|
I think it's just a really stupid way to say "If 64GB is not enough and 128GB is too much, you can save some money with 96GB."
|
#
?
Jan 18, 2023 16:41
|
|
|
|
| # ? Jun 8, 2024 08:46 |
|
|
Lockback posted:I think it's just a really stupid way to say "If 64GB is not enough and 128GB is too much, you can save some money with 96GB."
|
|
#
?
Jan 18, 2023 16:42
|
|
|
SwissArmyDruid posted:someone explain that article to me, I got to this part and my brain just ground to a halt. They're referring virtual machines running on a hypervisor in a data center. That is why they are mentioning price optimized memory allocation per core.
|
|
#
?
Jan 18, 2023 16:43
|
|
|
AMD's competitor is currently selling subscription models for hyperscalers to turn on CPU features on the fly whenever customers want it, so they can charge customers more for having access to various forms of compute acceleration. I'm not sure that just because someone wants it, that it's a good idea.
|
|
#
?
Jan 18, 2023 16:45
|
|
|
Alzion posted:They're referring virtual machines running on a hypervisor in a data center. That is why they are mentioning price optimized memory allocation per core. Except earlier in the article it said: quote:Meanwhile, support for AMD EXPO and Intel XMP 3.0 profiles designed primarily for enthusiasts in mind indicate that these modules are indeed aimed at desktops
|
|
#
?
Jan 18, 2023 16:47
|
|
|
The article also said these configurations were more "optimal" for modern CPUs in an entirely separate point from overspending by buying more than you need. So that's not what they meant. They think there's some other magic that makes these memory amounts better for modern CPUs.
|
|
#
?
Jan 18, 2023 16:49
|
|
|
BlankSystemDaemon posted:I'm not sure that just because someone wants it, that it's a good idea. What exactly is the downside of non-power-of-2 ram modules? If it's not a good idea, why?
|
|
#
?
Jan 18, 2023 16:49
|
|
|
Klyith posted:What exactly is the downside of non-power-of-2 ram modules? If it's not a good idea, why?
|
|
#
?
Jan 18, 2023 16:52
|
|
|
I'm pretty sure this gonna unearth some design flaws in memory controllers because they're assuming power of two sizes per rank and an even number of ranks per channel. Or some such.
|
|
#
?
Jan 18, 2023 16:57
|
|
|
Drakhoran posted:Except earlier in the article it said: It makes sense, they want everyone to buy it so they throw everything in. BlankSystemDaemon posted:It's not a bad idea, it's the PR speak that's absolute bullshit, as is trying to justify it. As someone that has bounced in and out of technical marketing everyone posting about this has put more thought into this than the authors of that ad copy.
|
|
#
?
Jan 18, 2023 16:57
|
|
|
Combat Pretzel posted:I'm pretty sure this gonna unearth some design flaws in memory controllers because they're assuming power of two sizes per rank and an even number of ranks per channel. Or some such. EDIT: Also remember that most of the server platforms are based on a Unix-like, and UNIX as well as BSD started on a 18-bit platform. BlankSystemDaemon fucked around with this message at 17:05 on Jan 18, 2023 |
|
#
?
Jan 18, 2023 16:58
|
|
|
Twerk from Home posted:Sure, I simplified it some and you need overhead for the OS, but once you've got enough RAM for your workload + the most active pages in cache / buffers, adding additional RAM is not going to do very much and memory is expensive. You don't want to just buy 50% more RAM than you can effectively use. So in reality it comes down to this: If your budget happens to line up that way, you might be able to get slightly more memory than you hoped for. And let's not kid ourselves, more memory is always good.
|
|
#
?
Jan 18, 2023 17:02
|
|
|
Speaking of which, anyone got any hot takes on the new CAMM memory for laptops? Every time I look at tech articles about it I see a lot of comments with people hating it, often for no other reason than Dell came up with it. It sounds pretty great to me! Why are we still putting memory in paired sticks when you could put both channels on one module? Does anyone even do 3-channel anymore? And it solves one of the most despicable things in laptops: BlankSystemDaemon posted:I doubt it, because it's not exactly uncommon to, as an example, amend 8GB soldered memory with 16GB socketed memory in laptops.
|
|
#
?
Jan 18, 2023 17:06
|
|
|
Drakhoran posted:Except earlier in the article it said: The article isn't particularly well written. It does that throwaway desktop line in the beginning then spends the rest of the article talking about how this is good for VMs. Completely ignoring that you can market a product line for different use cases.
|
|
#
?
Jan 18, 2023 17:07
|
|
|
Klyith posted:Speaking of which, anyone got any hot takes on the new CAMM memory for laptops? Every time I look at tech articles about it I see a lot of comments with people hating it, often for no other reason than Dell came up with it. Lenovo has been doing the soldered+socketed nonsense for years, and while I like my T480s, I don't like that it only has 24GB memory. 
|
|
#
?
Jan 18, 2023 17:12
|
|
|
BlankSystemDaemon posted:So in reality it comes down to this: If your budget happens to line up that way, you might be able to get slightly more memory than you hoped for. You hit a personal pain point for me, because the last person to be responsible for computing purchasing decisions felt like this, and that's how we ended up with nodes with Xeon Silver 4208s and 1TB of RAM. It's a really inefficient usage of resources to pair $800 of CPUs with $6,000 of RAM.
|
|
#
?
Jan 18, 2023 17:12
|
|
|
Twerk from Home posted:You hit a personal pain point for me, because the last person to be responsible for computing purchasing decisions felt like this, and that's how we ended up with nodes with Xeon Silver 4208s and 1TB of RAM. It's a really inefficient usage of resources to pair $800 of CPUs with $6,000 of RAM.
|
|
#
?
Jan 18, 2023 17:15
|
|
|
BlankSystemDaemon posted:CAMM is just a proprietary memory form factor though, right? I hadn't realized anyone was working on it but Dell. They've proposed it to jedec as a standard, and jedec seems on board. (No idea if dell will want royalties for it tho.)
|
|
#
?
Jan 18, 2023 17:25
|
|
|
I like calling it non-binary RAM since that terminology will trigger a certain type of nerd moron
|
|
#
?
Jan 18, 2023 17:25
|
|
|
Klyith posted:They've proposed it to jedec as a standard, and jedec seems on board. (No idea if dell will want royalties for it tho.) It's already going through the early processes of approval at JEDEC. It's pretty much destined to become a JEDEC standard at this point. Dell will be collecting royalties for its use, though I'm not sure if this is any different from the currently in-use standards out there.
|
|
#
?
Jan 18, 2023 17:36
|
|
|
Klyith posted:Does anyone even do 3-channel anymore? I'm not sure if it was ever a thing outside of LGA1366. Following generations of HEDT/server platforms went to 4+ and consumers stayed at 2. e: also LGA1356, the weird secondary Sandy Bridge server platform. Eletriarnation fucked around with this message at 18:30 on Jan 18, 2023 |
|
#
?
Jan 18, 2023 18:27
|
|
|
hobbesmaster posted:As someone that has bounced in and out of technical marketing everyone posting about this has put more thought into this than the authors of that ad copy. lol
|
|
#
?
Jan 18, 2023 18:48
|
|
|
Klyith posted:It's enough for your job and is cheaper. my I/O buffers! Twerk from Home posted:You hit a personal pain point for me, because the last person to be responsible for computing purchasing decisions felt like this, and that's how we ended up with nodes with Xeon Silver 4208s and 1TB of RAM. It's a really inefficient usage of resources to pair $800 of CPUs with $6,000 of RAM. Facebook at least has buckets of nodes of roughly that shape, as enormous memcached(ish) servers. They work great!
|
|
#
?
Jan 18, 2023 19:26
|
|
|
Klyith posted:They've proposed it to jedec as a standard, and jedec seems on board. (No idea if dell will want royalties for it tho.) Even if it wasn't, what company doesn't want what's essentially a passive income. Look no further AMDs compatitor for how far they're willing to cowtow to the hyperscalers just to get it. FuturePastNow posted:I like calling it non-binary RAM since that terminology will trigger a certain type of nerd moron
|
|
#
?
Jan 18, 2023 19:43
|
|
|
Twerk from Home posted:You hit a personal pain point for me, because the last person to be responsible for computing purchasing decisions felt like this, and that's how we ended up with nodes with Xeon Silver 4208s and 1TB of RAM. It's a really inefficient usage of resources to pair $800 of CPUs with $6,000 of RAM. We're about to save a bunch of yearly repetitive cost by moving to right-sized GCP instances and storage for the Cassandra clusters we need. All it took was the usual questions: "why are we using N TB and provisioning 4xN TB?", "why does a recommended 16 vCPU node run on 24 and 32 vCPU instances?" etc. Over-provisioning just because wastes a lot of money at scale. My favorite from the old days was when the megalomaniac we'd left in charge of purchasing twisted DEC's arm to deliver us 4-CPU 21264 Alphas hot off the presses, like no one had 21264 boxes yet. He was proud that NASA(!!) had to be told to wait because he pressured them. Thing is we couldn't use them effectively at our current workloads and they were a huge waste sitting around blocked on NFS I/O. I'd even feel guilty logging onto them and seeing all that wasted cpu that could have been spent on ~space travel~. But faster is better, right? Klyith posted:It's enough for your job and is cheaper. This. Optimization is fitting the solution to the problem, not figuring out how to go fastest or biggest.
|
|
#
?
Jan 18, 2023 21:37
|
|
|
v1ld posted:He was proud that NASA(!!) had to be told to wait because he pressured them. Thing is we couldn't use them effectively at our current workloads and they were a huge waste sitting around blocked on NFS I/O. Sounds like he should have pressured NetApp too!
|
|
#
?
Jan 18, 2023 23:07
|
|
|
Subjunctive posted:Sounds like he should have pressured NetApp too! Lol. We were an early NetApp customer too! So well within the realm.
|
|
#
?
Jan 18, 2023 23:20
|
|
|
But if they don't dump money wastefully into hardware, they'll just end up using it to do stock buybacks! (because they sure as hell aren't going to put it towards salaries...)
|
|
#
?
Jan 19, 2023 00:51
|
|
|
I am happy to report that ASRock is back on their poo poo https://www.youtube.com/watch?v=NrKZeEmu6UA
|
|
#
?
Jan 19, 2023 20:21
|
|
|
DoombatINC posted:I am happy to report that ASRock is back on their poo poo "yes"
|
|
#
?
Jan 19, 2023 21:45
|
|
|
DoombatINC posted:I am happy to report that ASRock is back on their poo poo I'd buy one just to get the 2 NVMe devices on the breakout board. Maybe. Depending on final cost vs a X670 board with 4 NVMe devices. Unless there are actually B650 boards with 4x4 bifurcation support by the time I'm ready to migrate to AM5.
|
|
#
?
Jan 19, 2023 23:20
|
|
|
mdxi posted:I'd buy one just to get the 2 NVMe devices on the breakout board. Maybe. Depending on final cost vs a X670 board with 4 NVMe devices. To me it sounded like it's also a very special edition motherboard that nobody but Wendell has.
|
|
#
?
Jan 19, 2023 23:37
|
|
|
ASROCK’s monthly cocaine delivery must have been contaminated with some other substance. I imagine they’re testing the waters on this concept with streamers first.
|
|
#
?
Jan 19, 2023 23:42
|
|
|
ASRock stands for A Shitload of (cocaine) Rocks, confirming now
|
|
#
?
Jan 19, 2023 23:46
|
|
|
BlankSystemDaemon posted:It's a very neat idea for a daughterboard, sure - but I think you might've missed that a special header is needed, one which is only found on that particular motherboard. Yeah I got it. I meant "If it were productized..." Which I didn't explicitly say, but the video ended with a call for Asrock to make it a real thing, and I was going from there as a premise.
|
|
#
?
Jan 20, 2023 05:04
|
|
|
After fTPM messed with AM4 CPUs and taking tons of convincing of AMD that it did, it got eventually fixed via BIOS updates. Guess what issue is back for AM5.
|
|
#
?
Jan 23, 2023 23:15
|
|
|
Combat Pretzel posted:After fTPM messed with AM4 CPUs and taking tons of convincing of AMD that it did, it got eventually fixed via BIOS updates. Guess what issue is back for AM5.
|
|
#
?
Jan 23, 2023 23:17
|
|
|
BlankSystemDaemon posted:Right, that reminds me that I need to ensure that whatever motherboard I get has a Port 80 header.
|
|
#
?
Jan 23, 2023 23:27
|
|
|
Combat Pretzel posted:After fTPM messed with AM4 CPUs and taking tons of convincing of AMD that it did, it got eventually fixed via BIOS updates. Guess what issue is back for AM5. Do you have a link for this? I was planning on doing an AM5 build sometime in March. Is this affecting all boards manufacturers?
|
|
#
?
Jan 24, 2023 03:18
|
|
|
|
| # ? Jun 8, 2024 08:46 |
|
|
So far people complaining over on Reddit only. Like last time, it’ll probably take a while for an official stance.
|
|
#
?
Jan 24, 2023 07:20
|
|