This channel is really starting to become my favorite tech channel. Not just reading benchmarks and drooling over the core count, but explaining the why and how. Giving perspective on the industry. Keep up the good work!
Please buy it, you can give me remote access then. Would be a shame if it is unused😉. You can also have all the profits, unfortunately i m doing non-profit research😂.
"It's something that's a little different than what you probably have deployed in your data centers today." I have a Dell R510 in my server rack so... yeah. It is a bit different.
So excited for the ARM revolution, can't wait to see how Nuvia Phoenix stacks up to these Neoverse M1 cores For the time being, the Ampere Altra is an actually shipping product and doesn't dissapoint at all. Exciting times
The Ampre Altra is good. The problem I see is that it's not so much better than Rome that it would justify a big transition to it. Unless the Ampere Altra is much cheaper than Rome but that's unlikely since it's a monolithic die (and a huge one at that) which results in higher fabrication cost (due to lower yields) than smaller chiplets like those from AMD. AMD went MCM specifically to increase yield and lower fabrication cost. And as you add more cores, the need for MCM becomes even more important to keep cost down. That's the weakness of all these ARM cpus.
@@1idd0kun Might actually be cheaper than Rome though, interposers are pretty expensive and traditionally RISC ISA chips have a reduced transistor count over CISC, but I'm not sure how an advanced μarch like Neoverse fits in to that tradition. Might just be a wash in the end? I think the primary attraction right now would be as a development platform for future processors that will be faster than competing processors from Intel/AMD like the aforementioned Nuvia Phoenix cores (+40% IPC over Rome at 33% of the TDP) Not sure exactly what's going on behind the scenes at the hyperscalers, as Amazon and now apparently Microsoft are betting big on ARM platforms right now already and they know more than we do.. At any rate exciting times.
@@hugevibez Rome doesn't use any interposer. It uses wired connection over substrate which is really really cheap. I do agree the main draw for Altra and similar ARM server CPUs is as a development platform, but for cloud gaming and mobile apps. I don't think anyone is gonna develop for future CPUs when no one really knows when (or even if) those CPUs are gonna be released. Those performance claims from Nuvia Phoenix look impressive, but they're just simulations. They don't even have a prototype yet. You won't see that CPU earlier than 2022 (maybe 2023) and at that point 40% over Rome won't be that impressive at all considering Milan is sampling right now and offering 20% over Rome.
@@1idd0kunYou're right, Rome isn't on an interposer, that's reserved for 3D stacking. Mixed things up here. That is coming to next generation AMD products however. What I said still stands though, and according to Anandtech the Q80-33 is priced at $4k, that's about half of what an equivalent Rome chip costs. Obviously hyperscalers don't pay MSRP, but it's indicative of the difference in pricing. I think the jump to ARM isn't as big as you imagine, a lot of the software we run today is ready for ARM already and everything else is moving towards accelerators anyway. The question is not if, but when. At least according to both Microsoft and Amazon.
@@hugevibez " The question is not if, but when." I think the question is why. Because, to be honest, I don't see any big transition to ARM happening if ARM can't offer something x86 can not. As it stands, ARM server CPUs can't beat x86 in raw performance for a significant margin, and even the efficiency lead over x86 is smaller than expected. Cost is also not cheaper enough, at least for hyperscalers who pay only a fraction of the list pricing anyway. What I see happening is ARM making a niche for itself as a platform for developing in the cloud, and maybe in the super computer market as well. But I don't think it will ever replace x86 in the dominant CPU in the data center. Not unless it can offer much more than what x86 can and I don't see that happening any time soon. In fact, if you look at market share numbers, AMD is gaining sever market share much faster than ARM.
Do you have any power consumption figures? How does power efficiency compare to the AMD Epyc Rome chips? Looking for something more specific than “this uses about 1KW at full bore”. I’d like to see some kind of scaled value to highlight the power efficiency like how many seconds of some task or benchmark vs how much power it used on that task.
LLVM is the future of compilers. I'd like to see those numbers as much as GCC/ICC/AOCC (especially since a lot of optimizations are contributed to the various LLVM optimization targets).
@@johannweber5185 you're right, I'd kinda assumed it was yet another proprietary fork of gcc. That said, the mainline should still be the standard I think since all of the surrounding ecosystem consumes it (like mainline Clang, Apple's stuff, the Rust compiler, Julia, some of the machine learning tools, etc.).
@@JMurph2015 Due to the GPL-Licence of gcc it is not possible to publish a non-GPL proprietary fork of GCC. This might be the reason why companies like to base their compilers on LLVM, which has a more liberal licence. Of course this does not mean that companies like AMD do not additionally contribute to to mainline LLVM and to GCC.
@@johannweber5185 I suppose I misspoke about "proprietary" when I really meant "corporate". Edit: also sorry to keep backtracking, lol, I distinctly remember using some custom versions of GCC for embedded microcontrollers and tend to associate "vendor compilers" with those.
Buy a macchiatobin while they're still available... but there are tons of arm boards out there, they're just not easy to find... mostly because arm is such an obscure and unique platform, every implementation is different.
RISC V servers will be where it is at I think as they can be finely tuned for specific workloads. They are also open with no royalties to pay to NVidia :)
I wonder when the lower cost new/used arm servers will start hitting the market. I would love to test them out as lower power draw VM and file servers at home.
ThunderX was not exactly low power draw. ThunderX2 was better but was designed for HPC. Ampere Altra is probably the first chip you would want to use for something like this. Given where we are in the macro cycle, these will come out of data centers in 2024-2026 after 3-5 years and by then we will have CXL switching and this type of architecture will seem extremely dated.
@@ServeTheHomeVideo dated is fine for most home use when you can not justify spending current prices for servers. My VM server is a Hyve Zeus server. Has a pair of 2650v2 chips and 64gb of RAM. Works well for my needs. But, in a few years as these servers start getting retired, they may be worth a look. I make a pretty good living, but all my hobbies are expensive. Lol
I wonder how Nuvia comes into the picture here. They seem to be after markets just like this and they have a outstanding core if thier briefing is to be believed
Perhaps the biggest difference is that this is shipping today to cloud customers. We will see Nuvia in time but for now it is comparing a product of today with a product of the future.
The subtitles are all messed up on this video. No English subtitles are available, not even autogenerated. The only subtitles available are listed as "autogenerated Vietnamese," but even that is producing seemingly random words from random languages.
Ah yes, Oracle linux. I was wondering about that at the start of this video. Surprisingly awesome OS.. only major downside is the lack of an in-place upgrade to the next major release, but that's a once-every-9-years issue.
Cool story on that one. It is actually due to the very tight manufacturing tolerances that Wiwynn has! Discussed that on p.2 of the main site article www.servethehome.com/ampere-altra-wiwynn-mt-jade-server-review-the-most-significant-arm-server/2/
No. I see on smaller screens it is harder to read. We kept the order the same to make it a bit easier. In the video we referred folks to the main site for more on the charts. Good point though. When we make big charts, people complain they are too big.
Which server do you recommend for a Magento 2 store with more than 1 million monthly visitors, hosting more than 10 million digital products? AMD EPYC 7502P or AMD 5950x or Altra Q80-30?
Great video and very informative! The ARM server solution has really made a lot of progress. Back in the ThnuderX1 days it seemed clear to me that Intel left a wide open hole for ARM to walk right in, and they sure did move in. I figured that the arrival of EPYC would greatly slow down ARMs progression into x86 server territory, Intel should be thanking AMD. Hopefully you can review a ThunderX 3 server one day. I do run some cloud instances, X86 so far, maybe one day it'll be ARM instead, I have no reason to use x86 other than for the tool chain.
Guess I should start learning Linux or pray Microsoft can get their codebase migrated to ARM. Wonder if the age of x86 is coming to a close after all; I've remained skeptical for a while that it would happen. Apple's M1 didn't impress me because of the soldered, locked down nature of the kit. This server on the other hand is amazing.
Wiwynn is more designed to deliver to hyper scale. Supermicro is designed for enterprise and channel sales. Somewhat different market segments they are playing in
So could someone tell me like what kind of uses would someone have for an ARM server? Is there even much out there in the way of software packages for ARM other Than PUBG mobile?
I am going out on a limb here and guessing the "last video" was the one from this summer on Power 10. Conceptually IBM is making a fabric in their CPUs that can span not just within a chassis, but to multiple chassis. As they scale out, they get more nodes that they can attach memory to. Also, with their OMI, they get more flexibility with what kinds of memory they can use, and how much can be attached per node, similar to what the rest of the market will see in the CXL or CXL 2.x generations.
@@ServeTheHomeVideo is that even possible with x86 because HP SuperDome Flex shows us that multiple nodes can be used to a total of 32 xeon processors and upto 48TB RAM, can this grow with x86? (a lot of ram and possibly even more processors)
@@hariranormal5584 sgi has this tech called Shared memory. Made programming easier since you can address all the memory. But if the memory reside on the other node. Big performance hit.
One of the big use cases for Arm servers is gaming, but a bit differently. One can run mobile games on these servers with native Arm and then stream to mobile devices.
They are awesome COU coolers and actually show the tight tolerances Wiwynn manufacturers to. Much tighter than Dell/ HPE. Put a bit around the bent find on p2 of the main site article.
It's interesting but this and all servers which are generally available are basically glorified PCs. I'm interested in what Oxide will produce in the next year or two as we've been stuck with these glorified PCs since Sun and boy do I miss Sun boxes.
I can’t wait for mips and arm to take over followed by risc-v opensparc .. X86 is garbage compared to the design of these. With exception of ppw instructions... but the cost is obvious.. we are pretty much hitting a wall with current tech
got my hand on one of those. they are JUNK, most of thins are cheaply made, you could press the power button and break it, firmware also junk, it's going to switch front VGA and back side VGA without anything change, all, both VGA ports are not working, the only way to access is with BMC, but BMC has ton of bugs too... save your headache... don't deal with this shit.
From my own experience with the Ampere platforms, Mt. Snow from Gigabyte is by far the most stable platform. The only caveat is it only has a single socket SKU. I will admit though that the power button on it also sucks. It's not as bad as Mt. Jade which can easily get lodged inside as the Mt. Snow has a membrane power button. However it can also get pushed in and get stuck. That being said though these units aren't really meant to be powered on and off that frequently. They're intended to be in a datacenter especially with how effing loud they get.
This channel is really starting to become my favorite tech channel. Not just reading benchmarks and drooling over the core count, but explaining the why and how. Giving perspective on the industry. Keep up the good work!
Glad you enjoy it! Thanks for the kind words.
This channel has a lot in common with Level1Techs. Definitely going here when I want to build a NAS or some kind of server.
Great overview of this server. Thanks for taking the time to put this video together and share it with us!
It's awesome how much movement there has come to the market again after basically over a decade of Xeon.
I want one of these servers. I have no need for one, but I want one.
SAME XD
Exactly the same .. it is just weirdly attractive to tinker with it.
Ha ha same😁
Please buy it, you can give me remote access then. Would be a shame if it is unused😉. You can also have all the profits, unfortunately i m doing non-profit research😂.
Claudius Smith
Your not getting a server begging for people like this. Mark my words
Man, I don't give a rat's ass about server stuff, but the sheer enthousiam you have for your job and tech is beyong inspiring. Keep going !
So disappointed when "were giving this wywin server" was followed by "our editor's choice award" rather than "away"! ;)
Loooool!
┻┳|
┳┻| _
┻┳| •.•)
┳┻|⊂ﾉ
┻┳|
yes and I would have many uses for it
Haha same, i thought they giving it out, shocked
I know I started getting so excited. The things I would do with that server.
I have no idea what I'd do with one of these, my parents would probably kill me if I brought this into our house.
i love the b roll in this video. putting those new cameras to good use!
Another Episode of "HOW MANY Cores?!"
I like
- Very many cores?
- YES!
Hey guys, this is super excited Patrick!
Sadly, this is super sleep deprived Patrick. I hope you are well Jon. Must be nice to see the fruits of your labor.
Isn’t it great to see Arm actually taking over? Judgement Day is inevitable 🤣
"It's something that's a little different than what you probably have deployed in your data centers today."
I have a Dell R510 in my server rack so... yeah. It is a bit different.
So excited for the ARM revolution, can't wait to see how Nuvia Phoenix stacks up to these Neoverse M1 cores
For the time being, the Ampere Altra is an actually shipping product and doesn't dissapoint at all. Exciting times
The Ampre Altra is good. The problem I see is that it's not so much better than Rome that it would justify a big transition to it. Unless the Ampere Altra is much cheaper than Rome but that's unlikely since it's a monolithic die (and a huge one at that) which results in higher fabrication cost (due to lower yields) than smaller chiplets like those from AMD. AMD went MCM specifically to increase yield and lower fabrication cost. And as you add more cores, the need for MCM becomes even more important to keep cost down. That's the weakness of all these ARM cpus.
@@1idd0kun Might actually be cheaper than Rome though, interposers are pretty expensive and traditionally RISC ISA chips have a reduced transistor count over CISC, but I'm not sure how an advanced μarch like Neoverse fits in to that tradition. Might just be a wash in the end?
I think the primary attraction right now would be as a development platform for future processors that will be faster than competing processors from Intel/AMD like the aforementioned Nuvia Phoenix cores (+40% IPC over Rome at 33% of the TDP)
Not sure exactly what's going on behind the scenes at the hyperscalers, as Amazon and now apparently Microsoft are betting big on ARM platforms right now already and they know more than we do..
At any rate exciting times.
@@hugevibez
Rome doesn't use any interposer. It uses wired connection over substrate which is really really cheap.
I do agree the main draw for Altra and similar ARM server CPUs is as a development platform, but for cloud gaming and mobile apps. I don't think anyone is gonna develop for future CPUs when no one really knows when (or even if) those CPUs are gonna be released. Those performance claims from Nuvia Phoenix look impressive, but they're just simulations. They don't even have a prototype yet. You won't see that CPU earlier than 2022 (maybe 2023) and at that point 40% over Rome won't be that impressive at all considering Milan is sampling right now and offering 20% over Rome.
@@1idd0kunYou're right, Rome isn't on an interposer, that's reserved for 3D stacking. Mixed things up here. That is coming to next generation AMD products however.
What I said still stands though, and according to Anandtech the Q80-33 is priced at $4k, that's about half of what an equivalent Rome chip costs. Obviously hyperscalers don't pay MSRP, but it's indicative of the difference in pricing.
I think the jump to ARM isn't as big as you imagine, a lot of the software we run today is ready for ARM already and everything else is moving towards accelerators anyway. The question is not if, but when. At least according to both Microsoft and Amazon.
@@hugevibez " The question is not if, but when."
I think the question is why. Because, to be honest, I don't see any big transition to ARM happening if ARM can't offer something x86 can not. As it stands, ARM server CPUs can't beat x86 in raw performance for a significant margin, and even the efficiency lead over x86 is smaller than expected. Cost is also not cheaper enough, at least for hyperscalers who pay only a fraction of the list pricing anyway. What I see happening is ARM making a niche for itself as a platform for developing in the cloud, and maybe in the super computer market as well. But I don't think it will ever replace x86 in the dominant CPU in the data center. Not unless it can offer much more than what x86 can and I don't see that happening any time soon. In fact, if you look at market share numbers, AMD is gaining sever market share much faster than ARM.
Throwback to the mips /alpha days ...
that's a distilled, detailed awesome review.
Do you have any power consumption figures? How does power efficiency compare to the AMD Epyc Rome chips?
Looking for something more specific than “this uses about 1KW at full bore”. I’d like to see some kind of scaled value to highlight the power efficiency like how many seconds of some task or benchmark vs how much power it used on that task.
Java HotSpot numbers would be interesting as well.
25:51 ~2020~ 2021 presumably?
Came to ask the same thing!
LLVM is the future of compilers. I'd like to see those numbers as much as GCC/ICC/AOCC (especially since a lot of optimizations are contributed to the various LLVM optimization targets).
As far as I know aocc is LLVM-based.
@@johannweber5185 you're right, I'd kinda assumed it was yet another proprietary fork of gcc. That said, the mainline should still be the standard I think since all of the surrounding ecosystem consumes it (like mainline Clang, Apple's stuff, the Rust compiler, Julia, some of the machine learning tools, etc.).
@@JMurph2015 Due to the GPL-Licence of gcc it is not possible to publish a non-GPL proprietary fork of GCC. This might be the reason why companies like to base their compilers on LLVM, which has a more liberal licence. Of course this does not mean that companies like AMD do not additionally contribute to to mainline LLVM and to GCC.
@@johannweber5185 I suppose I misspoke about "proprietary" when I really meant "corporate".
Edit: also sorry to keep backtracking, lol, I distinctly remember using some custom versions of GCC for embedded microcontrollers and tend to associate "vendor compilers" with those.
Will ICC even run on arm? Can't say I ever tried.
When first glancing at the thumbnail, I thought the cpu coolers were office buildings in a parking lot
This is great, but where are the developer platforms for ARM? I'm afraid that a Raspberry Pi just don't fit the bill.
Buy a macchiatobin while they're still available...
but there are tons of arm boards out there, they're just not easy to find... mostly because arm is such an obscure and unique platform, every implementation is different.
Benchmarks against m6gd.metal please!
Please :D
At about 3/4 through my brainwidth was saturated. I started to hallucinate and think, why don't you have a heatsink hat on?
In the "Table of Contents" the Wrap-up is at 27:75. I guess that's wrong.
Dyslexia strikes. 75->57 thanks!
RISC V servers will be where it is at I think as they can be finely tuned for specific workloads. They are also open with no royalties to pay to NVidia :)
Between Apple M1 on the desktop and this on the server side ARM is really moving now.
I wonder when the lower cost new/used arm servers will start hitting the market. I would love to test them out as lower power draw VM and file servers at home.
ThunderX was not exactly low power draw. ThunderX2 was better but was designed for HPC. Ampere Altra is probably the first chip you would want to use for something like this. Given where we are in the macro cycle, these will come out of data centers in 2024-2026 after 3-5 years and by then we will have CXL switching and this type of architecture will seem extremely dated.
@@ServeTheHomeVideo dated is fine for most home use when you can not justify spending current prices for servers. My VM server is a Hyve Zeus server. Has a pair of 2650v2 chips and 64gb of RAM. Works well for my needs. But, in a few years as these servers start getting retired, they may be worth a look.
I make a pretty good living, but all my hobbies are expensive. Lol
I wonder how Nuvia comes into the picture here. They seem to be after markets just like this and they have a outstanding core if thier briefing is to be believed
Perhaps the biggest difference is that this is shipping today to cloud customers. We will see Nuvia in time but for now it is comparing a product of today with a product of the future.
Hi can u make the charts bigger, it's difficult to read when watching on mobile phones
How much would that be?Is it openly available?
The subtitles are all messed up on this video. No English subtitles are available, not even autogenerated. The only subtitles available are listed as "autogenerated Vietnamese," but even that is producing seemingly random words from random languages.
Ah yes, Oracle linux. I was wondering about that at the start of this video. Surprisingly awesome OS.. only major downside is the lack of an in-place upgrade to the next major release, but that's a once-every-9-years issue.
7:20 am i the only weirdo, who got triggered by the heatsink beeing sloped?
Cool story on that one. It is actually due to the very tight manufacturing tolerances that Wiwynn has! Discussed that on p.2 of the main site article www.servethehome.com/ampere-altra-wiwynn-mt-jade-server-review-the-most-significant-arm-server/2/
@@ServeTheHomeVideo at first i thought, it would have gotten damaged in shipping or something ;)
The writing on the graphics are not readable. Was that intentional ?
No. I see on smaller screens it is harder to read. We kept the order the same to make it a bit easier. In the video we referred folks to the main site for more on the charts. Good point though. When we make big charts, people complain they are too big.
Which server do you recommend for a Magento 2 store with more than 1 million monthly visitors, hosting more than 10 million digital products? AMD EPYC 7502P or AMD 5950x or Altra Q80-30?
What is the price tho?
If you have to ask, you can’t afford it.
It’s priced as per-datacenter, not per-machine.
@@JasperJanssen doesn't harm asking them for black friday or new year deals though?, Right right?
What is the price tho?
Sound is much better
when will the terahertz be?
So how did you get to aquire one of their servers?
Whats the retail on one of these?
so it has PSP ?
What is the power consumption of this server?
Wait so can it run androed nativly can you teiple boot androed windows and linix
So one Arm core as usual equals to less than half of x86 core?
Great video and very informative! The ARM server solution has really made a lot of progress. Back in the ThnuderX1 days it seemed clear to me that Intel left a wide open hole for ARM to walk right in, and they sure did move in. I figured that the arrival of EPYC would greatly slow down ARMs progression into x86 server territory, Intel should be thanking AMD. Hopefully you can review a ThunderX 3 server one day. I do run some cloud instances, X86 so far, maybe one day it'll be ARM instead, I have no reason to use x86 other than for the tool chain.
Really interesting Video. Thanks!
The only problem is that I'm currently not in the market to use or buy such a thing.
Oracle hasn't cared about Solaris since they bought Sun
Excessive ARM fan and early "blood sweat and tears" adopter reviews new ARM server. It's cool, to no ones surprise.
Guess I should start learning Linux or pray Microsoft can get their codebase migrated to ARM. Wonder if the age of x86 is coming to a close after all; I've remained skeptical for a while that it would happen. Apple's M1 didn't impress me because of the soldered, locked down nature of the kit. This server on the other hand is amazing.
I really hope to these these on ebay in ~3 years, i'm getting one!
Also, are these actually 19" servers or 21"?
19"
@@ServeTheHomeVideo Nice, no need to get new racks then :)
In my opinion the Ampere Logo looks very similar to the Areva NP Logo
Is Oracle Linux a viable alternative to CentOS? I wonder if this servers could run ESXi for ARM.
Is Wiwynn cheaper than grabbing a Supermicro server?
Wiwynn is more designed to deliver to hyper scale. Supermicro is designed for enterprise and channel sales. Somewhat different market segments they are playing in
@@ServeTheHomeVideo Sad, how i wish someone would focus on SME market although most people are moving to the cloud haha.
But can it run Crysis?
@25:55 No! No, no, no! We don't want another 2020. I'm hoping for an exciting 2021, not a 24 month long 2020.
So could someone tell me like what kind of uses would someone have for an ARM server? Is there even much out there in the way of software packages for ARM other Than PUBG mobile?
Lel. Anything that can run on a RPI can run here then, there are many Linux packages and web servers, databases and so much more.
Sure, most open-source software can be compiled to run on ARM.
@@TheBackyardChemist
gentoo linux xD everything compiled at system itself
все круто ..но цены на классные железки делают их многим недосягаемыми...остается только играть на старом GTX
Well, you won't get the best ARM or AMD performance with Intel compilers, that's for sure.
2021 is interesting
In the last video you told about a IBM power10 getting 2PB of RAM can you tell me how is that possible
I am going out on a limb here and guessing the "last video" was the one from this summer on Power 10. Conceptually IBM is making a fabric in their CPUs that can span not just within a chassis, but to multiple chassis. As they scale out, they get more nodes that they can attach memory to. Also, with their OMI, they get more flexibility with what kinds of memory they can use, and how much can be attached per node, similar to what the rest of the market will see in the CXL or CXL 2.x generations.
@@ServeTheHomeVideo
is that even possible with x86 because HP SuperDome Flex shows us that multiple nodes can be used to a total of 32 xeon processors and upto 48TB RAM, can this grow with x86? (a lot of ram and possibly even more processors)
@@hariranormal5584 sgi has this tech called Shared memory. Made programming easier since you can address all the memory. But if the memory reside on the other node. Big performance hit.
can you game on these?
One of the big use cases for Arm servers is gaming, but a bit differently. One can run mobile games on these servers with native Arm and then stream to mobile devices.
The question is - Can I play Minecraft on this bad boy?
13:59 I remindber
I hope that Wiwynn didn't make those cpu coolers. It would be a red flag as to quality. If an item doesn't look good, it generally isn't good.
They are awesome COU coolers and actually show the tight tolerances Wiwynn manufacturers to. Much tighter than Dell/ HPE. Put a bit around the bent find on p2 of the main site article.
WELWYN? They do more than servers.. they used to make a lot of the oem equipment. Far superior equipment than china’s build quality.
When you have more cores than pixels 🤪😜
only amd and intel - boredom :(
bit giggly on this one, to the detriment of quality.
So I guess it's not something you can buy for your homelab
These systems are VERY loud. I would highly discourage anyone from using this in a homelab environment.
@@MagnumCarta
So it's much louder than dell R720, hp dl380 g8 ?
It's interesting but this and all servers which are generally available are basically glorified PCs. I'm interested in what Oxide will produce in the next year or two as we've been stuck with these glorified PCs since Sun and boy do I miss Sun boxes.
But can it run Cyberpunk 2077?
please please please it's "I am Patrick from Serve The Home'
Show me some mining with it 🥰
I can’t wait for mips and arm to take over followed by risc-v opensparc .. X86 is garbage compared to the design of these. With exception of ppw instructions... but the cost is obvious.. we are pretty much hitting a wall with current tech
That will be a very long wait, it might not even come.
got my hand on one of those. they are JUNK, most of thins are cheaply made, you could press the power button and break it, firmware also junk, it's going to switch front VGA and back side VGA without anything change, all, both VGA ports are not working, the only way to access is with BMC, but BMC has ton of bugs too... save your headache... don't deal with this shit.
From my own experience with the Ampere platforms, Mt. Snow from Gigabyte is by far the most stable platform. The only caveat is it only has a single socket SKU. I will admit though that the power button on it also sucks. It's not as bad as Mt. Jade which can easily get lodged inside as the Mt. Snow has a membrane power button. However it can also get pushed in and get stuck.
That being said though these units aren't really meant to be powered on and off that frequently. They're intended to be in a datacenter especially with how effing loud they get.