|
Gonna get the 16 core and disable hyperthreading / the AMD equivalent 
|
#
?
Jan 3, 2019 17:16
|
|
|
|
| # ? May 31, 2024 11:22 |
|
|
Icept posted:Gonna get the 16 core and disable hyperthreading / the AMD equivalent Please forgive the dumb question I am about to ask: Why would you want to disable hyper threading? I was under the impression (and casual google searching gives the same) that the computer just uses it when it can (video editing, etc.) and ignores it otherwise (gaming?) with no downside? Sounds like there is a downside though, since you want to disable it.
|
|
#
?
Jan 3, 2019 17:23
|
|
|
Broose posted:Please forgive the dumb question I am about to ask: Why would you want to disable hyper threading? I was under the impression (and casual google searching gives the same) that the computer just uses it when it can (video editing, etc.) and ignores it otherwise (gaming?) with no downside? Sounds like there is a downside though, since you want to disable it. The recent security vulnerabilities found in CPUs owe their existence in part to hyperthreading. Disabling it by default is currently the accepted and recommended way forward. Intel even in their latest line up only offer hyperthreading to their very high end CPU.
|
|
#
?
Jan 3, 2019 17:31
|
|
|
Volguus posted:The recent security vulnerabilities found in CPUs owe their existence in part to hyperthreading. Disabling it by default is currently the accepted and recommended way forward. Intel even in their latest line up only offer hyperthreading to their very high end CPU. Hope you run ECC RAM, cause rowhammer is *way* imore practical than the non-Spectre SMT exploits.
|
|
#
?
Jan 3, 2019 17:45
|
|
|
Broose posted:Please forgive the dumb question I am about to ask: Why would you want to disable hyper threading? I was under the impression (and casual google searching gives the same) that the computer just uses it when it can (video editing, etc.) and ignores it otherwise (gaming?) with no downside? Sounds like there is a downside though, since you want to disable it. Above posts answered this from a security perspective but the point is to disable working threads on a what is presumably a higher speed-binned chip for less power consumption and less heat which could possibly mean higher and more sustainable overclocking behavior. Cutting 32 threads back down to 16 is apparently okay if you don't do a lot of work that requires a lot of parallelism but is definitely one of those things that makes more sense on paper than in practice and also some posters on Serious Hardware/Software threads will roll their eyes at you for not getting the most value out of the features you splurge for.
|
|
#
?
Jan 3, 2019 18:02
|
|
|
Volguus posted:The recent security vulnerabilities found in CPUs owe their existence in part to hyperthreading. Disabling it by default is currently the accepted and recommended way forward. Intel even in their latest line up only offer hyperthreading to their very high end CPU. Can't Microsoft and Linux coders put out patches that mitigate those vulnerabilities?
|
|
#
?
Jan 3, 2019 18:03
|
|
|
spasticColon posted:Can't Microsoft and Linux coders put out patches that mitigate those vulnerabilities? They did and NVMe SSD throughput took a HUUUGE poo poo, because of the extra steps needed to switch contexts to system for a driver level disk read then back to userland for a program disk contents read.
|
|
#
?
Jan 3, 2019 18:09
|
|
|
And Microsoft and kernel.org handled the system level and their end of userland but software developers still have to patch their software to avoid some of the worst variants.
|
|
#
?
Jan 3, 2019 18:12
|
|
|
The SMT/HT vulns on bare-metal installs can be defeated by modifications to the resource scheduler. For security-sensitive threads/processes, reserve or block the logical virtual thread so you can't do any funny business along side it. Would suck on a 2 core system but with core counts of 8+ becoming common it shouldn't hurt too much these days. Might be possible to signal that all the way up to a hypervisor. Meltdown suuuuuuck for anything doing high iops, you're generally going to need to build a trusted platform/cluster/whatever and disable mitigations for that until new generations of silicon properly do (beyond the current "we pre-loaded the microcode for you" poo poo)
|
|
#
?
Jan 3, 2019 18:21
|
|
|
drat, that sucks. So I guess it's either have fewer threads and therefore less throughput by disabling SMT or have your poo poo in the wind.
|
|
#
?
Jan 3, 2019 18:22
|
|
|
spasticColon posted:drat, that sucks. So I guess it's either have fewer threads and therefore less throughput by disabling SMT or have your poo poo in the wind. You need to be running threads with sensitive data that you don't want to leak for it to really matter. Like maybe you could drop a payload on a workstation and recover the account hash from ram or get the FDE key or something but the return on attacking a workstation is pretty low unless you're the president or CEO holding corporate secrets. These vulns are a much bigger issue for servers, maaaaybe business systems. I suspect the mitigations aren't going to be worth enabling for a home user, far easier ways to crack that nut and there's little payoff. Especially true now that NVMes are becoming common.
|
|
#
?
Jan 3, 2019 18:28
|
|
|
spasticColon posted:drat, that sucks. So I guess it's either have fewer threads and therefore less throughput by disabling SMT or have your poo poo in the wind. Your poo poo is in the wind no matter what, companies like Equifax and others made sure of that. I'd say, don't worry about it and just enjoy the ride. Gonna be a bumpy one.
|
|
#
?
Jan 3, 2019 20:14
|
|
|
Worrying about these types of exploits as a standard home user is silly. This is the type of stuff you worry about on a workstation or server. The amount of effort to do this, there's easier ways to steal $5,000 from your bank account. It's likely the easiest way to steal corporate secrets or gain access to the millions on their account which you can slowly skim without being noticed. These are not trivial exploits, anyone undertaking them is going to do the mental math and realize phishing just works better and is cheaper. For anything to even be obtainable you'd need to have that info still in RAM someplace, so unless you just did your banking or something else what are they going to gain? Your WoW password? That you played 10,000 hours of Skyrim and that you named a character Lowtax? It's not like they gain admin they can just read sensitive info from RAM, and actually thinking about it it might even just be L1-L3 cache which limits it even more. pixaal fucked around with this message at 20:31 on Jan 3, 2019 |
|
#
?
Jan 3, 2019 20:28
|
|
|
Anybody have any insight as to why AMD's stock took a 10% poo poo today? Or maybe it's more like 9% right now. Whatever.
|
|
#
?
Jan 3, 2019 21:38
|
|
|
How good are you at deciphering stockmarketese? This was the only thing I could find. https://www.nasdaq.com/article/interesting-amd-put-and-call-options-for-march-15th-cm1077057 Other assumptions might be cashing out on short-term gains ahead of CES. Also, a HEXUS article on fixing Windows's performance regression on AMD. https://hexus.net/tech/news/cpu/125819-amd-ryzen-threadripper-2990wx-perf-boosted-2x-coreprio-tool/ edit: Oh hey, AMD got added to the NASDAQ 100! That said, Intel and Nvidia are also both down 5% at time of post, so, probably something wider-spread. SwissArmyDruid fucked around with this message at 21:52 on Jan 3, 2019 |
|
#
?
Jan 3, 2019 21:42
|
|
|
Methylethylaldehyde posted:They did and NVMe SSD throughput took a HUUUGE poo poo, because of the extra steps needed to switch contexts to system for a driver level disk read then back to userland for a program disk contents read. edit: Zen2 is also supposed to have the fixes implemented in hardware which should help largely eliminate any performance hits too I'd think. PC LOAD LETTER fucked around with this message at 21:51 on Jan 3, 2019 |
|
#
?
Jan 3, 2019 21:48
|
|
|
Sneeze Party posted:Anybody have any insight as to why AMD's stock took a 10% poo poo today? Or maybe it's more like 9% right now. Whatever. The market overall took a poo poo, investors aren't reading tech news and see AMD as speculative. With CES around the corner if you think AMD is going to go up it's probably a good time to buy some or buy more. It's basically down to politics and uncertainty due to Government shutdown. It's not a classic pick, it's something people are going to drop because their portfolio took a massive dump the last few weeks. People will instead stick to safer things like food and oil. People are less likely to drop a tech stock like Intel MS of Facebook because they have been doing good for decades and have a solid track record. I personally think AMD is on to something and Intel is scared and has nothing but that's all my personal feelings based on Zen+ and my gut. Do your research before you invest but the current drop in AMD is not related to AMD failing at anything, at least not that I can find. SwissArmyDruid posted:Other assumptions might be cashing out on short-term gains ahead of CES. What short term gains? It's at the lowest price it's been in about 5 months barely ahead of the 6 month mark. It was at $30+ for almost half that time. It is an odd sell time to realize a loss though considering the year just started but if you only bought AMD because you saw it go from $15 to $30 and it's back at $15 and you bough at $30, you are probably fed up with it. pixaal fucked around with this message at 22:07 on Jan 3, 2019 |
|
#
?
Jan 3, 2019 22:03
|
|
|
Apple revized their projections down citing a contracting Chinese market, and it's not something that's restricted to Apple, so there's some market panic about what that means.
|
|
#
?
Jan 3, 2019 23:01
|
|
|
pixaal posted:The market overall took a poo poo, investors aren't reading tech news and see AMD as speculative. With CES around the corner if you think AMD is going to go up it's probably a good time to buy some or buy more. If I divorce everything I know about AMD's future outlook from how the market has performed over the past five days?  I see it jump a few bucks like that, sure, I'll get what I can and bail.
|
|
#
?
Jan 3, 2019 23:34
|
|
|
PC LOAD LETTER posted:I thought this was more for the Intel fixes since the AMD chips were inherently much less susceptible to these sorts of attacks in general and so the fixes have a much more minor impact (1-3%) on performance? Microsoft turned the meltdown mitigations on for all non-server OS's by default. You can turn it off with some reg keys but its there and platform agnostic.
|
|
#
?
Jan 3, 2019 23:42
|
|
|
Didn't know that, thanks.
|
|
#
?
Jan 3, 2019 23:47
|
|
|
ufarn posted:Apple revized their projections down citing a contracting Chinese market, and it's not something that's restricted to Apple, so there's some market panic about what that means. Yeah, and AMD is very very intertwined in the China stuff with the Dhyana. Wouldn’t be surprised if that’s worrying folks. Also the generic disclaimer that the stock price is only tangentially related to anything happening with a company.
|
|
#
?
Jan 3, 2019 23:56
|
|
|
Also, and this is one of those "forest for the trees" moments, yo, every major index is down by at least 1.5%, the Dow by 600 points, and the NASDAQ down by 3%. Not out of the question for AMD's stock to be eating poo poo atm.
|
|
#
?
Jan 4, 2019 00:48
|
|
|
Before this, there was also a general tech stock panic for reasons I don't know. Nvidia losing like 50% and Facebook stepping on one million rakes probably didn't shore up confidence.
|
|
#
?
Jan 4, 2019 01:46
|
|
|
ufarn posted:Before this, there was also a general tech stock panic for reasons I don't know. Nvidia losing like 50% and Facebook stepping on one million rakes probably didn't shore up confidence. You're short on the number of rakes by an order of magnitude and then some.
|
|
#
?
Jan 4, 2019 03:03
|
|
|
Seamonster posted:https://www.npr.org/2018/10/13/657172112/facebook-says-14-million-accounts-had-broad-array-of-personal-data-stolen Also remember that facebook does sometime request people's birth certificate 
|
|
#
?
Jan 4, 2019 03:21
|
|
|
I like that "checkpoint" is part of the URL.
|
|
#
?
Jan 4, 2019 04:40
|
|
|
people putting their real names on the internet voluntarily will never not be funny to meufarn posted:Before this, there was also a general tech stock panic for reasons I don't know. Nvidia losing like 50% and Facebook stepping on one million rakes probably didn't shore up confidence.
|
|
#
?
Jan 4, 2019 06:31
|
|
|
I've reminded a sourpuss about the AMD rumors, so I'll just say I'll believe it when I see it.
|
|
#
?
Jan 4, 2019 07:49
|
|
|
Sneeze Party posted:Anybody have any insight as to why AMD's stock took a 10% poo poo today? Or maybe it's more like 9% right now. Whatever. 
|
|
#
?
Jan 4, 2019 09:32
|
|
|
https://twitter.com/KOMACHI_ENSAKA/status/1081174660136353792 Big if true. 12C AM4
|
|
#
?
Jan 4, 2019 16:06
|
|
|
That's interesting, H2 stepping matches up exactly with the list I posted on the previous page, making that data more and more likely to be real. My god, AMD might have actually done it.
|
|
#
?
Jan 4, 2019 17:13
|
|
|
NewFatMike posted:https://twitter.com/KOMACHI_ENSAKA/status/1081174660136353792 lol it's Epyc processors with one completely hosed numa domain
|
|
#
?
Jan 4, 2019 17:14
|
|
|
Arzachel posted:Hope you run ECC RAM, cause rowhammer is *way* imore practical than the non-Spectre SMT exploits.
|
|
#
?
Jan 4, 2019 22:00
|
|
|
It's like "does my ext4/5 need ECC, because btrfs doesn't need ECC" questions all over again. Bitch, if the use of ECC were important to you, you would be using it REGARDLESS of whatever file system you eventually arrive at.
|
|
#
?
Jan 4, 2019 22:07
|
|
|
BangersInMyKnickers posted:lol it's Epyc processors with one completely hosed numa domain
|
|
#
?
Jan 4, 2019 22:07
|
|
|
Anime Schoolgirl posted:Waste no part of the 7nm chiplet buffalo I unironically love it and wish nothing but the best for AMD with this architecture moving forward
|
|
#
?
Jan 4, 2019 22:13
|
|
|
Combat Pretzel posted:Yeah, I don't get the drama home users create about these exploits. Home users have to be very worried about concept level exploits. *Downloads and installs dozens of Skyrim mods*
|
|
#
?
Jan 4, 2019 22:28
|
|
|
Whatever it takes to keep pushing technological advancement forward. The past ten years of Intel sitting on its rear end with 5-10% more IPC and lies about "oh, we can't do more than 4 cores with HT in a mainstream desktop socket" are not ones that I want to revisit.
|
|
#
?
Jan 4, 2019 22:28
|
|
|
|
| # ? May 31, 2024 11:22 |
|
|
SwissArmyDruid posted:Windows also needs to fix their core assignment. Wendell @ L1T, Ian at Anandtech, and the writer of Process Lasso seem to have nailed down that the performance regressions that the 2nd gen Ryzen exhibits down to a kernel bug. I just tried their CorePrio program on my 1950X and was running Indigo myself. Without CorePrio and without /affinity, I was getting 1.6 on Indigo. With just /affinity 0xFFFFFFFE I was getting 1.73 in Indigo(Launching Indigo with Core 0 affinity off). With CorePrio and no /affinity, I shot up to 2.2 in Indigo with my 1950x OC'd to 3.9GHz. So not the same 50% increase in performance he was, but I did see a ~37% increase in performance in just Indigo. I wonder if this would solve my OBS encoder issues I've been having where I'll drop 1% of my frames when trying to render a 1080p image to 720p output. SlayVus fucked around with this message at 23:12 on Jan 4, 2019 |
|
#
?
Jan 4, 2019 23:08
|
|