.