GPUs & CPUs & Enthusiast hardware: Questions, Discussion and fanboy slap-fights - Nvidia & AMD & Intel - Separe but Equal. Intel rides in the back of the bus.

Wait wat? He is saying there are only sata expanders and no sas expanders? How is that even possible just type in 'sas expander' into amazon

This uses a chipset that is only compatible with SATA drives and not SAS drives, despite having the correct connector to connect to my servers backplates that my sas drives connect to.

The only compatible SAS expanders I found without them being 10x the price were raid cards.
 
  • Informative
Reactions: WelperHelper99
Well, good luck with that I guess
This one should work, and it can be flashed to IT mode to get the server direct access to the drives with no raid functions in the way. I'll probably hunt around for one that is already flashed

My raid card has a lsi 2000 series chipset which is just too old for modern cards. The 2100 series and 2200 series chipsets are compatible with 512e drives (if not 12Gb/s), and there are like 100 candidates that are each slightly different to look through

I just have to wait a couple weeks until payday as I jumped on a discount on the 8tb drives and bought two which cleaned out my budget. I'll be so relieved when I get my first full paycheck
 
Last edited:
  • Informative
Reactions: WelperHelper99
Explain what you mean, this sounds interesting.
It’s Spectre/Meltdown. Basically the CPU will guess at whether a conditional (“if…then”) will come true or not, which boosts performance quite a bit by letting the CPU continue to work while other parts of it are figuring out the conditional. It’s called branch prediction. The branch that doesn’t “win” is supposed to get flushed, but you can be lazy and skip this step, since the leftover data will just be overwritten anyway. Except until that happens, the data can still be read. Meltdown was similar and used a race condition to let code read any memory location in the entire system, which is a disastrous vulnerability since you could sniff out administrator or encryption keys that way. I’m not a computer scientist, I can’t really explain it all because I only understand the basics of it myself.

Anyway these are not a big deal if you’re sitting at home, any software that could sniff data this way might as well just issue a read to the file system and get tonnes more data with much less effort that way, but in a datacentre, where dozens of customers may be using a single processor at once in a VPS, I could sniff out your data from my share. It breaks memory protection barriers between containers/virtual machines. That’s catastrophic, considering how important VPS services like Azure and AWS are. Software fixes could be implemented, but affected Intel processors (something like 2011-2019) would be slowed down up to 30%. It was a huge hit to Intel’s competitiveness and credibility.
 
It’s Spectre/Meltdown. Basically the CPU will guess at whether a conditional (“if…then”) will come true or not, which boosts performance quite a bit by letting the CPU continue to work while other parts of it are figuring out the conditional. It’s called branch prediction. The branch that doesn’t “win” is supposed to get flushed, but you can be lazy and skip this step, since the leftover data will just be overwritten anyway. Except until that happens, the data can still be read. Meltdown was similar and used a race condition to let code read any memory location in the entire system, which is a disastrous vulnerability since you could sniff out administrator or encryption keys that way. I’m not a computer scientist, I can’t really explain it all because I only understand the basics of it myself.

Anyway these are not a big deal if you’re sitting at home, any software that could sniff data this way might as well just issue a read to the file system and get tonnes more data with much less effort that way, but in a datacentre, where dozens of customers may be using a single processor at once in a VPS, I could sniff out your data from my share. It breaks memory protection barriers between containers/virtual machines. That’s catastrophic, considering how important VPS services like Azure and AWS are. Software fixes could be implemented, but affected Intel processors (something like 2011-2019) would be slowed down up to 30%. It was a huge hit to Intel’s competitiveness and credibility.
Wow, seems like a bit of laziness on Intel's part. I'm at least glad it's fixed, but I can see how that would hit their reputation hard. If you're a data center and get attacked, they'll target the cores, something that you wouldn't think of protecting in the first place.
 
Security and performance are always at odds with one another. I don't consider it "shady" on their part. Customers want fast, then they complain about the damage later.
 
or they were just ignoring a huge security issue for better performance, i wouldn't put it past Intel with all the shady shit they have done over the years
I mean it definitely could be in their efforts to keep up with AMD. Raw performance over safety.
 
Wow, seems like a bit of laziness on Intel's part.

Designing complex systems is hard. The design flaw that Meltdown exploits was created in the 1990s. The Meltdown vulnerability wasn't discovered until 2017.

Intel's not the only chip maker that had these vulnerabilities, either, just the most prominent. AMD had some issues with Spectre, but not Meltdown.
 
Last edited:
Isn't this also a cisc vs risc issue? x86 based processor are risc and extremely complex, but arm based are risc and have simplified processes?
 
Isn't this also a cisc vs risc issue? x86 based processor are risc and extremely complex, but arm based are risc and have simplified processes?

All modern processors have speculative execution, including Arm and Sparc (today I learned Sparc is still around), which are RISC ISAs, so Spectre got just about everybody. Meltdown has to do with how the processor accesses memory, and this again is unrelated to instruction set - AMD and Oracle didn't have the issue, while Arm, Power, and Intel x86 all had it.
 
Always felt like it was rather over-academic of an exploit and meant to knock intel down a rung. This isn't really a real issue unless you are a VPS host or something.
 
  • Agree
Reactions: George Lucas
Always felt like it was rather over-academic of an exploit and meant to knock intel down a rung. This isn't really a real issue unless you are a VPS host or something.
or some bullshit small player that does 'cloud computing' like 'Amazon' or 'Microsoft' or 'Oracle'.
 
Unrelated to previous discussions - I'm finding that dual-socket Zen 4 EPYC server nodes are just kind of not really easy to get, not when you go for the high-core-count chips from the vendor. When I spec out what I need, I can only get single-socket. So with AMD, I can get 12 channels of DDR5 on the board, and with Intel, I can get 16. I did not expect this. The Xeon machine is still blazing hot, though - 2400W power supply vs 1400W for the EPYC. So I'm still better off with AMD for thermal reasons.
 
or some bullshit small player that does 'cloud computing' like 'Amazon' or 'Microsoft' or 'Oracle'.
Also known as a VPS host yes, hope that helps

this should work
Seems legit tbh, I was honestly just kindof concerned since a lot of raid controllers use a proprietary raid array that isn't always recoverable if the card dies, and then was a bit dubious of the whole '5 euro sas controller lmao'. To be honest I think I was being a little needlessly rude about it
 
Also known as a VPS host yes, hope that helps


Seems legit tbh, I was honestly just kindof concerned since a lot of raid controllers use a proprietary raid array that isn't always recoverable if the card dies, and then was a bit dubious of the whole '5 euro sas controller lmao'. To be honest I think I was being a little needlessly rude about it
They may be "RAID Controllers" but no one sane* uses the on-board RAID any more and they're just fancy JBOD to the OS.

* Kiwi Farms users may not be sane.
 
They may be "RAID Controllers" but no one sane* uses the on-board RAID any more and they're just fancy JBOD to the OS.

* Kiwi Farms users may not be sane.
I switched between software and onboard raid for a while since I was getting rubbish performance. Turned out it was all gigabyte's fault, among many other issues.

Their first AM5 outings were severely undercooked.
 
Their first AM5 outings were severely undercooked.
AM5 motherboards have really been eating it, IDK if they niggerified the PCB design department industry wide or what, I am on my second mini-itx board and this one at least boots reliably but the wifi card is dead (turns out its a very common issue for the ASRock AM5 (there is only one mini-itx ASRock board))

(first was in fact a gigabyte)
 
or they were just ignoring a huge security issue
Granted one of these examples isn't just Intel and it's more of a driver related vulnerability than solely hardware, but it wouldn't be the first time they've left a vulnerability to rot, nor is it going to be the last for Intel or many other "big tech" companies.

A lot of them will outright ignore any security issue they can until they either get blown out the water in the media because some bored dude managed to find the vulnerability in the first place and made a shitpost about it, or another party using their services suffer serious financial and reputational damage that can't be kept quiet and an internal employee lets it slip what really caused the compromise.

no one sane* uses the on-board RAID any more
It goes without saying that they're not sane in the slightest, but don't make eye contact with older enterprise on prem set ups if you can, far too many (at least where I've worked around) are still using it and the business often refuses to uplift even if they end up with such a terrible failure they have significant outages while getting services restored. It makes me weep
 
Back