Linus Torvalds rants against Intel's ECC market segmentation practices

[StDragon]

Summary

Linus Torvalds calls out Intel's BS market segmentation practices with regards to ECC memory usage and adoption within the industry

 

Quotes

Quote

ECC absolutely matters.

ECC availability matters a lot - exactly because Intel has been instrumental in killing the whole ECC industry with it's horribly bad market segmentation.

Go out and search for ECC DIMMs - it's really hard to find. Yes - probably entirely thanks to AMD - it may have been gotten slightly better lately, but that's exactly my point.

Intel has been detrimental to the whole industry and to users because of their bad and misguided policies wrt ECC. Seriously.

And if you don't believe me, then just look at multiple generations of rowhammer, where each time Intel and memory manufacturers bleated about how it's going to be fixed next time.

 

My thoughts

 

I've known this for 20+ years. Intel reserves ECC compatibility near exclusively for their Xeon lineup. There are rare SKUs for Core series, but those are usually for embedded appliance usage.

 

And while bit-flip errors are rare, they do happen. Its only going to get worse as density increase. ECC being a mainstream option is long overdue IMHO.

 

Google published a paper on this back in 2009 for anyone interested in the read.
 

 

Sources

https://www.realworldtech.com/forum/?threadid=198497&curpostid=198647

 

https://research.google/pubs/pub35162/

https://storage.googleapis.com/pub-tools-public-publication-data/pdf/35162.pdf

[TVwazhere]

-Moved to CPU's, Motherboards and Memory-

 

This does not comply with Tech News Posting Guidelines, namely:

Quote

• Your thread must include a link to at least one reputable source. Most of the time, this should be a respected news site.

A rant about certain company marketing practices on an archived forum from what appears to be an email chain does not count as a credible news source. If we had news articles written with such sources commonly, @jonnyGURU would likely be the #1 cited person. 

[Guest]

Since when was ECC hard to find?? Lol. 

 

Then ECC as a mainstream purchase for desktop computing?

 

And you think RAM is expensive now.... wait till it's all ECC memory. 

[StDragon]

1 hour ago, TVwazhere said:

-Moved to CPU's, Motherboards and Memory-

 

This does not comply with Tech News Posting Guidelines, namely:

A rant about certain company marketing practices on an archived forum from what appears to be an email chain does not count as a credible news source. If we had news articles written with such sources commonly, @jonnyGURU would likely be the #1 cited person. 

Fair enough. Your rules, your guidelines. Just that was the Linus Torvalds speaking. He could have scrawled that (very insightful rant) on a napkin and uploaded it as a photo. It would still be news worthy in of itself. But, whatever.

News shouldn't be who published it, rather where, or in this case, whom it was sourced from.

[StDragon]

1 hour ago, ShrimpBrime said:

Since when was ECC hard to find?? Lol. 

 

Then ECC as a mainstream purchase for desktop computing?

 

And you think RAM is expensive now.... wait till it's all ECC memory. 

 

Because of market segmentation practices that Intel does, ECC isn't mainstream on desktop computers. Yes, you can get an Intel workstation, but that has to be speced with a Xeon. In addition, the MB has to support it as well.

 

As such, due to the lack of mainstream support, there isn't much of a market for ECC outside of the server segment. So, scales of economy do matter in terms of keeping the costs low per DIMM with regards to mass-production.

 

The last desktop I built that had ECC was based on an AMD Phenom II. It's still actively in use by a relative. Never had a Windows crash (BSOD / kernel panic).

[Guest]

1 hour ago, TVwazhere said:

@jonnyGURU would likely be the #1 cited person. 

[Guest]

42 minutes ago, StDragon said:

 

Because of market segmentation practices that Intel does, ECC isn't mainstream on desktop computers. Yes, you can get an Intel workstation, but that has to be speced with a Xeon. In addition, the MB has to support it as well.

 

As such, due to the lack of mainstream support, there isn't much of a market for ECC outside of the server segment. So, scales of economy do matter in terms of keeping the costs low per DIMM with regards to mass-production.

 

The last desktop I built that had ECC was based on an AMD Phenom II. It's still actively in use by a relative. Never had a Windows crash (BSOD / kernel panic).

I've had systems last years with non ECC memory without a blue screen also. But that's not a good argument and redundant to what ECC memory is and what it does.

 

So first off, the loss in performance for the gamer types is an automatic turn away. ECC memory is meant to be redundant. Often times, sold at defaults, so DDR4 most kits would be 1067mhz (2133mt/s) and some 1200mhz (2400mt/s). High speed memory is not a thing with ECC kits. Plus the cost increase?

 

So for example, the redundancy, most ECC memory has an extra module. This is one factor that increases the pricing. But all these things are what make ECC memory very reliable and best suited for servers and workstations running software that would lean on the benefit of ECC memory.

 

And no, ECC does not prevent blue screens. I've seen several on my old K8N-DRE motherboard. They where not memory related. It was Cpu related. That's what overclocking does. Also, I wasn't running Windows Server operating system either. Which also the software running in that OS would be optimized to use ECC memory.

 

It's just not for desktops. Some desktop platforms you can buy ECC memory and run it, but at a higher cost for redundancy but at a loss for performance. It's a give and take thing.

 

I as a consumer would not want ECC memory in my high performance gaming rig. If I blue screen during game, no big deal. I'll adjust my OC and carry on. But that's my opinion obviously.

 

 

[StDragon]

2 hours ago, ShrimpBrime said:

I've had systems last years with non ECC memory without a blue screen also. But that's not a good argument and redundant to what ECC memory is and what it does.

Kernel Panics (aka, BSODs) are the result of buggy drivers (kernel space), or a hardware fault (bit-flip) causing a calculation error. But yes, OCing a CPU or having bad capacitors / VRMs feeding power into any of the surface mount components can effect MB and peripherals too. 

 

2 hours ago, ShrimpBrime said:

So first off, the loss in performance for the gamer types is an automatic turn away. ECC memory is meant to be redundant. Often times, sold at defaults, so DDR4 most kits would be 1067mhz (2133mt/s) and some 1200mhz (2400mt/s). High speed memory is not a thing with ECC kits. Plus the cost increase?

 

So for example, the redundancy, most ECC memory has an extra module. This is one factor that increases the pricing.

 

I think the argument is economies of scale. If Intel hadn't segmented the memory market between Core and Xeon, it's possible we would see a lot cheaper ECC DIMMs out there. In fact, you might even see "gamer" or "high performance" ECC DIMMS. But that market doesn't exist because no one is seriously gaming with a Xeon to have any measurable impact on the PC gaming DYI market. Now granted, there is a limit to how fast you can clock ECC as that's a factor with the MB, CPU, and whether or not they're buffered (extra latency)

2 hours ago, ShrimpBrime said:

 Which also the software running in that OS would be optimized to use ECC memory.

ECC is OS agnostic. Short of reading the BIOS, the kernel isn't aware if ECC is being used or not, nor would it care. It's completely abstracted from what is a hardware function.

You're confusing ECC for Persistent Memory (NVDIMM or Optane). To use Persistent Memory in Windows 10, you must be licensed for either Pro for Workstations, or Enterprise.

 

2 hours ago, ShrimpBrime said:

It's just not for desktops. Some desktop platforms you can buy ECC memory and run it, but at a higher cost for redundancy but at a loss for performance. It's a give and take thing.

It is. But not everyone uses their PC as a gaming machine. Some prefer a reliable platform with that extra assurance without paying for full server grade fault tolerance price. But like Torvalds, I agree this should be left to the free market, and Intel is anything but that. But at least AMD is playing nice.

[Guest]

44 minutes ago, StDragon said:

Kernel Panics (aka, BSODs) are the result of buggy drivers (kernel space), or a hardware fault (bit-flip) causing a calculation error. But yes, OCing a CPU or having bad capacitors / VRMs feeding power into any of the surface mount components can effect MB and peripherals too. 

 

 

I think the argument is economies of scale. If Intel hadn't segmented the memory market between Core and Xeon, it's possible we would see a lot cheaper ECC DIMMs out there. In fact, you might even see "gamer" or "high performance" ECC DIMMS. But that market doesn't exist because no one is seriously gaming with a Xeon to have any measurable impact on the PC gaming DYI market. Now granted, there is a limit to how fast you can clock ECC as that's a factor with the MB, CPU, and whether or not they're buffered (extra latency)

ECC is OS agnostic. Short of reading the BIOS, the kernel isn't aware if ECC is being used or not, nor would it care. It's completely abstracted from what is a hardware function.

You're confusing ECC for Persistent Memory (NVDIMM or Optane). To use Persistent Memory in Windows 10, you must be licensed for either Pro for Workstations, or Enterprise.

 

It is. But not everyone uses their PC as a gaming machine. Some prefer a reliable platform with that extra assurance without paying for full server grade fault tolerance price. But like Torvalds, I agree this should be left to the free market, and Intel is anything but that. But at least AMD is playing nice.

Thank you for some of the corrections there. 

Correctly you are accurate not everyone uses a PC as a gaming machine.

 

So OEMs then. mainstream PC. 

 

That'll be hard to sell to the old folks looking at a 450$ PC. 

If knowing they didn't have BSODs before with a machine that didn't have ECC, how would we market this memory to promise less or no BSODs??

But now would pay 600$ (just throwing numbers out there). What's the sale's pitch? "It's as stable as a server"?

 

Pardon me, just trying to find a marketing angle that Intel or AMD would use to BS the extra cost to save on BSODs....