Apple M1 Ultra - 2nd highest multicore score, lost to 64-core AMD Threadripper.

[Just that Mario]

On 3/28/2022 at 4:11 AM, Alex Atkin UK said:

Another annoying thing is MacOS unmounts the drives whenever Samba is restarted on the server, so I regularly wake up the Mac to a "drives disconnected" message because my server updates at 3am automatically

Ye, MacOS disconnects after even going to sleep for few minutes. Sometimes reconnects itself, sometimes doesn't. Not sure if it completely turns network access off or what. I do know that VPN on Windows will happily keep running even when PC is in sleep.

[Dracarris]

On 4/9/2022 at 8:06 PM, Just that Mario said:

I do know that VPN on Windows will happily keep running even when PC is in sleep.

Not true at least for the Cisco VPNs. They require a password after wake from sleep and annoy you with soundy alerts. However samba shares will immediately work again once reconnected. As it should be.

[Obioban]

Adobe released M1 native After Effects today. They also released a performance gains graph vs whatever 10 core Xenon is in the iMac Pro.

 

[Paul Thexton]

Apple need to hire someone who actually knows how to do performance comparisons. On paper they look like good gains, but I'll wait for the tech YouTube community to add some meat (or not) to the bones of those claims.

[Obioban]

1 hour ago, Paul Thexton said:

Apple need to hire someone who actually knows how to do performance comparisons. On paper they look like good gains, but I'll wait for the tech YouTube community to add some meat (or not) to the bones of those claims.

That graph aren't isn't Apple-- it's from Adobe.

[descendency]

On 3/9/2022 at 12:44 PM, Spindel said:

At a power draw of 65-70 W on the CPU

This is the amazing/frustrating part. I really wish Apple would make cores that are allowed to clock higher on desktop (where they will have the cooling). I think this has to be coming in future iterations of Apple's silicon. 

[Paul Thexton]

38 minutes ago, Obioban said:

That graph aren't isn't Apple-- it's from Adobe.

Somehow that's even more disappointing

[Paul Thexton]

32 minutes ago, descendency said:

This is the amazing/frustrating part. I really wish Apple would make cores that are allowed to clock higher on desktop (where they will have the cooling). I think this has to be coming in future iterations of Apple's silicon. 

Yeah I get where you're coming from, but I'll hold my hands up and say I know utterly sod all about CPU core design, is it really as simple as allowing the clock speed to go up another Ghz, or does that cause other issues that you need to accommodate in the design of the core itself?

 

This is what the clock rates & power consumption looks like when putting my 14" M1Max through a big code compile job (stats tool is asitop). It seems to max out @ 3Ghz for the P-cores and 2Ghz for the E-cores.  Peak 36W power draw.

[leadeater]

3 hours ago, Obioban said:

Adobe released M1 native After Effects today

M1 Max seems to be the best value overall for After Effects based on the graph

 

1 hour ago, Obioban said:

That graph aren't isn't Apple-- it's from Adobe.

It's Adobe, that makes it less trustworthy than from Apple because of how unreliable Adobe is 

[leadeater]

1 hour ago, descendency said:

This is the amazing/frustrating part. I really wish Apple would make cores that are allowed to clock higher on desktop (where they will have the cooling). I think this has to be coming in future iterations of Apple's silicon. 

Adding boosting, or higher clocks, would just push the chip in to the inefficient area of the silicon process and architecture limitations (in terms of perf gains). Also these are really big chips, I doubt the clocks could be pushed anywhere near 5Ghz or even 4.5Ghz. I don't think there is as much performance left on the table as you might be thinking, even if 3.5Ghz-3.8Ghz were possible it may come at the cost of being 10W per core rather than 4W, wouldn't seem like a good trade off and for something like a M1 Ultra 200W sustained power just for the CPU cores alone doesn't sound feasible at all once the GPU kicks in.

[hishnash]

10 hours ago, Paul Thexton said:

Yeah I get where you're coming from, but I'll hold my hands up and say I know utterly sod all about CPU core design, is it really as simple as allowing the clock speed to go up another Ghz, or does that cause other issues that you need to accommodate in the design of the core itself?

 

You correct you cant just turn up the frequency, the internal layout of transistors and the tolerances for noice will be tuned much more tightly to this given freqniey than the lines of chips from AMD, and Intel doing this saves on transistors and other structures and also saves power.  I would be very surprised if the chip would be stable even at 3.4GHz let alone 4GHz.  

Increasing speeds is not as simple as such increasing voltage and power, even if you can cool a chip you get to a point were the design of the chip just cant operate at that voltage (you might have a short within the chip, you might have electromagnetic interference issues, and the power delivery system within the chip will also have a limit). Apple will have designed all of these for a given spec and will not have wasted die area and fab time for running outside that spec.

Teams at intel and AMD do not have such a clear operating target when they design a internal structure of a chip, the team working on the ADDer transistor layout for intel will likly expect that to be used in a chip from 1GHz all the way unto 6GHz or more so will need to design it accordingly, if a single part of the chip cant run at the design frequency/voltage your hoping that is a limit.  Apples teams will have been told at the get go 'your designing a new ADDer layout that will operate at 3.0 to 3.2GHz' this is a much simpler task! And the teams working on paren delivery and regulation within the chip will also be given exact power draw for each part of the chip they are not going to bother with using up extra die area just in case there are some golden binned samples were logical units can run at a higher frequency as they know that will never go in a product.


Fundamentally the difference between AMD/Intel vs Apple is apple is not designing a chip and then building a product to use that chip, they are designing a product and then requesting a given chip from the chip teams to fit into that product.  If apple want a cpu with high single core perfomance they will not do this by increasing clock speeds they will do it by building a wider higher IPC core as while this costs more in transistors this is much better in pref/W, if they want a cpu with higher multi core perf.. well they will use more cores. 
 

[leadeater]

2 hours ago, hishnash said:

Fundamentally the difference between AMD/Intel vs Apple is apple is not designing a chip and then building a product to use that chip, they are designing a product and then requesting a given chip from the chip teams to fit into that product.  If apple want a cpu with high single core perfomance they will not do this by increasing clock speeds they will do it by building a wider higher IPC core as while this costs more in transistors this is much better in pref/W, if they want a cpu with higher multi core perf.. well they will use more cores.

I also think standardization is something Apple also wants. Imagine the problems with product marketing and customer communication if Apple had M1 family CPUs at different operating clocks based on platform installed in or based on total number of cores. We already get these odd ball situations on the PC side where a lower core count CPU performs better in some applications than higher core count variants due to being able to sustain higher clocks by way of each core being able to be allocated more power. Not having the hassle of trying to explain that for X application Mac Mini M1 Pro is actually better than MacBook Pro M1 Max because in the Mini it has better cooling or higher package power limit, or even worse same situation as before but with the Mac Studio M1 Ultra being slower because it clocks lower and the application just wants faster clocked cores and not more of them.

 

The uniformity on offer from Apple's SoC's makes things very simple for everyone.