The Intel Mac Pro used server class hardware. I was wondering if the M2 Ultra Mac Pro will be using ECC memory? I am guessing not, but hoping i missed it. Thanks.
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 ECC Memory
- Thread starter solardudesf
- Start date
- Sort by reaction score
I think its fine for most users.
ECC Ram is much slower than consumer DDR5 anyway.
If you want server environment, go to PC/Windows/Linux
Modern ECC memory isn’t that much slower. Also with the amounts of ram we have, errors are basically guaranteed on a daily basis now. At this point, ECC should now be standard on all computers.
Computer Device Makers Ignore Bit-Flip Errors and Data Corruption
John Kheit wants to know why all modern device makers haven't moved to ECC memory considering the problem of bit-flip errors.
www.macobserver.com
Modern ECC memory isn’t that much slower. Also with the amounts of ram we have, errors are basically guaranteed on a daily basis now. At this point, ECC should now be standard on all computers.
Computer Device Makers Ignore Bit-Flip Errors and Data Corruption
John Kheit wants to know why all modern device makers haven't moved to ECC memory considering the problem of bit-flip errors.www.macobserver.com
It's not going to happen ever on the M chips. Time to move on.
You're in the niche market, maybe move to a PC?
Interesting article. But does it really mean anything? There aren't millions of reports of memory errors which would be implied by the article. I've been running high memory systems for a long time and have never encountered a problem.
LPDDR can potentially do ECC but whereas regular DDR "sideband" ECC physically adds check bits to each word (and needs a wider data bus) LPDDR uses "inline" ECC which allocates a chunk of regular memory to hold the checksums. That's as much as I'm going to pretend to understand - more detail here: https://www.synopsys.com/designware-ip/technical-bulletin/error-correction-code-ddr.html
I think one of the snags for Apple Silicon is that it already has RAM capacity limited (by workstation standards) - at least partly - by the size/capacity of RAM chips that can be fitted on the package and, however you implement it, ECC needs more chips for the same amount of usable RAM. Also, Apple Silicon's superpower is having unified video RAM which allows the GPU to punch above its weight, so anything that slows down that RAM might not be a good idea.
Really, though, it is pointless to argue about ECC vs. no ECC without figures as to what the error rate is with Apple Silicon's somewhat unique RAM implementation vs. regular DDR. Anybody know?
I think the other hard fact is that the new Mac Pro simply isn't a drop-in replacement for the 2019 MP - it addresses one very specific niche of people who need more non-GPU PCIe bandwidth than you can get with external TB3, otherwise it is irrelevant. Presumably Apple think that's an important niche.
A lot of the errors will hit data and wont be noticed. Will you notice a color in a single pixel in a Jpeg change? Or a single color pixel in some icon gets changed? The problem is that the errors compound and over time lead to rot.
Of course some people may not care, but for me, I would really like ECC on all my devices. Particularly since most modern devices are ALWAYS ON. Just increases the corruption over time.
Bit-level errors are very exaggerated in compressed graphics and turn into whole streaks and patches.
Most people don’t care about ECC and wouldn’t notice, but there are people who really care about their data and the data path. There are many cases where a single bit error would cause a file to be completely corrupted, a computer operation to come out wrong, etc. Respectfully, I doubt that you have never encountered a problem, more likely never noticed a problem (which is good enough for most people). When there are problems like this people often write it off and it has become commonplace for people to just shrug their shoulders and reboot their system. People who manage clusters with many machines have ECC and you almost always encounter ECC errors with enough machines. With desktop class memory you wouldn’t ever see an error in the logs about it and would not know it happened, so of course there aren't millions of reports of “memory” errors. The average person doesn’t even know it is happening.
Last edited: