Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP All Models Is Apple ****ing themselves over the ThreadRipper 3 yet?
- Thread starter macjunkie2013
- Start date
- Sort by reaction score
Did you not even read the very post you just quoted? because I said that exact thing in the message. It's the second paragraph in your quote even.
Arm people first have to prove its option is a viable alternative or better one than the current intel andor amd in Mac OS. Unless proven, I really dont care.
It is viable for low-power thresholds, even more so than Intel Low-Power Y series..
Does it scale for high clock speeds and HEDT replacements? Remains to be seen.
Well then, better switch to AMD, because the only processor with anywhere near this many cores from intel is 300 plus watts TDP and requires OEM liquid cooling.
Right, many people don't know that the heart of an Intel x64 processor is a RISC engine.
...also applies to AMD x64 processors. https://www.anandtech.com/show/1057...lers-micro-op-cache-memory-hierarchy-revealed
The other thing to consider is:
1. the 512 bit AVX instructions are used by very little software
2. when you have > 2x the core count, you can (and threadripper normally does) brute force the performance with more cores to STILL be faster on those vector instructions - whilst being ~2x the performance in everything else.
And yes, there are some niche workloads where the xeon may be faster. But they are very much niche. And EPYC somes in dual socket configurations. 64 cores not enough to get it done? Build a 2 socket, 128 core EPYC mac pro. Apple have done multi-socket pro machines before.
Last edited:
Yeah - but the state of the art has advanced over the past 20 years.
[automerge]1581376587[/automerge]
How low power? The Ryzen 4000U series is at 15 watts.
Graviton 2 CPU has 32 cores/32 threads, 2.5 GHz(?) Clock speed and TDP of 105W. But I may be wrong on this.
The question that bothers me is whether anybody can scale their CPUs higher than 4 GHz, to be viable on desktops, and considered High performance. For those scenarios ARM may not be as viable because no architecture scales linearly with clock speeds when it goes to IPC. The higher you go with clock speeds, the less IPC you have. Apple may be able to target specific clock speeds range, that are possible on certain nodes, in certain power levels, but can they do the same thing and scale their CPUs beyond 3 GHz, without losing IPC?
Im very sceptical at this point unfortunately. I was genuine fanboy of ARM CPUs back in the day(like - 10 years ago). Currently, from what I have seen, the cores are not powerful enough, or rather I should say: physical designs are not powerful enough.
Even - Apple designs.
Unless you can show me an instance running mac OS that I can test with my work flow, I don't know. iPadOS is yet to have a true multitasking and I ask why??
If it's really that viable, Apple must have switched already. The very fact that they are not yet to do so shows otherwise.
I also don't see industry wide move toward ARM in PC sector. Only snippet of options and synthetic bench that it has future possibility. so I really don't care.
I am not really getting into the architecture aspect ( I mean you really can't compare one ISA to another in a benchmark sense & you can change a lot in microarchitecture so I don't know),
But I have to tell you that I was very optimistic (even about 2-3 years ago) about ARM's chance in HPC/Datacenter. Especially when AMD brought Jim Keller into their design team for designing K12. But ultimately it turned out to be a vaporware & other ARM product from AMD ended up being lackluster & they dropped the ball altogether in favor of Zen. Ampere CPUs never got any foothold in datacenter market & I don't know how many Graviton CPUs Amazon have deployed in their AWS datacenters. So it remains to be seen. Many other who ventured into the server segment with ARM backed off (including Qualcomm).
Theoretically ARM CPUs have so much potential in this segment, but there isn't a single thing ARM vendors have provided that can remove that skepticism.
But I have to tell you that I was very optimistic (even about 2-3 years ago) about ARM's chance in HPC/Datacenter. Especially when AMD brought Jim Keller into their design team for designing K12. But ultimately it turned out to be a vaporware & other ARM product from AMD ended up being lackluster & they dropped the ball altogether in favor of Zen. Ampere CPUs never got any foothold in datacenter market & I don't know how many Graviton CPUs Amazon have deployed in their AWS datacenters. So it remains to be seen. Many other who ventured into the server segment with ARM backed off (including Qualcomm).
Theoretically ARM CPUs have so much potential in this segment, but there isn't a single thing ARM vendors have provided that can remove that skepticism.
Last edited:
None. AWS has Intel, AMD, and their own Graviton processors which are Annapurna Labs designed ARM processors. AWS has no need to run ARM from a third party.
I meant to say How many Graviton Amazon have deployed in their AWS datacenter.
As far as I know they haven't stated how many, but they have been expanding the number of regions where they are available and the second generation is in preview right now, with AWS planning to offer many of their own services running on Graviton2. I'd say that indicates they are very appealing to customers.
Interesting. I know Microsoft made Az3166 in a collaboration with MXCHIP, But that's mainly as a Devkit & client side tools for their IoT & Embedded division. They were also talking a lot about including Marvell ThunderX2 into Azure, but haven't heard much from them.
Linus Tech Tips: Apple should have gone AMD
With actual numbers, tests, blah blah. If this has been referenced already, my bad.
With actual numbers, tests, blah blah. If this has been referenced already, my bad.
I think that’s the whole point, actually. ARM is just an instruction set. The instruction set is really probably not relevant. Apple’s CPU designs are the actual question.
People get sidetracked on ARM vs x86 or not having ARM benchmarks running real apps or blah blah blah... It’s like asking if a race car will go faster if the driver speaks English or Chinese. It’s really not the point.
The question is really if Apple can build a CPU that stands up to AMD or Intel. If it’s ARM it doesn’t really change anything. We know RISC is fine. Intel is RISC internally. PowerPC was RISC. There are other RISC CPUs. We don’t have to ask if ARM can be fast.
The ARM chips are bloody fast (according to some code that I ran on a quad core 64 bit Intel processor and an iPhone XR, which did beat the quad core. With two fast + six slow cores against four fast ones. The latest iPads are faster than that and have three fast cores.
And while speed went down on the iPhone after a few minutes, that wasn't more than ten or 15 percent, and it was still ahead. One reason is their unified 16 MB cache between processor and the outside world (RAM, GPU everything), another reason is that they can process more instructions in parallel than Intel can.
Apple has no reason to build a 64 core ARM monster, but others have done it. Four fast + 8 slow cores would likely beat all the quad core Macs, and I think 8 + 8 wouldn't be too difficult.
On the HPC side I think everyone underestimated how much scientists hate architecture changes ?. Dealing with POWER + x86 + various GPUs already strained people, adding another CPU arch didn't add enough to make it worth the pain for a lot of that field. Maybe if it hadnt come at a time when everyone was also ramping up GPU usage and figuring out how to best incorporate CUDA/etc into their code after years of not working as heavily on vector procs in big cluster deployments....
Am I missing something? Are the professional-grade AMD CPUs decent? Because, from my experience, the consumer lineup is still garbage—even post-Ryzen.
You should do some reading on the subject. As it stands AMD processors are less susceptible to many of the vulnerabilities that plague Intel processors. They have higher IPC, better SMT performance and are cheaper as well. The only downside right now is AVX512 performance.
I got convinced to buy a Ryzen processor for a build I was doing... and, well, I regretted it immediately; I was lucky that I was able to return the parts. No matter what OS I tried (Windows, OSX, and Linux), my system would intermittently lock up for bout 15 seconds at a time. A friend's system did the same thing. In addition, the motherboard I was using—which was not a cheap board—had input lag... of all things. So, when I plugged in a keyboard and tried typing on it, my strokes were noticeably delayed. I tried several other keyboards, and I still felt the lag. I had never experienced anything like it. So, I returned both pieces of garbage (the AMD CPU and the motherboard), and I settled for an equivalent Intel CPU and a cheaper LGA motherboard. And, problem solved. No issues since. The computer runs better than it ever had.