Why is ECC memory so expensive?

Now that DDR5 supports ECC over-the-wire by design, I’m starting to look at putting ECC into my next build. However, the price is insane. Looking at Crucial’s offerings, a non-ECC 16GB DDR5-4800 DIMM is $71, yet a 16GB DDR5-4800 ECC DIMM is $130. Sure, it has 25% more silicon, plus a bit of extra circuitry and lower volumes. But I’d think that should be closer to $100, not $130.
What gives?

DDR5 UDIMM ECC memory prices are still completely out of wack because production volumes haven’t ramped yet… If I had to guess its because there aren’t many platforms that take DDR5 ECC UDIMMs.

It is the DDR5 ECC RDIMMs that have ramped up in production and have reasonable prices at the moment.

The biggest reason ECC memory is so expensive is because Intel killed off support for it for consumer grade hardware over a decade ago to push those who explicitly know they need it to Xeons to make more money, so the DRAM manufacturers shifted towards marketing ECC as only for businesses because they knew they could copy the strategy and increase prices on it as well.

3 Likes

I think I’ve read once (was before DDR5) that many consumer memory was of lower class than ECC memory.
ECC memory achieve their rated performance at JEDEC spec level, while consumer memory would actually run slower than ECC memory if they would be run at JEDEC spec without XMP overvolting

ECC should have outcompeted non-ECC a long time ago. At the very least ECC support should be ubiquitous across all motherboards just like XMP is now.

The fact that it still isn’t, tells me that either Intels pricing shenanigans worked, or ECC is just not as big of a deal we were led to believe for most tasks. My money is on the latter.

This assumes that customers had a choice in the matter though. When the dominant CPU manufacturer (at the time, this was during the Nehalem days I believe) decides to completely remove support for a feature in their chipsets, you no longer have the ability to choose. AMD providing unofficial ECC support on the Ryzen platform and official support on the Ryzen Pro platform has certainly been a boon for them and helped them sell products to those who are knowledgeable with computers and understand the value of ECC. Your average computer user has incredibly little understanding of how a computer works, but their ineptitude doesn’t mean that they deserve an inferior product that’s capable of having severe issues due to manufacturers cutting corners to increase profits. Many countries have laws against that very practice when it comes to other products, like electronics that won’t burn your house down, cars that can roll over too easily due to dangerous designs, etc.

As large-scale studies have shown, memory errors happen quite regularly in the wild. As someone who has worked on a mature piece of software for a few years that is used by hundreds of thousands of customers on consumer grade hardware, I have seen plenty of absolutely insane one-off bugs that have never been seen again and have resulted in many developer and QA hours spent going down rabbit holes trying to reproduce them with no results. I have a strong feeling and supporting evidence that if ECC DIMMs were ubiquitous in consumer hardware, a feature that doesn’t cost all that much more to implement, that many of those one-off bugs would have never happened.

4 Likes
  • Lower demand
  • Higher quality demands (enterprise don’t want faliure - what may pass as acceptable for consumer may not be stringent enough QC for enterprise - individual consumers are more willing to deal with returns and aren’t buying as much - thus they won’t personally be able to compare failure rates across brands, etc)
  • Target market willing to pay more (enterprise)
  • 12.5% additional capacity included for error detection

I’ve had systems saved by ECC before, but not as far as I am aware, bit-rot related. Systems that have detected errors and then run just fine until the memory was replaced.

I suspect many crashes that people just write off as Windows being Windows are in fact memory glitches.

When your normal uptime is measured in hours or days (typical consumer machine) you’re a lot less aware of stability problems than if say, your hypervisor has been up for 6 months or more.

I have had systems with uptime measured in years running ECC.

Because there are multiple kinds of ECC memory. You have unbuffered, registered and load-reduced ECC memory. All of them requires extra circuitry to ensure RAS so that’s why it’s expensive.

Seems to me like it’s somewhat of a “market will bear it” scenario. Those that need it will pay for it. If you don’t need it then its cost prohibitive and no one cares about you. Not trying to be mean here… Just trying to justify the high prices. Couple all of this with the low yield and slow ramp up as previously mentioned and it seems like that would explain it.

1 Like

Since it wears the label ECC, the QA/QC will be far more aggressive
Between the tiers of ECC, leads to more stuff/material on the DIMMs
Price spiking is also the enterprise realm, [generally] willing to fork out the money

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.