[ / / / / / / / / / / / / / ] [ dir / girltalk / hikki / leftpol / monarchy / nofap / s / sw / v4c ][Options][ watchlist ]

/tech/ - Technology

You can now write text to your AI-generated image at https://aiproto.com It is currently free to use for Proto members.
Name
Email
Subject
Comment *
File
Select/drop/paste files here
* = required field[▶ Show post options & limits]
Confused? See the FAQ.
Expand all images

File (hide): 81801cdf11e1592⋯.png (304.86 KB, 1202x768, 601:384, Itanic1.png) (h) (u)

File (hide): af219d68671b370⋯.png (266.41 KB, 1460x781, 1460:781, Itanic2.png) (h) (u)

File (hide): 723c272ad19562d⋯.png (260.58 KB, 1384x672, 173:84, Itanic3.png) (h) (u)

[–]

 No.851824>>851831 >>851980 >>852033 >>852036 >>852399 [Watch Thread][Show All Posts]

IMMUNE TO THE BOTNET

So it occurred to me that now might be a good time to pick up Itanium server hardware on the used market to re-purpose as a personal workstation.

>Pros:

>

>No ME. These processors were designed for multi-socket systems, so they relied on the system to provide remote management, eg HP iLO. Typically this has its own separate ethernet port which can be left disconnected.

>

>Totally invulnerable to Meltdown and any future Spectre-class vulnerabilities. The Itanium architecture is based on explicit parallelism via VLIW. It doesn't do speculative execution.

>

>Gentoo port available

>

>

>Cons:

>

>Horrendous power draw and bloody noisy

>

Best option looks to be systems about 10 years old. By then it was only HP making them. These systems started around $10K and could run up to $100K+

There are still a few refurb IT outfits asking big money for these, but you can find them cheap enough on ebay. The blade servers are the most ubiquitous and cheapest, although this obviously presents the problem of acquiring an enclosure. Some of the blades sell cheap enough that it might work out to cannibalise them for CPUs and RAM to populate another server.

I've never worked with these systems myself. Does anyone have an insights into whether this is a viable idea? Performance comparisons to current hardware are a bit of an unknown. I believe it depends a lot on the workload, particularly how parallelisable it is.

 No.851826>>852033

File (hide): 118b2b6671c7f4d⋯.png (335.95 KB, 1216x672, 38:21, Itanic4.png) (h) (u)

File (hide): 1fa3d1d8e9d7d1a⋯.png (643.46 KB, 1729x828, 1729:828, Itanic5.png) (h) (u)


 No.851828>>851831 >>851835 >>851871 >>852031 >>852034

maybe just get a beaglebone black


 No.851829>>851840 >>852031

fucking retard never come back here


 No.851831>>851835 >>851840 >>852031

>>851824 (OP)

For what purpose?

For most purposes that an average kid on this site, they don't need that high power server stuff. They could just easily do with a SBC like Raspberry pi or what >>851828 said and not put a huge dent in their neetbux pile.

Besides, I wouldn't imagine the supply for these computers being enough for any substantial switchover.


 No.851835>>851837

>>851831

> They could just easily do with a SBC like Raspberry pi or what >>851828 said

It would be a viable option if shit was useable as a desktop PC or at least worked.


 No.851837>>852031 >>852553

File (hide): 12562a478353b0a⋯.gif (752.58 KB, 303x224, 303:224, theChase.gif) (h) (u)

>>851835

more usable and functional than a fucking itanium, what the fuck are you thinking


 No.851840>>851849

>>851829

Well, I guess that confirms my theory that CIAniggers can't into Itanium.

>>851831

As a non-botnet, non-spectre-vulnerable, general-purpose workstation. I think this "high power server stuff" from 10 years ago might be fairly comparable to modern consumer CPUs performance wise - or at least it can be made to be, as these systems range up to 16 sockets. Of course it's not directly comparable for single-thread performance, this architecture thrives on parallel workloads.

And who said anything about a "substantial switchover"? Remaining a niche platform not in CIAnigger's sights is an advantage.

There is another negative - EFI BIOS, no core/libreboot available. Weighing this against ME and spectre vulnerabilities, I still think it comes out ahead.


 No.851849>>851851

>>851840

>intel

>no backdoor

try again, sweetie.

you're looking at pretty expensive space-heaters.


 No.851851

>>851849

I wondered about down-clocking them.


 No.851856>>851860 >>851873 >>852024 >>852031 >>852406

Keep in mind, you won't be able to run x86 programs.


 No.851860>>851973

>>851856

And how would lack of x86 binary compatibility be an issue when running Gentoo, which compiles from source?


 No.851871>>851885 >>852022 >>852031

>intanium

It maybe non-botnet, but it's fucking weird. Get SPARC/MIPS64/POWER hardware instead. Or >>851828


 No.851872>>852022

Wouldn't it be better to use SPARC server? But I don't know how much those cost.


 No.851873>>851946

>>851856

Why should I care? I have free software! This means I have the freedom to make all my programs run without x86!


 No.851881>>851885 >>852000 >>852031 >>852557

File (hide): 04c5f506e6502e6⋯.png (26.58 KB, 120x196, 30:49, 1328036157357.png) (h) (u)

That's a lot of work when I can just deal with AMD.

I know someone's gonna say "but that's backdoored too!", but honestly, I trust AMD with this kind of access. They're just not in the position to fuck over consumers. They don't have the money to pull it off. If there's ever a huge controversy surrounding the security of their CPUs, the company is done, period. Intel can tank bad press like it's nobody's business.


 No.851885>>851892 >>851946

>>851871

POWER is vulnerable to spectre.

SPARC status unconfirmed, I'm investigating.

Are there MIPS64 systems with reasonable workstation-level performance? I thought MIPS was relegated to embedded these days.

>>851881

AMD is vulnerable to spectre.

Spectre is not just one hole, it's a new class of vulnerabilities of which Meltdown was just the first example found. Itanium is immune by design due to it's EPIC/VLIW architecture.


 No.851892>>851900

>>851885

So far all of the press I read says that the manner in which AMD is vulnerable is either easily patchable, or not a security risk. But I honestly haven't been following this shit too closely, so if you have an article or something suggesting otherwise, I'll read it.


 No.851900>>851915

>>851892

They haven't found a spectre vuln on AMD or POWER yet, but any processor that does speculative execution is potentially vulnerable to this newly conceived vector of attack.


 No.851915>>851928

>>851900

...so you're deciding not to buy chips because they are POTENTIALLY vulnerable?

The fuck is that? Literally every piece of silicone on the market is vulnerable to SOME kind of future exploit. You might as well stop using computers altogether if you're that paranoid.


 No.851928>>851935

>>851915

It's not just "potentially vulnerable" in a hypothetical "there are always new exploits" sense. It's potentially vulnerable in the "It has a fundamental design flaw which opens up an entirely new vector of attack" sense. Speculative execution was a mistake.


 No.851935

>>851928

A dam could have a "fundamental flaw" somewhere and still stay up for decades. The chips are a security risk when there's an exploit in the wild, and not a second before.


 No.851946

>>851873

And this is why you go all FOSS or nothing, you can use it on any platform you compile it for. OP a Itanium with a pci-e nvidia gpu with nouveau would be very nice.

>>851885

Older powerpc CPU's like those in the powermac G4's are not vulnerable. And appearently itanium is not either.


 No.851973>>851975

>>851860

You do know most programs cant just be compiled for other architectures right?


 No.851975>>852133 >>852143

>>851973

Yes they can you giant faggot. Unless it is something like a emulator with hardcoded assembly in memory or using specific things to the architecture like cryptodev it will compile if it uses C code and the target architecture supports the features neccessary for C. Which is to say a MMU, superscalarness, having enough RAM, and having had a general C library ported to it. All of which itanium supports or has.


 No.851980

>>851824 (OP)

That second one is probably worth 3 times its sale price for the RAM alone. Whoever's selling that is retarded.


 No.852000

>>851881

AMD-PSP: fTPM Remote Code Execution via crafted EK certificate

http://seclists.org/fulldisclosure/2018/Jan/12

¯\_(ツ)_/¯


 No.852022

SPARC ftw

>>851871

>>851872

All it needs to fix a problem is a literal slap.


 No.852024

>>851856

Not necessarily. Various combinations of software such as Bochs, QEMU, KVM, and WINE will allow you to emulate other architectures and OSs, often at near-native speed for very library-heavy software.


 No.852031

>>851828

>>851829

>>851831

>>851837

>>851856

>>851871

>>851881

>ayy lmao, just get a raspberry pi or a Mac

The 4chan in here is palpable, or are they just glowing?

TBH OP, It's not a bad idea, might be a fun little project if you can pick something up cheap enough. Although personally I'd like the Sun Fire SPARC server myself, just because Intel sickens me so much. I would love an update thread with pics and vids to see how you progress if you do this.


 No.852033>>852386

>>851824 (OP)

>IMMUNE TO THE BOTNET

That's where you're wrong.

Speculative execution was introduced in Poulson (~2012) to improve the performance.

But as long as you buy something older you're fine.

>>851826

>blades

Do you realize those need an enclosure that costs an order or two of magnitude more?


 No.852034>>852048 >>852049


 No.852036>>852042

>>851824 (OP)

OP, get a sunfire E25K

The E25K supports up to 72 dual-core UltraSPARC IV+ processors (up to 1.95 GHz). As with UltraSPARC IV-based 15K systems, the "MaxCPU" option was not offered for E25K systems. Overall system bandwidth is claimed as up to 172.8 GB/s aggregate, up to 115.2 GB/s peak, and up to 43.2 GB/s sustained. For overall I/O bandwidth, up to 35.8-GB/s sustained. Up to 64 GB of RAM per board is possible with a maximum of 1.15 TB of RAM for a single domain. Up to 72 hot swappable PCI-X I/O slots; 54 slots are 90 MHz, 18 slots are 33 MHz. It also supports 10/100 BaseT Ethernet, Gigabit Ethernet, UltraSCSI (LVD and HVD), ATM, FC-AL, HSI and SCI.


 No.852042>>852215

>>852036

>up to 1.95 GHz

>SPARC

lol

might as well get a pocket calculator because that's what low-clocked RISC amounts to


 No.852048

>>852034

What's wrong with Beaglebone exactly? I know some boards need non-free software to even boot-up, but that page only covers a small subset of all ARM SBCs.


 No.852049>>852083

>>852034

if you can live without 3d acceleration, that's as free as you'll get


 No.852083>>852085 >>852104

>>852049

Yeah, I don't care about 3D stuff at all. Only problem is, some of these boards have very limited RAM, and I need to be able to run Firefox or Chromium once in a while. On my amd64 laptop, Firefox sucks up nearly 170 MB after just starting up. Homepage is set to my home directory, so it's not even rendering any webpage, CSS, javascript, and so on. Maybe 512 MB would be sufficient, but I'd feel safer with a 1 GB board to avoid swapping (which would be pretty horrible especially if your SBC is running with an SD card rathar than hard disk).

ARM Cortex-A7 seems to be recommended choice as per https://marc.info/?l=openbsd-misc&m=151527756600887&w=2 so that rules out a lot of other boards. Of the ones that remain, Banana Pi (and M2/M3 variants) stands out as a possibility, since it's based on Allwinner Axx SOC, which isn't listed under "fatal flaws" in https://www.fsf.org/resources/hw/single-board-computers so presumably that means it can boot without non-free software. Then again, they don't directly confirm this, so who knows...

Also here's a useful page to compare specs of various ARM SBCs: https://en.wikipedia.org/wiki/Comparison_of_single-board_computers


 No.852085>>852091

File (hide): a036e5a6cea87e1⋯.png (238.72 KB, 674x1062, 337:531, 60264741_p0.png) (h) (u)

>>852083

>I need to be able to run Firefox or Chromium once in a while

>I need my hardware to be absolutely backdoor free so I can run backdoored spyware on it


 No.852091>>852115

>>852085

I'm not at all a fan of those bloated browsers, but I'm forced to use them on occasion, because web developers insist on making sites that only work in those kind of browsers. I try to minize my interaction with Web 2.0, but I'm not at the stage where I can just cut it off completely. Personally I've no use for any computer more powerful than a non-accelerated Amiga 500. And I really hate upgrading OS and other shit constantly. Modern computeting sucks ass, but I'm kind of stuck using it. My aim is simply to minimize my exposure and interaction with the modern browser shit, that's all. If I could do everything Unix-related in text terminal, I would.


 No.852104

>>852083

I have an Olinuxino A20 Micro that works with 100% free software and with the mainline kernel, has SATA and 1GB of RAM. I've never tried using X though, but apparently 2D works.


 No.852115

>>852091

I'm being 100% serious when I say this: if you have some website that's so important to you that it makes you use CHROME, maybe you should consider just writing your own standalone reader for that site and that site alone. Not a general purpose web browser, but a single purpose standalone app viewer. FOSS niggers do it all the time on Android, there's nothing stopping you from doing it on desktop.


 No.852133

>>851975

... which is a small subset of popular programs.


 No.852134

You're still a Fagette


 No.852143>>852154

>>851975

could you even run firefox on it?


 No.852150>>852151 >>852153 >>852158 >>852159 >>852219 >>852232

>1.5GHz

Why are you faggets focusing on this dead ass architecture when you could be looking at Lemote MIPS. Same cockspeed and it's gonna go up soon, fully libre PMON boot firmware, made for GNU/Linux.


 No.852151>>852198

>>852150

where can you buy them?


 No.852153

>>852150

>>1.5GHz

Are you retarded? Itanium does a lot more per cycle than other architectures.

>Lemote MIPS

>Same cockspeed and it's gonna go up soon

Oh, right, you are retarded.


 No.852154

>>852143

http://en.znurt.org/search.php?search=&q=firefox&x=0&y=0

52.5.2 is at "testing" status. This means someone has compiled and is using it. IA64 stuff rarely makes it to "stable" status due to the small userbase.

57.0.4 is "unknown" status, which means you can go ahead try it and see.


 No.852157

well its not going to be immune if its no longer obscure as it is


 No.852158>>852198

File (hide): e6af7df83192e40⋯.jpg (Spoiler Image, 120.92 KB, 801x1200, 267:400, C0QZqPdVQAA4XD2.jpg) (h) (u)

>>852150

>MIPS

aka chinese freedom cpu

revolutional comrade!


 No.852159>>852160 >>852161 >>852179

>>852150

Weren't those dropped by everyone because too slow?

Even Stallman moved from Lemote Yeeloong to a T400s.


 No.852160

>>852159

that was their yeeloong netbook.


 No.852161>>852400

>>852159

>Even Stallman moved from Lemote Yeeloong to a T400s.

IIRC that was because his Lemote was stolen in some south-murican shithole


 No.852179>>852198 >>852398

>>852159

He moved to a t60? Or an x60? Something with libreboot anyhow


 No.852198>>852398

>>852179

x60.

>>852151

Currently I don't know, but I do know that there was a german anon on halfchan who said he was gonna start a business reselling them into western markets. Dunno if he's actually gonna go through with that though.

>>852158

Well considering that the boot firmware used on the boards is way more free than 90% of what's out there, yeah it's pretty damn good in that respect

also thats a cute doll, anon


 No.852215>>852245

>>852042

Are you fucking retarded?

>The E25K supports up to 72 dual-core UltraSPARC IV+ processors (up to 1.95 GHz).

>72 processors

Do the math.


 No.852219

>>852150

Actually they offer an mitx board that looks really good, was thinking of getting one. However I am holding out hope for POWER9 or Risc-V, the MIPS stuff they use seems very weak.


 No.852232

>>852150

>Lemote

WEW


 No.852245

>>852215

Stop LARPing that you know what you're talking about.

That thing is 72 low-clocked RISC dual-cores from the year 2003. Ancient technology.

Might have been great if you were running enterprise shit like Oracle or SAP in mid 2000s, but useless for a personal workstation in 2018.

It's large, draws incredible amounts of power, produces fuckload of heat, and still sucks performance-wise.

Fucking autist.


 No.852386

File (hide): 50425cdf42dc6b8⋯.png (280.66 KB, 1335x615, 89:41, Enclosure1.png) (h) (u)

>>852033

I did in fact mention the need for a blade enclosure in the OP...

Here's one that I found for a reasonable price. The problem I see with blades is that they don't have standard PCIe slots, they take XMC Mezzanine modules instead. This would necessitate a hard to find, probably expensive adaptor to bring the PCIe signalling out to an external enclosure for a graphics card.


 No.852397>>852400

POWER9 and TALOS II


 No.852398

>>852179

>>852198

Nope. T400s now.


 No.852399>>852617

>>851824 (OP)

G5 outperforms it and is more easily available and is better supported.

It's going to be a dinosaur though, if you can lay your hands on a working setup and preserve it real nice it'll be worth a mint in 25 years. I've sunk thousands of dollars into speculative retro-computers which aren't quite old enough to be worth fuck you money like an Apple I but which are destined to appreciate.

I just picked two specific genres of stuff too, you can't just start buying every heap of shit out there, if you want to get into this hobby which now, oddly, might have a practical side, learn and become an auteur. Specialize.


 No.852400>>852567

>>852161

Enriched.

>>852397

Supposedly vulnerable to SPECTRE as well. Has that been confirmed for sure? I'm still interested in the things for other freedom reasons.


 No.852406

>>851856

70-80% of the software that the average /tech/nitian on here uses could be ported by the better 25% of us easily. If it aint foss, then fuckem.


 No.852461

>time to raise the itanic?

Go advertise on 4chan.


 No.852553>>852592

>>851837

You have to go back you fucking newfag sucm


 No.852557

>>851881

kill yourself shill


 No.852567>>852574

>>852400

POWER9 has been prereleased patched, SPECTRE and MELTDOWN have been fixed with zero performance loss according to Raptor Engineering.


 No.852574>>852752

File (hide): 192b1535c9d8865⋯.jpg (10.31 KB, 300x225, 4:3, 1435913206924.jpg) (h) (u)

>>852567

>thing has zero performance loss according to people that sell the thing


 No.852592

>>852553

>sucm

Now that's rage.


 No.852617>>852703 >>852706

>>852399

>G5 outperforms it

Are you sure about that?

Most powerful G5 had 2 x 2.5Ghz PowerPC 970MP dual core CPUs. 2005 vintage.

The Itanium systems I'm looking at have Montvale Itanium 9140M processors from 2007.

Itanium performance was famously disappointing, but you have to take that in the context of the market it was aimed at and the price. It was a much more expensive piece of silicon than the IBM chip. Performance comparisons are hard to come by, it's a very different architecture and it's said to depend heavily on the application, but I'd be surprised if the G5 kicked the Itanium's ass on a core-for-core basis.

Even if core-for-core it does outperform, these HP Itanium systems go up to 16 x quad cores. I could potentially get 4 x dual cores in my price range.


 No.852703>>852706

>>852617

Yeah, the G5 would get obliterated, mainly because it's a scaled-down POWER4 arch meant for workstations and Itanium is a full-blown big iron chip. The massive amounts of cache alone would make a huge difference.

More interesting (and fair) would be comparing Itanium and POWER.


 No.852706>>852780

>>852617

>Even if core-for-core it does outperform, these HP Itanium systems go up to 16 x quad cores. I could potentially get 4 x dual cores in my price range.

You're probably going to need 10+ amps for it though.

>>852703

These systems are probably all on the SPEC site, maybe later I will look into them. Anyway if you're not 3D gaymen anything post-2000 is probably powerful enough for most uses. I use an old X41 for shitposting and it can handle modern JS-heavy sites even if it's a little slow.


 No.852752>>852766

>>852574

It's their claim, the only companies on the planet who can say otherwise are Google and IBM.


 No.852766

>>852752

And everyone else using POWER9 systems right now. Wikipedia lists a several other high-profile customers, I'm sure there are many more.


 No.852780>>852823

>>852706

Yes, I'm aware that power consumption, and the noise from all the associated cooling, may be the biggest issue.

Fortunately the 9100 series were among the best for this. TDP is 104W. Some others went up to 260W. I don't know what the total system power usage would be, obviously with 4 of them it's still going to be juicy, but I'd guess it might be comparable to some of the mental SLI gaming rigs out there - not beyond reason.


 No.852823>>853212

>>852780

>need 104w X 4

>Tannoy's Law says you need at least the sum of cores for bus power

Even if they perfectly engineered the things you probably will draw 2kw idle.


 No.853212

File (hide): 5c9278c8ca5e68a⋯.png (30.82 KB, 668x313, 668:313, TannoysLaw.png) (h) (u)

>>852823

>Tannoy's Law says

That's a new one on me and on Google.




[Return][Go to top][Catalog][Screencap][Nerve Center][Cancer][Update] ( Scroll to new posts) ( Auto) 5
78 replies | 12 images | Page ?
[Post a Reply]
[ / / / / / / / / / / / / / ] [ dir / girltalk / hikki / leftpol / monarchy / nofap / s / sw / v4c ][ watchlist ]