Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
Klyith
Aug 3, 2007

GBS Pledge Week

Cygni posted:

Also... what in the world is with IO die having its own 2 crestmont cores?

They are there to run the all new next-generation RGB effects on connected devices.

(Real answer: ultra-low-power efficiency cores.)

Adbot
ADBOT LOVES YOU

FuturePastNow
May 19, 2014


Will these e-cores actually be efficient or will they also be juiced to the highest voltage they can handle for a 1% higher Cinebench score?

Cygni
Nov 12, 2005

raring to post

FuturePastNow posted:

Will these e-cores actually be efficient or will they also be juiced to the highest voltage they can handle for a 1% higher Cinebench score?

Its mobile only, so prolly not. The Redwood cores tho...

borkencode
Nov 10, 2004
gently caress Everything, We're Doing Five Blades Caches

SwissArmyDruid
Feb 14, 2014

by sebmojo

borkencode posted:

gently caress Everything, We're Doing Five Blades Caches

is that literally not what the point of optane was?

Anime Schoolgirl
Nov 28, 2002

RIP Optane, killed by falling registered ram prices and dumping memory into fast sequential NAND

BlankSystemDaemon
Mar 13, 2009



SwissArmyDruid posted:

is that literally not what the point of optane was?
Nah, Optane wouldn't have achieved its final form until operating systems had been modified to no longer be installed onto a disk and then use the main memory via a VM subsystem.

Anime Schoolgirl posted:

RIP Optane, killed by falling registered ram prices and dumping memory into fast sequential NAND
Optane wasn't just killed by that - one of the more fundamental issues with it was that even with all the development put into it, the write endurance of Optane DIMMs was still many times less than that of non-volatile memory DIMMs.
So the above notion was simply out of reach for the foreseeable future.

gradenko_2000
Oct 5, 2010

HELL SERPENT
Lipstick Apathy

borkencode posted:

gently caress Everything, We're Doing Five Blades Caches

cache rules everything around me

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

BlankSystemDaemon posted:

Nah, Optane wouldn't have achieved its final form until operating systems had been modified to no longer be installed onto a disk and then use the main memory via a VM subsystem.

reject modernity, return to javacard :q:

https://www.youtube.com/watch?v=31D94QOo2gY&t=608s

https://www.youtube.com/watch?v=khgT5dwKvOo

Paul MaudDib fucked around with this message at 16:31 on Apr 24, 2023

WhyteRyce
Dec 30, 2001

Optane was killed by CXL. Once customers had a path to stick commodity DRAM on an non proprietary or platform locked bus to expand memory and get acceptable performance then it was basically over for Optane. The entire endeavor was only financed so that Intel could keep platform locking people into Xeons

Yes, you can put Optane on CXL but why would intel care about that

priznat
Jul 7, 2009

Let's get drunk and kiss each other all night.
^^ what timing! haha

CXL 3.0 memory pooling solution with multiple hosts sharing memory appliances :getin:

Perplx
Jun 26, 2004


Best viewed on Orgasma Plasma
Lipstick Apathy
Optane is still the best for qd1 iops, but I guess that's some nerd stat that normal people don't care about.

WhyteRyce
Dec 30, 2001

Perplx posted:

Optane is still the best for qd1 iops, but I guess that's some nerd stat that normal people don't care about.

There are very much power users/companies that actually do low QD workloads but how much do you want to pay for it vs. just buying some NAND SSDs for like a tenth the cost per GB and calling it a day.

BlankSystemDaemon
Mar 13, 2009



Perplx posted:

Optane is still the best for qd1 iops, but I guess that's some nerd stat that normal people don't care about.
That's the SSDs, not the DIMMs.

phongn
Oct 21, 2006

WhyteRyce posted:

There are very much power users/companies that actually do low QD workloads but how much do you want to pay for it vs. just buying some NAND SSDs for like a tenth the cost per GB and calling it a day.

My ZFS SLOG :(

hobbesmaster
Jan 28, 2008

Intel made Bitcoin ASICs?

Rexxed
May 1, 2010

Dis is amazing!
I gotta try dis!


Looks like it:
https://en.wikipedia.org/wiki/Blockscale

in a well actually
Jan 26, 2011

dude, you gotta end it on the rhyme

I refuse to put this in my search history but how does that compare to industry standard miner performance?

Rexxed
May 1, 2010

Dis is amazing!
I gotta try dis!

I don't really know about comparisons but from the intel press release that was covered by Tom's Hardware:
https://www.tomshardware.com/news/intels-blockscale-cryptominer-ships-ahead-of-schedule-late-to-the-party

quote:

Going by the numbers, this ASIC can achieve a hash rate of 580 GH/s, with power consumption between 4.8 and 22.7W, and power efficiency of 26 J/TH. The flip chip LGA package size is a very modest 7.5 x 7mm. There are on-chip sensors for both temperature and voltage to help monitor systems and tune performance.

edit: can't find it on the list of miners in the nicehash profitability calculator, seems they were sold to companies making mining asic hardware as standalone units.

from: https://www.newsbtc.com/news/bitcoin/intel-ships-blockscale-asic-chips-to-selected-bitcoin-mining-companies/

quote:

Intel AXG Custom Compute team is now shipping the Blockscale ASIC! First product will always be unforgettable, congratulations team?
Excited to see how @ArgoBlockchain @griid and @HiveBlockchain improvise around Blockscale and our open design. pic.twitter.com/0rxtNTLMfw

— Raja Koduri (Bali Makaradhwaja) (@RajaXg) June 29, 2022

So I guess things made by ArgoBlockchain, griiid, and HiveBlockchain used them.

Rexxed fucked around with this message at 23:26 on Apr 24, 2023

ijyt
Apr 10, 2012

Seems to have launched a fair few years too late to matter.

wet_goods
Jun 21, 2004

I'M BAAD!
That was definitely an all time low level of debasement for them, I remember being so god drat embarrassed when the headlines came out that they were making that garbage

KKKLIP ART
Sep 3, 2004

Is there a new embedded fun server board like the Atom 3xxx series that intel makes that could be good for a NAS setup? I haven’t looked at new hardware anytime recently.

mobby_6kl
Aug 9, 2009

by Fluffdaddy

KKKLIP ART posted:

Is there a new embedded fun server board like the Atom 3xxx series that intel makes that could be good for a NAS setup? I haven’t looked at new hardware anytime recently.
The Alder Lake-N family has the x7425E but no idea if it's available in any server boards yet. I've seen Tremont-based boards though on Aliexpress:


https://www.aliexpress.com/item/1005004757954007.html

mobby_6kl fucked around with this message at 14:16 on Apr 25, 2023

hobbesmaster
Jan 28, 2008

The newest Denverton type SoC replacement is Elkhart lake (atom x6000) if that’s what you’re asking.

They’ve been very slow to make it into sales channels and onto interesting boards you can buy though.

KKKLIP ART
Sep 3, 2004

mobby_6kl posted:

The Alder Lake-N family has the x7425E but no idea if it's available in any server boards yet. I've seen Tremont-based boards though on Aliexpress:


https://www.aliexpress.com/item/1005004757954007.html

I was looking at these but apparently there are issues with the 225/226 NICs and not a lot of documentation on folks actually using them. That was my top pick but until I know the NIC will work well, I don’t want to drop the money.

E: Brian Moses apparently made one from the N5105 motherboard. Don’t see why the N6005 boards won’t work. He used TrueNAS Scale which seems to have more robust hardware support.

KKKLIP ART fucked around with this message at 15:35 on Apr 25, 2023

BlankSystemDaemon
Mar 13, 2009



hobbesmaster posted:

The newest Denverton type SoC replacement is Elkhart lake (atom x6000) if that’s what you’re asking.

They’ve been very slow to make it into sales channels and onto interesting boards you can buy though.
The Atom x6000 series don't do ECC.

hobbesmaster
Jan 28, 2008

BlankSystemDaemon posted:

The Atom x6000 series don't do ECC.

It does in band, but yeah. The other low power, low cost ones don’t do ECC either unless I’m forgetting or missing something. Intel’s non i series and “big xeon” naming makes absolutely no sense even by Intel’s standards

Cygni
Nov 12, 2005

raring to post

BlankSystemDaemon
Mar 13, 2009



hobbesmaster posted:

It does in band, but yeah. The other low power, low cost ones don’t do ECC either unless I’m forgetting or missing something. Intel’s non i series and “big xeon” naming makes absolutely no sense even by Intel’s standards
The idea of ECC where you don't receive at least an NMI on correctable errors is just an entire waste of money.

The Denverton SoCs had ECC memory support.

SwissArmyDruid
Feb 14, 2014

by sebmojo
I am still traumatized by circa-2016 Atoms committing suicide, (in my case, MY loving NAS) are we finally safe from that?

Lowen SoDium
Jun 5, 2003

Highen Fiber
Clapping Larry

SwissArmyDruid posted:

I am still traumatized by circa-2016 Atoms committing suicide, (in my case, MY loving NAS) are we finally safe from that?

Your NAS and 120 Cisco ASAs I had just finished deploying over the previous 6 months.

SwissArmyDruid
Feb 14, 2014

by sebmojo

Lowen SoDium posted:

Your NAS and 120 Cisco ASAs I had just finished deploying over the previous 6 months.

Jesus gently caress, do they have a portal to the Plane Of Suiciding Silicon that they can't close and are just trying to mitigate or something?

hobbesmaster
Jan 28, 2008

SwissArmyDruid posted:

Jesus gently caress, do they have a portal to the Plane Of Suiciding Silicon that they can't close and are just trying to mitigate or something?

Ah, that’s why there were so many 14nm updates.

Kazinsal
Dec 13, 2011

SwissArmyDruid posted:

Jesus gently caress, do they have a portal to the Plane Of Suiciding Silicon that they can't close and are just trying to mitigate or something?

In the case of Cisco ASAs, it was a pretty significant list of branch firewall devices that were all sold all the way up to August 2021: https://www.cisco.com/c/en/us/support/docs/field-notices/642/fn64228.html

Supposedly the post-2019 revisions are free of this issue.

Lowen SoDium
Jun 5, 2003

Highen Fiber
Clapping Larry

Kazinsal posted:

In the case of Cisco ASAs, it was a pretty significant list of branch firewall devices that were all sold all the way up to August 2021: https://www.cisco.com/c/en/us/support/docs/field-notices/642/fn64228.html

Supposedly the post-2019 revisions are free of this issue.

That is what pissed me off the most: Cisco knew about the flaw before it was recalled and sold that poo poo to us anyways.

New Zealand can eat me
Aug 29, 2008

:matters:


hobbesmaster posted:

72W for a ridiculous overclocking CPU seems quaint but think of that like 72 W for a single core

:hellyeah: That's drat near a 10x multiple over contemporary 'flagship' CPUs, I don't even use twice as much for 16 cores at 4.5ghz.

My dreams of doing a custom waterloop sous vide system are dying more and more every day. Can't even keep this gen's GPU over 60C if its actually using the fans. Hopefully nvme storage keeps getting hotter

Potato Salad
Oct 23, 2014

nobody cares


priznat posted:

^^ what timing! haha

CXL 3.0 memory pooling solution with multiple hosts sharing memory appliances :getin:

we're rapidly normalizing capabilities previously reserved for highly specialized HPC systems

Cygni
Nov 12, 2005

raring to post

$1.5B loss for the quarter from Intel. From the financials:

Intel 4 + Meteor Lake (mobile only at first?) still a 2nd half 2023 launch. CPU tile is Intel 4, GPU, SoC, and IO are TSMC 6. Future nodes "on track":

quote:

Intel continues to be on track to meet its goal of achieving five nodes in four years, with two of the five
nodes nearly complete. Intel 7 is in high-volume manufacturing and CCG's Meteor Lake product on Intel 4
is ramping production wafer starts for an expected launch in the second half of 2023. Intel 3, Intel 20A, and
Intel 18A remain on track.

Emerald Forest (P core) this year, with Sierra Forest (144 E core) in first half of 2024.

quote:

DCAI also announced it expects to deliver Intel’s 5th Gen Xeon Scalable processor, Emerald Rapids, later
this year. In addition, the business narrowed the delivery window for Sierra Forest, which is expected to ship
to customers in the first half of 2024, with Granite Rapids expected to follow shortly thereafter. Clearwater
Forest, the follow-on to Sierra Forest, is expected to ship in 2025, and will be manufactured on Intel 18A,
the node designed to achieve process leadership and representing the culmination of the company’s five nodes-in-four-years strategy.

Projecting another loss in Q3, no mention of Arc in the financials at all.

SwissArmyDruid
Feb 14, 2014

by sebmojo

Kazinsal posted:

In the case of Cisco ASAs, it was a pretty significant list of branch firewall devices that were all sold all the way up to August 2021: https://www.cisco.com/c/en/us/support/docs/field-notices/642/fn64228.html

Supposedly the post-2019 revisions are free of this issue.

okay, but seriously, though, what's the goldilocks zone for atoms that I can use for an OPNsense box, but won't either explode in 1.5-3 years or have i225/i226 issues? Or just wait? (because I just got done raging at trying to diagnose intermittent internet problems that turned out to be said i225/i226 issues at work, and was barking up the wrong tree as our modem being bunk for THE LONGEST TIME.)

Adbot
ADBOT LOVES YOU

Twerk from Home
Jan 17, 2009

This avatar brought to you by the 'save our dead gay forums' foundation.

SwissArmyDruid posted:

okay, but seriously, though, what's the goldilocks zone for atoms that I can use for an OPNsense box, but won't either explode in 1.5-3 years or have i225/i226 issues? Or just wait? (because I just got done raging at trying to diagnose intermittent internet problems that turned out to be said i225/i226 issues at work, and was barking up the wrong tree as our modem being bunk for THE LONGEST TIME.)

If you want one more datapoint about which Atoms are usable, you need to be aware that the entire N5095 - N6005 Jasper Lake family has some microcode-related kernel panics in guest VMs that resulted in 33 pages of comedy as people reproduce it on every kernel under the sun and hardware from 20 different manufacturers: https://forum.proxmox.com/threads/vm-freezes-irregularly.111494.

I gave up and used mine as a physical box without any virtualization because of this.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply