Read more.This processor will target laptop designs as well as high-end Android devices.
Read more.This processor will target laptop designs as well as high-end Android devices.
I'm still not convinced that the M1 is that great a chip. Only good because of the tie in and the fact that Apple can leverage all their resources to make it work. Windows isn't great on ARM because of many reasons and the chips would have to be absolute beasts to turn most away from X86
Old puter - still good enuff till I save some pennies!
This come up all the time.
Yes, Apple are willing to pay for the latest manufacturing process and do throw transistors at the problem.
But throwing transistors at the problem is what semiconductors are all about and I don't see any evidence of them wasting die space.
However, M1 and A7-A14 before that are in a place which Qualcomm, Samsung, and Nvidia never reached.
Qualcomm and Samsung have given up on making their own cores after years of trying really really hard.
Nvidia, well they still seem to be doing their own thing but the Nintendo Switch is their only real success.
So that only leaves ARM who seem to focusing on high-performance now but are still really far behind Apple.
I hate Apple's closed walled-off-garden as much as the next PC enthusiasts, but there's no denying how well Apple have executed since buying PA Semi.
DanceswithUnix (11-05-2021)
.....Qualcomm, Samsung, and Nvidia never had a market for their version of M1. Apple is both an operating system and hardware company while those other guyz have to rely on Microsoft because of market share.
That's totally my take on it. I won't be buying any of their products, but I have to applaud their engineering here.
It is pretty hard to compare amd64 and Arm cores when the Arm cores always come on a highly integrated SoC, but I suspect if you could pair an M1 up with an RTX 3090 it would be pretty potent. OFC that assumes you can run the software you want on it.
This is it exactly....
Others say it's great but it's so closely tied in it's not the chip that makes it great but both the hardware AND the software
Dances - mate the M1 has soooooo much removed (transistor AND otherwise) to use in an entry level system that you could never pair it with an rtx card or anything else. Heck it's single channel memory with a couple of usb channels so even getting great usb performance is difficult without adding extra chips and supporting hardware.
I'm not saying the results aren't great but that's because of all that Apple can do - NOT the chip. There isn't anything on a windows/linux level because nobody wants that hardware (though one could argue that now Apple have the M1 this WILL happen) because of the lack of flexibility.
Samsung putting in AMD gfx tech is one step towards that. Using a single x1 core won't help though as for many applications we're back to great 1 thread performance and so so multi-thread. And that is before we throw in the windows scheduler that can't really cope with how ARM uses BIG.little cores anyhoo.
I have a few Pi's here. $35 buys you a heck of performance on a Pi4 with 2 gig ram etc. Linux runs great on it - perfectly useable for an office based pc and pretty close to an M1 setup. It can be done - just it wouldn't perform anywhere near as good in an X86 situation which is what I mean. And that's it in a nutshell. Windows + ARM is nowhere at the moment and it's still what a good 90% of people look at and need right now
Old puter - still good enuff till I save some pennies!
I'm interested to see what Apple do with the M1.
I wonder how many of those extra transistors can be powered at the rated TDP. Power requirements for transistors is decreasing slower than other improvements from node improvements. So I hear from someone who actually knows about these things and I just parrot to sound clever.
The m1 is a 15w chip in many situations but can be upped a bit (18w and 20w are easily done so I believe). The m2 is rumoured to be a 35w part so would fit in with industry standards for many laptop designs. Samsung has decent SoC's but they have lacked with thermal throttling and thus reduced battery life compared to qualcomm designs. The Apple cores are a bit ahead of the qualcomm ones. Rated TDP isn't really a decent metric anymore as it is so different from manufacturer to manufacturer and nearly always can be configured per design
Old puter - still good enuff till I save some pennies!
So finally an AMD GPU you can buy??!!
Yes, and in newer designs, in order to comply with the TDP a load of the chip consists of dark transistors which can't be active and also comply with TDP. It's a fact of life. So, if Apple throw a load of transistors at the problem, either they'll violate their own TDP spec (unlikely) or a load of the chip will have to be switched off. Given the modular design, it's likely that is doable with this kind of core configuration. I think, in the mobile space, rated TDP is still valuable as violating this too much will simply lead to insane cooling or throttling issues.
Aye Phil - but my point was that simply talking about a TDP is kinda simplistic. Point in case - on this very website this morning is a piece about Intel laptop cpus. TDP is rated at 45w yet they can use up to 110w burst. That's 2 and a bit times the "rated" TDP
More interestingly I've just noticed that if you have a Pi4/400 there is now a pretty easy way to get Win 10 ARM edition running on them.
This opens up the whole ARM cpu v x86 debate in a more meaningful way and means that if someone can get a decent ARM chip for Win 10 up and running in the vein of the M1 we can start and see if it will be better in the future. Pi400 was used in the tests I saw, I imagine that someone will port it to say a Mac Mini to see how it performs but the Pi400 came in at around the sort of spec of a thin and light from say 4 years ago. I'd imagine that a new AMD equipped Exynos would be around the speed of an M1 chip in many ways and could, if done right, kickstart this whole movement. I wonder how many people would be interesting in a Mac Mini with Win 10 installed as for example their work pc
Old puter - still good enuff till I save some pennies!
Or to take it further - how many people would be interested in a truly decent Win 10 SBC? I know I would for certain applications/useage scenarios
Old puter - still good enuff till I save some pennies!
New markets for AMD here, how long until they are producing graphics for all the ARM chips?
Or maybe not?
https://www.cpubenchmark.net/compare...62vs3930vs3938
Not that passmark is the most accurate benchmark and multi-platform is always tricky to bench.
The Exynos 2100 is pretty recent:
- 5nm (5LPE) Samsung process[145]
- 6MB System Cache
- CPU features
- 1 + 3 + 4 cores (2.91 GHz Cortex-X1 + 2.81 GHz Cortex-A78 + 2.2 GHz Cortex-A55)
- 19% better perf on single thread[146]
- 33% better perf on multi threads[147]
Yet has a single threaded score nearly half of the M1.
I believe Samsung or Qualcomm can beat Apple but the issue is marketing. What is the proportion of the market that is willing to fork off dollars? I mean Apple has their fan base but Samsung has been struggling to sell off their Exynos because of stiff competition (Qualcomm & Mediatek).
Almost every generation, Exynos was way behind Snapdragon. So much so that Samsung got rid of their custom CPU team.
Qualcomm similarly gave up on Kryo and now uses standard ARM cores with a few tweaks:
Now ARM has big plans to go after the high performance market with X1 and similar, but as the Passmark scores I posted show they have a long way to catch M1 or even the older A12X. Some of that is due to the power budget, some due the process, but note that the 2021 Snapdragon 8350 is 5nm Samsung so possibly not that far behind TSMC 5nm and Apple's A12X was of TSMC 7nm. Can't find anything about transistor count or die size of the Snapdragon.Snapdragon 888 (SM8350) made with 5nm (5LPE) Samsung process, features:
- 1 Kryo 680 Prime (ARM Cortex-X1-based), up to 2.84 GHz. Prime core with 1 MB pL2 and 64 KB pL1
- 3 Kryo 680 Gold (ARM Cortex-A78-based), up 2.42 GHz. Performance cores with 512 KB pL2 each
- 4 Kryo 680 Silver (ARM Cortex-A55-based), up 1.8 GHz. Efficiency cores with 128 KB pL2 each
- Move to instruction set ARMv8.4-A (From ARMv8.2-A)
- DynamIQ with 4 MB sL3,
- 25% performance uplift and 25% power efficiency improvement
There are currently 1 users browsing this thread. (0 members and 1 guests)