Intel Readies Atom "Grand Ridge" 24-core Processor, Features PCIe 4.0 and DDR5

erek

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
12,115
"The "Grand Ridge" silicon is slated to be built on Intel's 7 nm HLL+ silicon fabrication node, and features 24 "Gracemont" cores across six clusters with four cores, each. Each cluster shares a 4 MB L2 cache among the four cores, while a shared L3 cache of unknown size cushions transfers between the six clusters. Intel is deploying its SCF (scalable coherent fabric) interconnect between the various components of the "Grand Ridge" SoC. Besides the six "Gracemont" clusters, the "Grand Ridge" silicon features a 2-channel DDR5 integrated memory controller, and a PCI-Express gen 4.0 root complex that puts out 16 lanes. It also features fixed function hardware that accelerates network stack processing. There are various USB and GPIO connectivity options relevant to 5G base-station setups. Given Intel's announcement of a delay in rolling out its 7 nm node, "Grand Ridge" can only be expected in 2022, if not later."

https://www.techpowerup.com/270718/...-24-core-processor-features-pcie-4-0-and-ddr5
 
Rockenrooster said:
I stopped reading at Atom...lol
Click to expand...

Agreed.

There are applications for this though. They are not MY applications, but there are applications where many small cores are useful.

I mean, this was essentially the philosophy behind Larrabee I believe.
 
What would this be for? Webserver/database/rackspace? All I can think of with high storage and network bandwidth, and somewhat high processor latency, but power-efficient.

"Please don't replace all your racks with AMD, we promise to give you a bunch of chips in a couple of years and your electric bill will drop, like, crazy, man!"
 
Axman said:
What would this be for? Webserver/database/rackspace? All I can think of with high storage and network bandwidth, and somewhat high processor latency, but power-efficient.

"Please don't replace all your racks with AMD, we promise to give you a bunch of chips in a couple of years and your electric bill will drop, like, crazy, man!"
Click to expand...

RTFA, or even just the part quoted here:

" It also features fixed function hardware that accelerates network stack processing. There are various USB and GPIO connectivity options relevant to 5G base-station setups "
 
I did, that seems like an all too specific market; either that or they’re really scraping the bottom of the barrel for headlines.
 
Axman said:
I did, that seems like an all too specific market; either that or they’re really scraping the bottom of the barrel for headlines.
Click to expand...

There're about 220k cell towers in the US, which means millions globally; all with networking gear by the rack not just a single system. That's a billion dollar market even before you factor mmWave needing zillions of new base stations.
 
Should be cool and lots of uses once it arrives. Just think what you could do with a raspberry pi powered with one of these things.

Might be overkill but by then who knows :)
 
These could be pretty awesome chips for network edge controllers, it really depends on what optimizations they have put in there for the network stack performance with only 2 channels I can't see this being a part for large servers but for small rackmount setups if this is cool enough that it can run in one of those rackmount IoT servers (1U short depth) than this would sell like hotcakes.
 
EniGmA1987 said:
Atoms can be ok for many things. I stopped reading at Intel 7nm, that will be a LONG ways away.
Click to expand...
Not so long, their 7nm line is actively making chips just not at great yields, well terrible yields actually that's why they have contracted TSMC for their 5nm line to make up the difference so they can get their DoE contract complete on time. 10nm was more a huge outlier in terms of their fabrication capabilities, they made some bad choices early into designs and they worked those choices as a requirement into a bunch of designs then later on turned out those bad choices bit them hard in the butt. Left them with 2 options, scrap the designs take the bad tech out and launch it that way, or power forward, the egos of some key people made them choose the power forward option and it set them back years. This is why Intel now says they won't be basing their designs around a specific process anymore because that is really what hurt them the most.

Anywhoo, I am looking forward to seeing what these chips could do inside a network stack, depending on how they actually perform I could see them as solid chips for SSL Decrypt mid network, putting them in my closets instead of having it all done top of rack works but increases latency by a fair bit.
 
[Spectre] said:
and the security vulnerabilities they have in them as well. Intel inside means something different these days.
Click to expand...
Not an issue so far on their Atom chips, not too worried there. I would love to see how they work compared against the Epyc 3000's though.
 
Lakados said:
Not so long, their 7nm line is actively making chips just not at great yields, well terrible yields actually that's why they have contracted TSMC for their 5nm line to make up the difference so they can get their DoE contract complete on time.
Click to expand...
Intel is not currently producing 7nm chips on their own fabs, even in small quantities. Intel has not yet gotten that node working and even just announced the other week that 7nm was being delayed till 2022 at this current time.

Intel's first 7nm product, a client CPU, is now expected to start shipping in late 2022 or early 2023, CEO Bob Swan said on a conference call Thursday. The company expects the initial production shipments of its first 7nm data center CPU design in the first half of 2023.
Click to expand...
 
EniGmA1987 said:
Intel is not currently producing 7nm chips on their own fabs, even in small quantities. Intel has not yet gotten that node working and even just announced the other week that 7nm was being delayed till 2022 at this current time.
Click to expand...
I was under the impression that they had some production under way
" “We have identified a defect mode in our 7-nanometer process that resulted in yield degradation,” said Swan. “We’ve root-caused the issue and believe there are no fundamental roadblocks, but we have also invested in contingency plans to hedge against further schedule uncertainty.”" Is Intel's response to their 7nm delay before they announced that TSMC's 5nm node would be picking up the slack. The DoE's supercomputer has been in production for a while and parts have been delivered it was just falling behind schedule which Intel has a firm date that things must be delivered by so falling behind is a non-option, that would at least indicate that they have had some parts built on their 7nm successfully to this point.
 
Remember when AMD couldn't make high performance chips and they released the e-350 platform to try and sell the "but can Intel do this with 15w"?
This reminds me of that.

Have to say I like the atom systems, own one and looking at another to build a high performance CNC plasma. Want it fanless.
 
travm said:
Remember when AMD couldn't make high performance chips and they released the e-350 platform to try and sell the "but can Intel do this with 15w"?
This reminds me of that.

Have to say I like the atom systems, own one and looking at another to build a high performance CNC plasma. Want it fanless.
Click to expand...
I have 6 whose job is to scrub out adds and redirects on websites in a way that the websites hopefully don’t detect. Since getting them in the virus/malware activity has dropped significantly and I get very few reports from users that they can’t load a site because of a detected add block. They do well but are almost always pinned on heavy traffic days. A more powerful Atom would be welcome, these things are installed in my off site racks so I needed a maximum depth of 18” so they would fit.
 
[Spectre] said:
and the security vulnerabilities they have in them as well. Intel inside means something different these days.
Click to expand...

Intel_Inside_logo-job.jpg
 
travm said:
Remember when AMD couldn't make high performance chips and they released the e-350 platform to try and sell the "but can Intel do this with 15w"?
This reminds me of that.

Have to say I like the atom systems, own one and looking at another to build a high performance CNC plasma. Want it fanless.
Click to expand...
I had one! lenovo x120e. Really like that machine. Much much faster than intel atoms at the time. Also amusingly faster than my cousin's lenovo core 2 duo's for games (simply because the gpu in the e350 was that much better than whatever the core 2 duo had).
Too bad they started stuffing e350's into super budget laptops and desktops though. Not a great image for AMD.
 
Lakados said:
I have 6 whose job is to scrub out adds and redirects on websites in a way that the websites hopefully don’t detect. Since getting them in the virus/malware activity has dropped significantly and I get very few reports from users that they can’t load a site because of a detected add block. They do well but are almost always pinned on heavy traffic days. A more powerful Atom would be welcome, these things are installed in my off site racks so I needed a maximum depth of 18” so they would fit.
Click to expand...

Lol a Raspberry pi 3 b+ can do that using pihole for an entire enterprise
 
tangoseal said:
Lol a Raspberry pi 3 b+ can do that using pihole for an entire enterprise
Click to expand...

In what universe does a raspberry pi 3 b+ have dual gigabit network adaptors? Or even 1.

You would want any good firewall to have two separate gigabit nics ...isolating your internal network from the external device and offering full gigabit speeds. (i have google fiber and gigabit is good...i'd expect any respectable business to need gigabit speeds too)

i went with an up^2 nuc-like device. It's not 24 core but does the job easily. Though, like mentioned above, this is probably more for much more massive bandwidth requiring networks than just a simple gigabit pass-thru
 
Darth Ender said:
In what universe does a raspberry pi 3 b+ have dual gigabit network adaptors? Or even 1.

You would want any good firewall to have two separate gigabit nics ...isolating your internal network from the external device and offering full gigabit speeds. (i have google fiber and gigabit is good...i'd expect any respectable business to need gigabit speeds too)

i went with an up^2 nuc-like device. It's not 24 core but does the job easily. Though, like mentioned above, this is probably more for much more massive bandwidth requiring networks than just a simple gigabit pass-thru
Click to expand...

I probably misread you, I do that A LOT. Pihole doesnt scan it just controls DNS requests keeping you safe by using listing.
It doesnt need gigabit speeds to answer simple dns queries. But it sounds like your firewallong which is another beast.

I used to run PfSense on a Sandy Bridge Core I3 2100 and it ran like a dream. I used Pf Blocker and other AV plugins. Had dual intek 1GiB nics as well

Now I just use UBNT Dream Machine Pro with IPS enabled. It catches quite a bit. For AV I stopped using network wide for home. I just use malwarebytes subscription on my machines.

For corporate stuff I used to use all Cisco stuff.
 
Last edited:
i usually associate pi-hole with a firewall, but i guess you could just rock it as the lan's dns box.
 
tangoseal said:
Lol a Raspberry pi 3 b+ can do that using pihole for an entire enterprise
Click to expand...
They also do live image filtering which is really what eats them, they scan any pages as they come in and if the image content is determined to be pornographic it replaces the image with a cuddly kitten. Mostly it just filters out bathing suits and stuff because the actual porn is blocked but it also blocks out the "medical" images that would come up if somebody searched "Gross butt farts boobies". But I have considered getting some cheap 1U racks and building Pi4's into them to take over the add blocking.
 
Eh, it would probably be great in a low to mid budget Chromebook.
 
Rockenrooster said:
I stopped reading at Atom...lol
Click to expand...
Then it sounds to me, like you don't own a NAS ........... cause this hardware sounds like fun to me.
I have an Atom based NAS and it's been wonderful but the processor inside it was first designed in 2016, it'd be nice to see some updates to this tech.

I'd especially like to see the AMD v3000 or Epyc 3004 series, compete in this space.
I just want an ITX board with 8+ SATA ports, ECC memory, IPMI, 45W Max and a not crap CPU for under $650 US. Not that I need to replace mine but it'd be nice.
 
Axman said:
What would this be for? Webserver/database/rackspace? All I can think of with high storage and network bandwidth, and somewhat high processor latency, but power-efficient.

"Please don't replace all your racks with AMD, we promise to give you a bunch of chips in a couple of years and your electric bill will drop, like, crazy, man!"
Click to expand...
Only 16 lanes limits connectivity. Looks like they're targeting cell tower base stations; could probably do some other network edge style work where space and power are limited. Maybe load balancers / TLS proxy (if encryption is accelerated enough). Could be a webserver if you go for lots of small servers. IMHO not a great fit for database; you usually want fewer, but more powerful database boxes because cross-system database work is a drag. Rackspace is going to depend on how this looks overall, and the customers... Clearly 'core' based cores are going to be faster and better individually, so you'd need a lot more nodes if you replaced your Xeons with this, but if $/compute and W/compute are better, some organizations are going to be happy to save money and power. Could be somewhat useful for cloud hosters: you might be more likely to rent a whole node to a single customer and avoid the security fun of having multiple customers on the same hardware.

But all this is speculation until they actually release something we can look at.
 
wyqtor said:
How do Atom cores compare to the E cores of Alder lake?
Click to expand...
From Intel’s most recent Architecture Day:

“The smaller core as part of Intel’s hybrid Alder Lake design is called an E-core, and is built on the Gracemont microarchitecture. It forms part of Intel’s Atom family of processors, and is a significant microarchitectural jump over the previous Atom core design called Tremont.”

So the E cores are their new Atom cores.
 
AbRASiON said:
Then it sounds to me, like you don't own a NAS ........... cause this hardware sounds like fun to me.
I have an Atom based NAS and it's been wonderful but the processor inside it was first designed in 2016, it'd be nice to see some updates to this tech.

I'd especially like to see the AMD v3000 or Epyc 3004 series, compete in this space.
I just want an ITX board with 8+ SATA ports, ECC memory, IPMI, 45W Max and a not crap CPU for under $650 US. Not that I need to replace mine but it'd be nice.
Click to expand...

For a dedicated NAS I think it could be great. I mean, 24 cores may be a bit overkill, but filesystem loads generally spread quite nicely over many cores, and don't really need big core performance.

That said, my NAS is also my VM server which hosts several VM's and containers for me and for those applications I think I'd like big cores.
toast0 said:
Only 16 lanes limits connectivity. Looks like they're targeting cell tower base stations; could probably do some other network edge style work where space and power are limited.
Click to expand...

Yeah, that seems like a serious limitation.

On the one hand, who knows, these modern Atom cores may be as fast or faster than my old dual Ivy Bridge era Xeon E5-2650v2's, but on the other hand I am using quite a lot of PCIe lanes on that thing. 8x lanes for the SAS controller, 8x lanes for the dual 10gig ethernet, 56x lanes on the 14 separate m.2 NVMe drives, and that's just the ones I've manually added and doesn't include the on board devices on the Supermicro X9DRI-F.

As small cores get more and more powerful, I imagine they will eventually be more than sufficient for my NAS/VM/Container needs, but I don't think I'll ever get away from wanting massive quantities of PCIe lanes.
 
Weird chip design tiled chip layouts are inefficient for general workloads. (Memory scheduling becomes a problem) and these seem too weak for hpc stuff.

I run a bunch of the phi 7210s (64 core 256thread @1.3ghz on 14nm) and they are VERY specific as far as workload.

Additionally the cache kinda rings alarms. The phis have 16gb of mcdram on chip they use as a shared L3. It seems weird nothing about on chip memory is marketed.

This looks like old news. Perhaps just a old false rumor?
 
You must log in or register to reply here.
Back
Top