The 2TB M never made any sense. The CPU supports 6-channels and 2DPC, so that means 1.5TB for 128GB, 3TB for 256GB, and 4.5TB with 256GB DRAM and 512GB Optane is the most logical configuration.
In order to reach 2TB, you need some channels to be in asynchronous channel mode with total of 16 slots. 16 doesn't evenly divide into 6.
And speaking philosophically, I bet the economy will have less drastic ups and downs if companies with market dominance stabilized their prices rather than pricing it to the level of limit and quickly reducing it when faced with stiff competition.
What if say they priced the top chip "L" at $6k for example?
Then investors cry when the market crashes massively. Let's say if the drops were less severe(conversely that means the gains are less severe too). Stability is better even if it means the potential wins are reduced.
"And speaking philosophically, I bet the economy will have less drastic ups and downs if companies with market dominance stabilized their prices"
you must be some left-wing Red Nut!!! :) oligopolies through to out-right monopolies always do exactly the opposite, in order to max profit: it's called market segmentation and it max profit by taking what the econ types call 'consumer surplus'. if you look at econ supply/demand curves with the 'price' determined at the crossing point, you see that there are some (perhaps a lot) consumers to the left of the equilibrium price who are willing to pay more. identify them and charge them. most STEM graduates are engaged in that very task for corporations.
Actually if I were to pick a side I'd be right wing. I'm in Canada but if I were to vote I'd probably have voted for trump.
Realistically speaking though, I believe politicians in general are mediocre to rule at best. After all, they are looking for their own best interests.
I agree about your view, but taken to the extreme, it just makes me question things. Does it always have to be *that* way?
Cheers user, I see it the same way you just wrote the statements you have made.
Just cause it always has been done a certain way, and the massive companies want to price things as they do, should they, and why on earth does anyone support such "game play"
I am sure it hurts EVERYONE, maybe not the company making the massive profits, but everyone who is not, as they certainly pay for it, likely many times over.
Dead by trillions of paper cuts this world is coming to O.o
Yes, to go past 1.5TB you need to use 256GB DIMMs, or 512GB Optane DIMMs - both extremely expensive options - 256GB DIMMs are so rare and expensive that they are not even listed online. The sales of L models has to be very low.
If you have a system with 2 or 4 sockets, each socket's memory is visible to the other sockets. So a 4 socket system can have 3TB with ordinary 64GB DIMMs (4 x 12 x 64GB).
If you want to know about security vulnerabilities on Intel hardware and the resulting performance losses as a result of mitigations, then Anandtech is not going to tell you. The security vulns and performance losses on Intel hardware continue to rack up, literally month by month, but not a peep out of Anandtech. Do a deep dive on that, Dr Ian!
Do you have a link, as I don't find any such article on Anandtech that addresses in any significant detail the vulnerabilities in Intel CPUs (and now you can add Intel iGPUs to the list) - perhaps you have this site confused with a more editorially independent publication? You don't even see the issue mentioned on Anandtech when discussing new silicon designs, it's as though the issue doesn't exist...
Thanks for the link but we've had far more vulnerabilities since Spectre & Meltdown, none of which have been discussed by Andandtech - perhaps not surprisingly, as they almost entirely reflect badly on Intel.
As it was mentioned on Toms Hardware in a far more recent article from Nov 2019:
"Intel currently has 242 publicly disclosed vulnerabilities, while AMD has only 16 (a 15:1 difference in AMD’s favor)"
Sadly even that figure is already out of date, and vulnerabilities are now being discovered in Intel GPUs (the performance losses seen so far from the - admittedly - early GPU mitigation patches - although in development since August 2019 - are absolutely catastrophic for Gen 7/7.5, see phoronix.com for details).
I guess it's so much easier to have a photo op with a silicon wafer than it is to write about the real issues affecting users and their ever slower hardware. Or is a die shot the price for ignoring these issues? I think we all know the answer on that one!
Well, this is a response to EPYC, and nothing else. Intel doesn't normally reduce prices, so these rounds of price reductions have been a pleasant surprise.
It's a win compared to their previous stance, but they're still charging extra to use the processor's full RAM support capabilities, and that's one of the more grotesques examples of how their product stack is fundamentally compromised by artificial segmentation.
not quite. there is no cpu, to my knowledge, that addresses 64 bit address space. building such a chip/motherboard with SCM and an OS that treats SCM as single-level store rather than a filesystem surrogate will change transactional applications by light-years. ultimately, a transaction would need only a memory latch and execute at memory speed. existing OSes can't handle that protocol, I'd wager.
They had hit 5% server market share in September 2020. I haven't seen new numbers since then but I wouldn't doubt they are at 6-7% now. Which still isn't great but I'm certain its moving in a positive direction.
This would require some support on the Epyc side of things as well. I know a few features are there (memory encryption etc.) but it is not clear if everything is there on-die already to support byte addressable NVDIMMs on Epyc.
It is very likely and would be mutually beneficial for both AMD and Micron. AMD is taking an open-ecosystem strategy, and would like to adopt whatever the best technology out there, meanwhile Micron also wants to crack the dominance of Intel in high-margin server memory market.
AMD's EPYC 7742 which is 64 cores, supports 4TB of max memory and supports 2P(two processors) is selling for under $7K. Only time will tell but I do wonder if these costs will be enough. These larger chips I have to assume will be used for VM's / server consolidation and supporting 4TB is a huge plus and having 128 cores in 2P configuration is also a huge plus.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
35 Comments
Back to Article
spikebike - Thursday, January 16, 2020 - link
Makes sense, anything to slow the migration to the AMD Epyc CPUs. Why pay a premium just to be able to address more RAM, AMD doesn't.IntelUser2000 - Thursday, January 16, 2020 - link
The 2TB M never made any sense. The CPU supports 6-channels and 2DPC, so that means 1.5TB for 128GB, 3TB for 256GB, and 4.5TB with 256GB DRAM and 512GB Optane is the most logical configuration.In order to reach 2TB, you need some channels to be in asynchronous channel mode with total of 16 slots. 16 doesn't evenly divide into 6.
IntelUser2000 - Thursday, January 16, 2020 - link
And speaking philosophically, I bet the economy will have less drastic ups and downs if companies with market dominance stabilized their prices rather than pricing it to the level of limit and quickly reducing it when faced with stiff competition.What if say they priced the top chip "L" at $6k for example?
Then investors cry when the market crashes massively. Let's say if the drops were less severe(conversely that means the gains are less severe too). Stability is better even if it means the potential wins are reduced.
FunBunny2 - Thursday, January 16, 2020 - link
"And speaking philosophically, I bet the economy will have less drastic ups and downs if companies with market dominance stabilized their prices"you must be some left-wing Red Nut!!! :) oligopolies through to out-right monopolies always do exactly the opposite, in order to max profit: it's called market segmentation and it max profit by taking what the econ types call 'consumer surplus'. if you look at econ supply/demand curves with the 'price' determined at the crossing point, you see that there are some (perhaps a lot) consumers to the left of the equilibrium price who are willing to pay more. identify them and charge them. most STEM graduates are engaged in that very task for corporations.
IntelUser2000 - Thursday, January 16, 2020 - link
Actually if I were to pick a side I'd be right wing. I'm in Canada but if I were to vote I'd probably have voted for trump.Realistically speaking though, I believe politicians in general are mediocre to rule at best. After all, they are looking for their own best interests.
I agree about your view, but taken to the extreme, it just makes me question things. Does it always have to be *that* way?
Dragonstongue - Thursday, January 16, 2020 - link
Cheers user, I see it the same way you just wrote the statements you have made.Just cause it always has been done a certain way, and the massive companies want to price things as they do, should they, and why on earth does anyone support such "game play"
I am sure it hurts EVERYONE, maybe not the company making the massive profits, but everyone who is not, as they certainly pay for it, likely many times over.
Dead by trillions of paper cuts this world is coming to O.o
lefty2 - Thursday, January 16, 2020 - link
Yes, to go past 1.5TB you need to use 256GB DIMMs, or 512GB Optane DIMMs - both extremely expensive options - 256GB DIMMs are so rare and expensive that they are not even listed online.The sales of L models has to be very low.
btkcsd - Tuesday, February 9, 2021 - link
If you have a system with 2 or 4 sockets, each socket's memory is visible to the other sockets. So a 4 socket system can have 3TB with ordinary 64GB DIMMs (4 x 12 x 64GB).AshlayW - Thursday, January 16, 2020 - link
Thanks to AMD, essentially. Intel's stranglehold on the industry is starting to break.Irata - Thursday, January 16, 2020 - link
Well, at least we know it can't be due to their iGPU security bug (check Phoronix) as it's a server CPU.Competition that offers extended memory at standard prices perhaps ?
CityBlue - Thursday, January 16, 2020 - link
If you want to know about security vulnerabilities on Intel hardware and the resulting performance losses as a result of mitigations, then Anandtech is not going to tell you. The security vulns and performance losses on Intel hardware continue to rack up, literally month by month, but not a peep out of Anandtech. Do a deep dive on that, Dr Ian!Korguz - Thursday, January 16, 2020 - link
um.. i believe anandtech did an article on that a little while ago.. say.. around november...CityBlue - Thursday, January 16, 2020 - link
Do you have a link, as I don't find any such article on Anandtech that addresses in any significant detail the vulnerabilities in Intel CPUs (and now you can add Intel iGPUs to the list) - perhaps you have this site confused with a more editorially independent publication? You don't even see the issue mentioned on Anandtech when discussing new silicon designs, it's as though the issue doesn't exist...Korguz - Friday, January 17, 2020 - link
i think this was the one i was referring to :https://www.anandtech.com/show/13659/analyzing-cor...
CityBlue - Friday, January 17, 2020 - link
Thanks for the link but we've had far more vulnerabilities since Spectre & Meltdown, none of which have been discussed by Andandtech - perhaps not surprisingly, as they almost entirely reflect badly on Intel.As it was mentioned on Toms Hardware in a far more recent article from Nov 2019:
https://www.tomshardware.com/features/intel-amd-mo...
"Intel currently has 242 publicly disclosed vulnerabilities, while AMD has only 16 (a 15:1 difference in AMD’s favor)"
Sadly even that figure is already out of date, and vulnerabilities are now being discovered in Intel GPUs (the performance losses seen so far from the - admittedly - early GPU mitigation patches - although in development since August 2019 - are absolutely catastrophic for Gen 7/7.5, see phoronix.com for details).
I guess it's so much easier to have a photo op with a silicon wafer than it is to write about the real issues affecting users and their ever slower hardware. Or is a die shot the price for ignoring these issues? I think we all know the answer on that one!
FunBunny2 - Friday, January 17, 2020 - link
isn't Tom's owned by the same folks??? Yes, yes it is.Korguz - Friday, January 17, 2020 - link
um.. .the one i linked to.. is dated Dec 3rd.... so its newer then the toms article you posted..CityBlue - Friday, January 17, 2020 - link
@Korguz Look again... Small matter of the year... Anandtech Dec 3rd 2018... Tom's Nov 4th 2019... 2019 > 2018.Korguz - Saturday, January 18, 2020 - link
doh..... didnt see that...yetanotherhuman - Thursday, January 16, 2020 - link
Well, this is a response to EPYC, and nothing else. Intel doesn't normally reduce prices, so these rounds of price reductions have been a pleasant surprise.Spunjji - Thursday, January 16, 2020 - link
It's a win compared to their previous stance, but they're still charging extra to use the processor's full RAM support capabilities, and that's one of the more grotesques examples of how their product stack is fundamentally compromised by artificial segmentation.Sahrin - Thursday, January 16, 2020 - link
It’s a testament the failure of unregulated capitalism that Intel can charge these premiums.cheshirster - Thursday, January 16, 2020 - link
This should be."Intel cuts price premium for full memory support on Xeons by 60%".
FunBunny2 - Thursday, January 16, 2020 - link
"full memory support"not quite. there is no cpu, to my knowledge, that addresses 64 bit address space. building such a chip/motherboard with SCM and an OS that treats SCM as single-level store rather than a filesystem surrogate will change transactional applications by light-years. ultimately, a transaction would need only a memory latch and execute at memory speed. existing OSes can't handle that protocol, I'd wager.
ksec - Thursday, January 16, 2020 - link
Just to keep everyone updated AMD still only has 4% of Market Shares.eva02langley - Thursday, January 16, 2020 - link
6 months ago, Lisa Su was saying they where mid single digit, your numbers doesn't hold up.FreckledTrout - Thursday, January 16, 2020 - link
They had hit 5% server market share in September 2020. I haven't seen new numbers since then but I wouldn't doubt they are at 6-7% now. Which still isn't great but I'm certain its moving in a positive direction.FreckledTrout - Thursday, January 16, 2020 - link
September 2019.... ugh new year.Rudde - Thursday, January 16, 2020 - link
For reference, AMD 2nd gen EPYC supports 4 TB of RAM memory (Anandtech 2019).chris.london - Thursday, January 16, 2020 - link
A wild stab in the dark: is Micron about to announce its 3D XPoint based product line that is compatible with AMD Epyc?Kevin G - Thursday, January 16, 2020 - link
This would require some support on the Epyc side of things as well. I know a few features are there (memory encryption etc.) but it is not clear if everything is there on-die already to support byte addressable NVDIMMs on Epyc.hunter168 - Thursday, January 16, 2020 - link
It is very likely and would be mutually beneficial for both AMD and Micron. AMD is taking an open-ecosystem strategy, and would like to adopt whatever the best technology out there, meanwhile Micron also wants to crack the dominance of Intel in high-margin server memory market.eva02langley - Thursday, January 16, 2020 - link
28 cores, 13k$... still not enough.FreckledTrout - Thursday, January 16, 2020 - link
AMD's EPYC 7742 which is 64 cores, supports 4TB of max memory and supports 2P(two processors) is selling for under $7K. Only time will tell but I do wonder if these costs will be enough. These larger chips I have to assume will be used for VM's / server consolidation and supporting 4TB is a huge plus and having 128 cores in 2P configuration is also a huge plus.repoman27 - Thursday, January 16, 2020 - link
“Most of these models are ‘standard’, without any letter prefix.“That would be a suffix, not a prefix, seeing as it comes after the rest of the model number.