|
wargames posted:I think its the multiple menus/expo in fighting with one another one, or multiple areas adding an offset. like expo is adding 0.2v offset and something else is adding another 0.1v offset or something. One of the most important things about writing configuration into software is for there to only be one place to configure things; on some sort of non-volatile storage. You can then either restart the software, or in some instances cause it to reload itself, based on that configuration. Runtime configuration is always the bad option. It doesn't matter if it's user-space, kernel-space, or firmware - and it doesn't even matter if it's obfuscated like it is in Windows or macOS. Klyith posted:Pretty much yes. They took a Neoverse IP core from ARM, because that's what AWS Gravitron is, and Googles butt service is just a Altra (Max?) from Ampere. They're just paying licensing costs to ARM, instead of paying for CPUs from AMD or Intel. If you hate money, you can buy one from Gigabyte with 256 cores and no SMT.
|
#
?
May 4, 2023 18:51
|
|
|
|
| # ? Jun 3, 2024 11:44 |
|
|
Twerk from Home posted:Did AMD fully shut down / sell off / give up at making ARM CPUs? I feel like Ampere, AWS, and Apple have already done the hard work, and software portability to ARM is the best that it's ever been. Klyith posted:Pretty much yes. So… the question here is what do you mean by “AMD”, “shut down” and ARM CPUs. I don’t recall and a quick search didn’t reveal anything other than AMD saying that their custom silicon solutions can use ARM processors. Anyway… Xilinx is AMD and sells plenty of MPSoCs which are 2-6 core ARM SoCs with an FPGA.
|
|
#
?
May 4, 2023 19:42
|
|
|
https://www.theregister.com/2022/06/20/jim_keller_arm_cpu/ They had a ARM project that got canned when Keller left.
|
|
#
?
May 4, 2023 19:46
|
|
|
Twerk from Home posted:Did AMD fully shut down / sell off / give up at making ARM CPUs? I feel like Ampere, AWS, and Apple have already done the hard work, and software portability to ARM is the best that it's ever been. They sold off during the bad time they sold off their arm stuff, they are however using arm on their x86 cpus as a security enclave.
|
|
#
?
May 4, 2023 19:49
|
|
|
ConanTheLibrarian posted:I know Nintendo don't target high specs but that seems like a really underpowered APU. Assuming they go with the 1024 cuda core version of the Orin like the rumors say, it is a pretty significant step up from the X1. Likely at least 3x the GPU performance, and its all but certain that they are going to lean heavily on DLSS. The Switch launched at $299 with an included dock, and I imagine Nvidia would like to land there again.
|
|
#
?
May 4, 2023 19:49
|
|
|
Pablo Bluth posted:https://www.theregister.com/2022/06/20/jim_keller_arm_cpu/ Pablo Bluth posted:https://www.theregister.com/2022/06/20/jim_keller_arm_cpu/ quote:In the talk, available online via YouTube, Keller discusses how when planning the Zen 3 core – now at the heart of AMD's "Milan" Epyc processor chips – he and other engineers realized that much of the architecture was very similar for Arm and X86 "because all modern computers are actually RISC machines inside," and hence according to Keller, "the only blocks you have to change are the [instruction] decoders, so we were looking to build a computer that could do either, although they stupidly cancelled that project." The question would be what is the value add over the ARM IP cores.
|
|
#
?
May 4, 2023 19:53
|
|
|
BlankSystemDaemon posted:That probably is it, but why are there multiple places to set it in the first place? Because they had a suppppper talented bios writer/programer leave like 4-5 years ago because of lack of raises if i remember right and why would you pay your employees what they are worth?
|
|
#
?
May 4, 2023 19:55
|
|
|
wargames posted:Because they had a suppppper talented bios writer/programer leave like 4-5 years ago because of lack of raises if i remember right and why would you pay your employees what they are worth? Hey fella some of those poor executives are struggling to pay third mortgages and boat loans! Won't somebody think of the C suite?
|
|
#
?
May 4, 2023 20:05
|
|
|
repiv posted:the GBA, DS, 3DS, wii and wii u could all play previous generation titles with near-100% compatibility, and the switch gets a pass since the form factor change made backward compatibility impractical And the Gameboy Color, that could also play Gameboy games. There was also a Gameboy adapter for the SNES & GameCube. As you say, Nintendo definitely has a legacy of reasonable backwards compat
|
|
#
?
May 4, 2023 20:09
|
|
|
i think people tend to think of the GBC as a "pro" refresh of the gameboy, not a distinct generation, likewise with the DSi but yeah more often than not nintendo has made their systems backwards compatible
|
|
#
?
May 4, 2023 21:01
|
|
|
Cygni posted:Assuming they go with the 1024 cuda core version of the Orin like the rumors say, it is a pretty significant step up from the X1. Likely at least 3x the GPU performance, and its all but certain that they are going to lean heavily on DLSS. The Switch launched at $299 with an included dock, and I imagine Nvidia would like to land there again. A 3 - 4x boost to total system performance is about what I'm expecting, which is kind of disappointing to be honest, but it is what it is. I think the switch maxes out at 10 watts, docked. If Nintendo wants to meet that same target again, then a 3x boost to performance is probably the best you can hope for, and it might even be a little lower. It's all about the efficiency improvements going from Maxwell on TSMC 20nm to Ampere on Samsung 8nm, which is probably less than you'd think. But if Nintendo decides to up the power budget, then that changes things. If they really wanted to, they could do a much bigger difference between docked and handheld power targets. Doing 20 - 30 watts docked instead of 10 would help them output to 4K TVs using DLSS. The catch is that they'd have to make the handheld bulkier to accommodate a better cooling solution, but I think the ROG Ally is showing that decent cooling can be had in an acceptable form factor (youtuber hands-on impressions said the device wasn't very loud at 30 watts) edit: why are we talking about this in the amd cpu thread again? Dr. Video Games 0031 fucked around with this message at 21:17 on May 4, 2023 |
|
#
?
May 4, 2023 21:09
|
|
|
Dr. Video Games 0031 posted:Doing 20 - 30 watts docked instead of 10 would help them output to 4K TVs using DLSS. The catch is that they'd have to make the handheld bulkier to accommodate a better cooling solution, but I think the ROG Ally is showing that decent cooling can be had in an acceptable form factor (youtuber hands-on impressions said the device wasn't very loud at 30 watts) Build a 120mm fan and a baffle into the dock, baby 
|
|
#
?
May 4, 2023 22:57
|
|
|
Klyith posted:Pretty much yes. Yep. MS has farmed out a custom ARM cpu to marvell to design and manufacture. doesn’t have to be to fancy. just more cost efficient and cheaper than x86 from the big vendors.
|
|
#
?
May 5, 2023 00:33
|
|
|
Klyith posted:
The insanely competitive state of the ARM market seems like a biiiig problem for the x86 vendors. ARM doesn't have to win in every way or all of the time to take a big chunk out of Intel and AMD's highest margin server market, and the competition in the ARM space means that they are iterating and evolving really fast. Over the last few years it's gone from "yeah ARM may be efficient but it doesn't clock high or perform well" to "Sure ARM has decent performance but none of the platforms have a lot of I/O or memory bandwidth" to "Well ARM doesn't have wide vector extensions and can't compete on vectorized workloads". Now SVE is here and ARM pricing is competitive even for small time purchasers. If your stuff can run on ARM it's cheaper to do so on AWS now, even if performance is a little worse. You can buy those Ampere Altra boxes from several vendors for reasonable prices. ARM is here not to conquer all spaces, but to eat away at both AMD and Intel profits.
|
|
#
?
May 5, 2023 02:20
|
|
|
Twerk from Home posted:The insanely competitive state of the ARM market seems like a biiiig problem for the x86 vendors. ARM doesn't have to win in every way or all of the time to take a big chunk out of Intel and AMD's highest margin server market Could very well be, but I don't know and am not quite ready to go chicken little just yet. The highest margin space for AMD and Intel is still the HPC stuff. An alternate hypothesis is like, for a long long time power efficiency wasn't the biggest demand from server customers. It is now because the clown succeeded in eating the world, and is now competing against itself and hunting for more profit margin. ARM was much better positioned to take advantage of that change in emphasis than x86. And in the non-server space, it's still ... very difficult to get good comparisons for gaming performance between normal enthusiast PCs and high end ARM (ie M1/2). Twerk from Home posted:The insanely competitive state of the ARM market seems like a biiiig problem for the x86 vendors. ARM doesn't have to win in every way or all of the time to take a big chunk out of Intel and AMD's highest margin server market Personally I'm happy for ARM to be competitive in areas besides cell phones and chromebooks. The stretch where AMD poo poo the bed and we had near-total Intel monopoly was the worst. Now even if one x86 company goes kablooie they'll still have to watch their back and compete.
|
|
#
?
May 5, 2023 04:31
|
|
|
Twerk from Home posted:The insanely competitive state of the ARM market seems like a biiiig problem for the x86 vendors. ARM doesn't have to win in every way or all of the time to take a big chunk out of Intel and AMD's highest margin server market, and the competition in the ARM space means that they are iterating and evolving really fast. It's considerably better clocked than the highest core count processors from AMD and Intel, which have 96 cores at 2.4GHz base clock (3.55GHz all-core boost, hypothetically) and 60 cores at 1.9GHz base clock (3.5GHz all-core boost, hypothetically), respectively. All-core boost is hypothetical because no HPC cluster I know of enables turbo-boost, as the systems aren't validated for it. They also disable SMT, so the number of threads AMD and Intel have is irrelevant.
|
|
#
?
May 5, 2023 10:06
|
|
|
Nearly six months to the day after building a system with the Gigabyte B650E Aorus Master, I think the board might be toast. Aside from a weird integrated graphics issue I posted about somewhat recently, it has been performing well without issues. I tried restarting my PC a bit ago and it refuses to boot. Sits at code 15 (north bridge mem init) for a bit for the training, eventually cycles over to 97 (console output devices connected) for 20 sec where I get GPU output to the monitor (just a blinking underscore) and then it restarts again. Oddly, even when I hold power to force power off, the RGB on the motherboard is still lit. Historically, it has turned off when the PC is off. I can't even get into the BIOS anymore. Gonna try to clear CMOS, I guess.  edit: Okay, after resetting CMOS, the computer restarted 3 times in a row before outputting any video. Finally, I saw the AORUS splash screen on the third boot and could get into the BIOS. After being greeted by the CMOS reset screen, I enabled advanced settings, turned on the EXPO profile (still states 1.3v in BIOS, HWinfo says 1.245v), disabled memory context restore, disabled SATA hotswap, disabled Gigabyte bloatware installer, and disabled integrated graphics. It still took a couple of restarts, but I knew I was making headway because at least the splash screen was appearing whenever it rebooted. Thankfully, after about an hour of futzing around, Windows is back up and running. I am still not sure what the issue could have been. Seems CMOS corrupted itself? I'm glad it's up and running again and can only hope it's not a sign of issues to come. My fans were going 100% as the PC booted and then they'd all basically turn off as it hung on qcode 97, kicking back up to 100% at reboot. My best educated guess is that it corrupted itself during the restart, as it wasn't ever disconnected from AC power and there's also no way the cr2032 battery on the board is dead. Bloodplay it again fucked around with this message at 11:54 on May 5, 2023 |
|
#
?
May 5, 2023 11:24
|
|
|
Quote from Lisa Su on the Moore's Law debates going on (Jenson said its dead, Gelsinger said its not):Lisa Su posted:I would certainly say I don’t think Moore’s Law is dead. I think Moore’s Law has slowed down. We have to do different things to continue to get that performance and that energy efficiency. We’ve done chiplets—that’s been one big step. We’ve now done 3-D packaging. We think there are a number of other innovations, as well. Software and algorithms are also quite important. I think you need all of these pieces for us to continue this performance trajectory that we’ve all been on. So basically I guess her take is "Moore's Law isn't dead, but yes it is dead". Really a lot of the discussion seems to be people using "Moore's Law" to mean everything from the original transistor counts will double definition (which Intel is still using), to redefining it as ICs will continue to get faster through one pathway or another (AMD's definition apparently) or continue to get cheaper at a given performance level (Nvidia's definition apparently). It seems everyone fundamentally agrees that cheap and cheerful process improvements are over, though.
|
|
#
?
May 5, 2023 20:50
|
|
|
AMD's definition seems even vaguer, considering they're throwing software into the mix with that statement. Now it's just "computer performance will continue to improve in some capacity going forward." Which... cool? Very insightful.
|
|
#
?
May 5, 2023 21:15
|
|
|
I took the software mention to be a reference to software possibly needing to be changed in order to make the most of hardware improvements. We saw this with multi-core pushing people into threading, and cache-oblivious algorithms, etc.
|
|
#
?
May 5, 2023 21:43
|
|
|
Moore's Law is dead because Wright predicted it much more accurately long before.
|
|
#
?
May 5, 2023 22:29
|
|
|
Cygni posted:Quote from Lisa Su on the Moore's Law debates going on (Jenson said its dead, Gelsinger said its not): The bit about Nvidia 
|
|
#
?
May 5, 2023 22:49
|
|
|
The gently caress did "AMD Software Adrenalin Edition" or whatever the gently caress it was install itself here? I don't have a Radeon card, nor did I consent to it installing. --edit: At least the annoying Razer stuff, that was regularly triggered by that old DeathAdder, had the courtesy to ask whether to continue installing or not.
Combat Pretzel fucked around with this message at 23:27 on May 5, 2023 |
|
#
?
May 5, 2023 23:22
|
|
|
But you do have a Radeon GPU in your CPU. It's actually a really good software suite that somehow doesn't bog down your computer like everyone else's software suite, but I don't know why it's installing automatically on your system.
|
|
#
?
May 5, 2023 23:27
|
|
|
Combat Pretzel posted:The gently caress did "AMD Software Adrenalin Edition" or whatever the gently caress it was install itself here? I don't have a Radeon card, nor did I consent to it installing. If you have a 7000 series cpu, it comes with an igpu and that's what those gpu drivers are for.
|
|
#
?
May 5, 2023 23:27
|
|
|
Combat Pretzel posted:The gently caress did "AMD Software Adrenalin Edition" or whatever the gently caress it was install itself here? I don't have a Radeon card, nor did I consent to it installing. --edit: At least the annoying Razer stuff, that was regularly triggered by that old DeathAdder, had the courtesy to ask whether to continue installing or not. If you are annoyed by windows installing drivers you didn't ask for, you can fix that with settings -> search for "device installation" -> change to No.
|
|
#
?
May 6, 2023 01:11
|
|
|
Glofo wafer agreement rearing its ugly head again: https://videocardz.com/newz/amd-reportedly-resumes-production-of-ryzen-3000g-series
|
|
#
?
May 6, 2023 03:11
|
|
|
Wibla posted:The bit about Nvidia Nvidia believes (their definition of) Moore's Law is dead though, so they believe that future performance increases will require more silicon costs and higher prices. Which is what we've been seeing from them, so it tracks.
|
|
#
?
May 6, 2023 03:32
|
|
|
My interpretation: Moore's Law isn't dead, but it's in the hospital in the ICU on life support. The way that the semiconductor industry navigates the next few process nodes will determine whether or not it actually dies, or if it gets discharged from the hospital to go back into hospice care at home.
|
|
#
?
May 6, 2023 05:45
|
|
|
I have an idea that will raise yields, lower costs, and navigate around several of the physical limitations on what are more or less stagnant process nodes: chiplets. If only someone would come up with some kind of bus to link them together...
|
|
#
?
May 6, 2023 05:55
|
|
|
SwissArmyDruid posted:My interpretation: Moore's Law isn't dead, but it's in the hospital in the ICU on life support. The way that the semiconductor industry navigates the next few process nodes will determine whether or not it actually dies, or if it gets discharged from the hospital to go back into hospice care at home. the real question is what does the whole economy do when it can't just continue to throw money at the tech sector relying on exponentially increasing compute power, but this is probably the wrong place to ask that
|
|
#
?
May 6, 2023 09:16
|
|
|
Anime Schoolgirl posted:Glofo wafer agreement rearing its ugly head again: but wafer agreement is fulfilled by having glofo make all the I/O die on their 12nm node.
|
|
#
?
May 6, 2023 10:38
|
|
|
as said before, be ... extra careful with downloading bios updates, especially for msi motherboards https://twitter.com/tomshardware/status/1644350198276149248 https://twitter.com/PCMag/status/1654535238779895808 https://twitter.com/hn_frontpage/status/1654785723478732800
|
|
#
?
May 6, 2023 10:56
|
|
|
as if this platform wasn't loving cursed enough
|
|
#
?
May 6, 2023 11:05
|
|
|
Kibner posted:If you have a 7000 series cpu, it comes with an igpu and that's what those gpu drivers are for.
|
|
#
?
May 6, 2023 11:50
|
|
|
Cygni posted:Nvidia believes (their definition of) Moore's Law is dead though, so they believe that future performance increases will require more silicon costs and higher prices. Which is what we've been seeing from them, so it tracks.  (from https://www.techpowerup.com/274720/tsmc-achieves-major-breakthrough-in-2-nm-manufacturing-process-risk-production-in-2023) 3nm is something like 25-30% more expensive than 5nm iirc. Plus there's ugly news like this: https://www.phonearena.com/news/tsmc-us-chips-30-percent-premium_id147305 wargames posted:but wafer agreement is fulfilled by having glofo make all the I/O die on their 12nm node. Zen 4 doesn't use 12nm for the IO chiplet so AMD may need to make up the numbers by ordering other chips. They extended the wafer supply agreement in 2021, so it's a problem of their own making: https://www.anandtech.com/show/17132/amd-and-globalfoundries-wafer-supply-agreement-updated-once-more-now-21b-through-2025
|
|
#
?
May 6, 2023 12:27
|
|
|
ConanTheLibrarian posted:
are they using 7nm for the i/o die?
|
|
#
?
May 6, 2023 12:57
|
|
|
TSMC N6, so yeah basically.
|
|
#
?
May 6, 2023 13:20
|
|
|
Combat Pretzel posted:The gently caress did "AMD Software Adrenalin Edition" or whatever the gently caress it was install itself here? I don't have a Radeon card, nor did I consent to it installing. --edit: At least the annoying Razer stuff, that was regularly triggered by that old DeathAdder, had the courtesy to ask whether to continue installing or not. Most don't implement it, but just in case - which vendor + model do you have? It could also just be Windows Update, which you should be able to find in the log files, via Event Viewer: Applications and Service Logs\Microsoft\Windows\WindowsUpdateClient\Operational
|
|
#
?
May 6, 2023 13:41
|
|
|
|
| # ? Jun 3, 2024 11:44 |
|
|
New Gigabyte F8d bios today, but still on AGESA 1.0.0.6. It does explicitly mention the SOC voltage changes though.
|
|
#
?
May 6, 2023 14:48
|
|
