Bye Intel, hi AMD! I’m done after 2 dead Intels
(michael.stapelberg.ch)
from Pro@programming.dev to technology@lemmy.world on 07 Sep 08:12
https://programming.dev/post/37020007
from Pro@programming.dev to technology@lemmy.world on 07 Sep 08:12
https://programming.dev/post/37020007
cross-posted from: programming.dev/post/37019722
Comments
- Hacker News; - Lobsters.
threaded - newest
I built a new PC recently. All I needed to see were the benchmarks over the last 5 years. There’s currently no contest.
I went from Ryzen 1000 to intel 12000 since I need single threaded performance above all else (CAD). Plus it was a steal of a deal.
If Intel ever sorts out their drivers or it gets cheap enough I might for at 14000 chip but no further.
Interesting, so it’s not only their recent-ish (either 12th or 13th gen and up, iirc) laptop CPUs that die under normal load.
I’d never heard of arrow lake dying like raptor has been? wild.
Looks like they didn’t have adequate cooling for their CPU, killed it… Then replaced it without correcting the cooling. If your CPU hits 3 digits, it’s not cooled properly.
What if it hits around 90°C during Vulkan shader processing? 😅 Otherwise like 42–52 idle. How’s that? I’m wondering if my cooling is sufficient.
This is an AMD 9950X3D + 9070 XT setup, for reference.
Any way to do Vulkan shader processing on the GPU perhaps, to speed it up?
It’s fine, modern CPUs boost until they either hit amperage, voltage, or thermal constraints, assuming the motherboard isn’t behaving badly then the upper limits for all of those are safe to be at perpetually.
If you’re talking about the Steam feature you can safely turn it off, any modern hardware running mesa radv (the default AMD vulkan driver in most distros) should be sufficient to process shaders in real-time thanks to ACO.
How far back does that go? My AMD 6000 series GPU probably doesn’t need it, but what about my old laptop APU (3500U?).
What does it mean to “process shaders in real-time”? Wouldn’t it be objectively faster to process them ahead-of-time? Even if it’s only slightly faster while running the game?
I mean processing takes like a minute or so, so it’s no big deal. I’m just curious for the fun of it, if I can compile it on the GPU. Not sure it’s even possible.
Processing them as they’re loaded, quickly enough that there’s no noticeable frame drop. Usual LLVM based shader compilers aren’t fast enough for that but ACO is specifically written to compile shaders for AMD GPUs and makes this feasible.
Pre-compilation would in theory always yield higher 1% lows yes, but it’s not really worth the time hit anymore especially for games that constantly require a new cache to be built or have really long compilation times.
I think the one additional thing Steam does in that step is transcoding videos so they can be played back with Proton’s codec set but using something like Proton-GE, Proton-cachyos or Proton-EM solves this too.
Disclaimer: I don’t know how the deeply technical stuff of this works so this might not be exact.
Huh.
Well like I said it only takes like a minute with half of my 32 threads utilized at 100 % (so all of my cores I guess?). Might as well keep doing it I suppose.
AMDs 7000 series CPUs were designed to boost until they hit 95c, then maintain those temps. 9000 series behaves differently for boosting, but the silicon can handle it.
Okay cool, then I feel more confident. This is only my second build, ever, so I’m a little bit nervous. I didn’t buy any extra fans apart from the ones that came with my case. But I did get that beasty Noctua gen 2 air cooler, and it seems to be holding so far, even in the hot summer air.
Slight under volt, or upgrade cooler. 90c is too hot sustainably. Idle high 40s to 50s is not the best. Find a better air cooler or use a 240 AIO atleast.
Not true, 95C is the rated safe operating temperature
Thats what the manufacturer says but… 95c is damn near boiling at 203f. That is too hot to sustain any good longevity of a part, and any good workload for any component in a PC. That is a lot of heat. You will not get the best performance for a processor at its maximum temperatures running it like that all the time or even close to its max operating temp. I’m not saying you can’t hit that number but ideally you really really shouldn’t.
So what I said I think stands. Upgrade to a better air cooler and if need be a water cooler at least a 240AIO nothing smaller period. Keep temps lower and parts last longer. Performance boosts during core loads hold clocks longer. No question.
Near the boiling point of water, sure. I don’t have any water in my system, though. Unless there’s water in the cooling paste? I dunno.
Anyway, there’s a lot of saying this and saying that in these replies to my question without any links to references, so I think I’ll do some proper research instead. 😅
You got a better suggestion for an air cooler than this one that I have?
noctua.at/en/nh-d15-g2-lbc
Much appreciated.
The Noctua NH 15 is a great air cooler one of the best air coolers actually. What are you pairing it to? What CPU?
Almost all things noctua are good. However. They are expensive and if you’re not plotting on future cpu high TDP chips it’s overkill. I don’t honestly see the need to drop that much money on a CPU cooler. There’s near equal more cost effective options. Unless your doing major overclocks, cpu heavy loads at near max clock speed constantly where you think major wear and tear will happen I don’t for see the need to spend that much. I’ve built many many PCs.
You can find comparable air coolers much cheaper. Think 50 to 80 range and technically you can score them on eBay for cheaper than that. Look at the Phantom spirit 120 cheapest 30 to 40 usd, AK620 1 to 3c cooler than phantom 50 to 65usd, frost commander 140 80usd, noctua DH 15 150 to 180usd.
Of course prices fluctuate. Those get you the best ranked air coolers for the most part. Their all within roughly 5C of each other. All good coolers. Take the extra money savings and add to a GPU or whatever part you really need. I just seen you already have the noctua dh15 so in that case you were testing me to see what I knew. LOL. Enjoy you got a great cooler.
I wrote the CPU and GPU in the very first comment…
I already bought these components and built the computer already. 😄 It’s a pretty much maximum spec’d AMD system, and I’m privileged to buy the parts tax free through work, so I just went ham.
The only thing is that my 3900X in my previous build was running about 39–42°C idle, so I was worried I’d made some kind of error with the cooling paste or messed up the mounting of the cooler or something.
Definitely not; was genuinely a bit concerned, that’s all.
Your CPU isn’t made of water. Yes, this is safe to do. The manufacturer is on the hook for warranties if this goes wrong, and they know it.
The main concern would be lower quality electrolytic capacitors on the motherboard VRM, but they tend not to use low quality caps these days except maybe on budget boards.
The VRM temperature is a different sensor. You can have 69C VRMs when your processor is 95C
Irrelevant. If your CPU is chugging hard, then the VRM is chugging hard. That’s what causes high VRM temps.
Not really, my CPU is a 65W CPU. When it chugs hard the VRM is not even warm. My cooler is much weaker
The processor is not made out of water so the boiling point of water doesn’t affect it
In fact you will get the best performance at 95C because to decrease it means to decrease the clocks
If you mean to get a better cooling solution that would be better, but then running it at 95C with that better cooling solution would be better again
Once you reach a cooling solution that can take it to the maximum clocks, then there will be no performance benefit between 70C and 95C at the same clocks and you can reduce the fan speed to hit the higher temperature at less noise
I did some quick research.
I feel confident now. My CPU and cooling seems perfectly nominal. 😌👌
If your running that CPU then you almost certainly have done research or have money. Either way. Enjoy the setup you’ve got a good cooler and CPU it seems.
That’s not the case. 100% for new CPUs, but also for old ones too.
My father’s old CPU cooler did not make good contact, got lose in one corner some how, and the system would throttle (fan at 100% making noise and PC run slow). After i fixed it, in one of my visits, CPU was working fine for years.
System throttles or even shuts down before any thermal damage occures (at least when temperatures rise normally).
Pretty much anything with a heat spreader should be impossible to accidentally kill. Bare die? May dog have mercy on your soul.
If your CPU hits 3 digits, therm throttling isn’t working properly, because it should kick in before it hits that point.
The article (or one of the linked ones) says the max design temperature is 105°C, so it doesn’t throttle until it hits that.
Which makes me think it should be able to sustain operating at that temperature. If not, Intel fucked up by speccing them too high.
I’d expect it to still throttle before getting to 105C, and then adjust to maintain a temp under 105C. If it goes above 105C, it should halt.
Then you misunderstand the spec. That’s the max operating temperature, not the thermal protection limit. It throttles at 105 so it doesn’t hit the limit at 115 or whatever and shut down. I can’t find a detailed spec sheet that might give an exact figure.
The chip needs to account for thermal runaway, so I’d expect it to throttle before reaching max operating temperature and then adjust so it stays within that range. So it should downclock a little around 90C or whatever, the increase as needed as it approaches 105C or whatever the max operating temp is. If it goes above that temp, it should aggressively throttle or halt, depending how how far above it went and how quickly.
Again, you misunderstand. The max operating temperature is where Intel has stated that the CPU can safely operate for extended periods of time, including accounting for situations like thermal runaway (though ideally they engineer the chip that that doesn’t happen in the first place).
If that situation does occur, the chip attempts to throttle at 105, and if that fails then it presumable halts at whatever the protection threshold is before it hits the actual damage point, as I said.
Interesting, so it only throttles at that temp? That’d a bit different than how AMD handles it IIRC, which think stops boosting around 80C or so and throttles around 90C, and the max operating temp is closer to 100C.
Yes.
Whether Intel fucked up by saying “oh yeah works great up to 105” if that isn’t actually true is another question, as I mentioned.
It’s not really that different, the exact temperatures are slightly higher but most intel processors will boost up to 105C, then start throttling to maintain that 105C as a maximum, and if that’s not possible they’ll halt at 110C.
AMD does the same, just the temps are (for the one specific CPU I remember them for) 80-85C for starting dialing down the boost, 90C for throttling below the normal freq, and 95C for TjMax which either halts the system or just drops the power usage so low it doesn’t matter - I’m not about to take a heatgun to my CPU to see what it does as it wasn’t capable of hitting that on its own.
But it shouldn’t be possible to break your CPU from over temperature, no matter what those temps are, because they should be capable of protecting themselves, even if that means dropping to 386 speeds when you are running them in the Death Valley with not cooler whatsoever.
The 7000 series had the Intel behavior of just clocking up until like 95C and staying there indefinitely
That’s why people thought the 9000 series was disappointment - AMD went back to balancing power efficiency and performance
No it shouldn’t slow down at 90C, it should clock up until it can sustain exactly 105C and stay there. That’s the optimal performance point.
Why? It’s designed to run up to 105c.
I think it was when AMDs 7000 series CPUs were running at 95c and everyone freaked out that AMD came out and said that the CPUs are built to handle this load 24/7 365 for years on end.
And it’s not like this is new to Intel. Intel laptop CPUs have been doing this for a decade now.
CPUs should throttle as they approach the limit to prevent thermal runaway. As it gets closer to that limit, it should adjust the frequency in smaller increments until it arrives at that temp to keep the changes to temps small.
105c is the max operating temperature. It’s not going to run away the second it hits 106.
Your CPU starts throttling at 104c so that way it almost never hits at 105c for long If it can’t maintain clocks then it drops them until 104c can mostly be maintained.
If you have an improperly mounted cooler, you could very well get to 105C incredibly quickly, and 115C or whatever the halt temp is shortly after.
My intel mac’s cpu (i5-5250U) throttles to maintain 105 C
laughs in 8700k
When I overclock this old chip (which it was built for) it can hit over 100 with proper cooling. Some chips are hot as fuck. I think this one shuts off at 105.
damn… its sitting at 301 Kelvin currently…
Somehow I figured out Intel was shit early on. Been AMD for like 15-20 years. I think it was a combo of childhood shit computers running Intel, and a lot of advice pointing out what garbage it was and not worth the cost for PC builds.
Similar reasons I hate Hitachi and Western Digital hard drives. They always fucking fail.
I was in team AMD in the 2000s for two reasons: price and competition to Intel. Intel had a massive anti-trust loss to AMD around that time, and I wanted AMD to succeed. I stuck with them until Zen was actually competitive and stayed with them ever since because they actually had better products. Intel was the king in both performance and power efficiency until that Zen release, so I really don’t know where that advice would’ve come from.
As for Hitachi and Western Digital, WTF? Hitachi hasn’t been a thing for well over a decade since they sold their HDD business to WD, and WD is generally as reliable or better than its competition. It sounds like you were impacted by a couple failures (probably older drives?) and made a decision based on that. If you look at Backblaze stats, there’s not a huge difference between manufacturers, just a few models that do way worse than the rest.
You misspelled Seagate.
My WD drives have been great, but my Seagates failed multiple times, causing data loss because I wasn’t properly protecting myself.
All manufacturers have bad batches. Use diversity and keep backups.
How likely is it that I got 4 to 5 bad batches over the space of as many years?
Raid and offline backups these days, I eventually learned my lesson. One of which is stay away from Seagate.
Within the realm of possibility. Especially if you treat them harshly (lots of start-stop, and low airflow and high temps). Backblaze collects and publishes data, and the AFR for Seagate is slightly higher than other manufacturers, but not what I’d consider dangerous.
Seagate has more than bad batches. When every single one of their 1tb per platter barracuda drives have high failure rates then that’s a design/long term production issue.
15-20 years is silly. Intel was the clear leader for a long time before Ryzen in 2017, and arguably a few years after that too.
Yuuup! The last time AMD was better before Ryzen was the Athlon 64 era.
Ah ha ha. I had my second ryzen die yesterday in a row. No load, no overclocking, just in the middle of coding. Fack AMD and fack Intel. I’m gonna go buy a Mac Mini.
Those M chips ARE pretty amazing.
Probably a bad motherboard then. CPUs generally don’t just die, unless there’s some kind of excess voltage or something. If you weren’t aggressively overclocking, that sounds like the mobo isn’t doing a great job at controlling voltage. It could also be a bad PSU, the CPU is the last thing I’d suspect on the second failure.
Boards are different, Asus and Asrock, power supplies cheap Zalman and expensive DeepCool. It doesn’t matter. It’s not supposed to happen! And it has never happened before, until they started making some wild voltage controls.
CPUs don’t die very often without something being very wrong with your system.
Could be the PSU or motherboard
Most likely your motherboard. Asus and ASRock (they are related companies) have been fucking up recently, so if you use one of them it’s likely that.
It just happens, it’s a lottery
I knew Michael Stapelberg from other projects, but I just realized he is the author of the i3 Window Manager. Damn!
“Do you need to transcode video?
Then leave Intel the fuck alone.”
Been my rule for 20 years, and it’s worked good so far.
It’s odd, their GPUs are doing fine, a market they are young in, but their well established CPU market is cratering
Business majors suck.
they always did, even back in college.
Their GPU situation is weird. The gaming GPUs are good value, but I can’t imagine Intel makes much money from them due to the relatively low volume yet relatively large die size compared to competitors (B580 has a die nearly the size of a 4070 despite being competing with the 4060). Plus they don’t have a major foothold in the professional or compute markets.
I do hope they keep pushing in this area still, since some serious competition for NVIDIA would be great.
Sure, if by doing fine you mean looking alright in benchmarks while having zero supply because they don’t make money selling them and thus don’t want to produce them in any significant amount.
They are in stock near me.
yeah quicksync is the only reason i put an intel in my NAS.
I transcode video with nvenc, I don’t need Intel
Can nvenc do dual pass encodings these days?
I honestly don’t get why anyone would have bought an Intel in the last 3-4 years. AMD was just better on literally every metric.
Idle power is the only thing they are good at, but for a homeserver a used older cpu is good enough.
Was that even true for comparable CPU’s? I feel this was only for their N100’s etc.
Nah all the am4 cpus have abysmal idle power, the am5 got a little better as far as i know but the infinity fabric was a nightmare for the idle power.
Well I concede, I guess there was one metric they were better at. Doing absolutely nothing.
What about the monolithic socs? Idle power on my 5825u is better than my i5-1135g7. My i7-8550u is in between.
The problem is just the infinity fabric that connects the chiplets. Afaik this is because its clock runs at a consstant speed and its made on a way bigger node. The monolithic parts dont have that problem.
My AM4 box draws 0 watts on idle. I turn it off when I don’t use it
Ahh yes idle=not using computer and using computer=100% cpu usage. There is no such thing as “reading a pdf document” or “surfing the web”.
Browsing the web is not idle
Enlighten me! When is a cpu idleing?
When it’s not doing anything
If the system is powered the cpu is always doing something. Idle doesnt mean that your pc is off it just means no “real” load. Modern cpus are so powerfull that browsing the web is no real cpu load. It may be a few % difference than just the os itself but windows backround tasks will propably make more impakt on the cpu load.
My CPU boosts to 4Ghz at 1.5V loading a website
That’s hardly nothing. When it’s not doing anything it goes back to 500Mhz at less than 1V
The CPU is not doing anything when the system is turned on! I don’t run Windows spyware so it only does something when it’s told
Yea exactly, you load a page your cpu will boost for 1% of the time you spend on that page while loading it and then idle at a few Mhz for your os. I dont get the “it doesnt idle”. With a slim linux system it propably idles a lot more than with win11.
So clearly it will be using much more power when it’s loading the site. These days as you scroll down the page it will infinitely load more stuff. Can’t call that idle
Dude go measure your cpu usage while scrolling a site i can watch a high res video with cpu usage under 10% measuring the difference between no active program and only the os is nearly impossible. Why dont you go and log your cpu power for doing this vs only your os running and report back? If your cpu usage keeps being above say 25% while loading a normal site without funky stuff going on, something is weird. Also keep in mind that even if one of your cores is boosting for a second you cpu usage as a whole over time is still low and low power draw at low clocks will benefit you.
The CPU usage is 10%, but the energy usage is NOT 6.5 watts (with 65 watts being 100% usage)
That’s not how CPU usage works
First energy usage is not watts, second you can look via tools like hwinfo or others how much power your cpu draws. Where did i say it would draw 6.5w at 10%?
Just checked the laptop first:
It’s 9.8w on idle (including laptop screen and RAM and everything) and 16.8w scrolling a website
That’s 50% more power usage. Over 20 seconds up from 196 J to 316 J
Like turn off the PSU switch? Computers can draw more than you think when they’re “off”, but the PSU switch should shut out the whole thing.
If your use case benefited from Quicksync then Intel was a clear choice.
Older Intel CPUs are the only ones that can play 4K BluRays on the player itself and not just ripping to a drive. Very niche use case but that is one I can think of.
They can’t even do that anymore. sgx had a bunch of vulnerabilities and as a result, that service has been disabled.
sgx.fail SGX.Fail
Well shit
This section contains his reasoning.
It was ok until he said the AMD chip consumed more power. It is a X3D chip that is pretty much a given, if he’d gone for a none X3D chip he’d have saved quite a bit of power especially at idle. Plus he seems to use an AMD chip like an Intel chip with little or no idea how to tweak its power usage down
I’ve got a 9700X and it absolutely rips at only 65W
Same here, that thing fucks and stays very cool doing it
As I said - a bit of knowledge is a dangerous thing
Just for interest. Why did you buy Intel in the first place. I don’t know about many use cases where Intel is the superior option.
Better energy efficiency overall.
Other than that - maybe some habit.
They have better idle efficiency, but if your PC is not on and idle 24/7 for some reason, AMD is more efficient.
I find it funny that they mention noise level, as if the CPU itself were making noise. I’ve bought silent fans all my life, separately from CPUs.
I’d probably just warranty the CPU and assume it was a defect instead of blame the entire company.
But yeah amd is the better choice for everything atm except x86 power efficiency laptop chips.
care to explain that last part?
for example if I want to by a new laptop right now, do I buy intel? because it has better power usage? is that only idle and/or lower usage power? what about if I just plus the laptop to power socket always (like a simple desktop), then amd is better with respect to performance per watt?
Intel has its core ultra line up which improved its power usage a lot. This isnt great for intensive workloads because there is more efficeny cores and less preformance cores but its great for general everyday use on battery.
When you’re plugged in power usage no longer matters so having an Intel CPU sucks because yoy are getting less preformance for your dollar.
Also amd is more price to preformance and still has good battery life so you really do get the most value.
I get the intel lower power when not doing stuff (wish amd had high/low config for cores too) but what I mean was, in laptop cpus that are not on battery (just connected to power) does amd do more with the same power usage?
if the comparation can not be done with the same gen cpus from two companies, then maybe a similar power usage cpu from amd and one from intel (laptop of course), do they for example have similar geekbench benchmark results? (for lack of better tool)
so what I am asking is I dont care that is the same gen amd and intel laptop cpu with both connected to socket power, if amd is better. I want to know if for the same power usage (not idling but working) amd is better or not.
I havent seen the exact data on this but i would say yes. Amd current gen CPUs are very strong right now preformance wise especially when directly compared against intel.
Which is fine. Potentially part of the huge known issue with the last couple of generations of Intel chips which affected a huge swath of CPUs, fixes have been released, but damage has been done - that alone would make me dubious about them going forward.
The more immediate issue though is, my CPU failed, I need to find some time to take the PC apart, safely box up the the CPU, figure out the intel rma procedure, ship it off, wait for intel to assess the cpu, hope they accept responsibility, ship me a new CPU and then find the time, once again, to take the PC apart to put the CPU back in. Twice. And I’ve been without PC for the entire time. And they most likely knew about the issues before the second gen of defective chips they launched. And it’s not even the better chip as you mention. I’d be sufficiently pissed off to stay away.
I was an intel guy most of my life, Intel on all the hand-me-downs I got from my grandfather’s appliance store, Intel on my first gaming PC in 2008 til 2012, Intel on the 2012-2019 PC, it wasn’t until I built my current PC in 2019 that I Switched because of the Meltdown / Spectre / Etc issues, largely just out of reputation not actually understanding them.
Sufficed to say, I left in 2019 and have had no reason to return.
The computer I bought should last me about 10 years. I spent a fuckload of money on it. The next comp will have to be done entirely with as little starting google and privacy violating shit as possible.
And I am certain AMD will make better stuff by then.
I’m still rocking an i7 4790k and its >10 years old! Judging from the other comments it seems the intel issue is more of a recent one though. If I ever configure a new PC, I’ll check out AMD for sure.
I’m still using the same chip on an Asus mobo. No problems here.
I’ve swapped back and forth between brands since I built my first computer almost 30 years ago. It was intel forever until AMD showed up with their early Athlons, amazing CPUs for the price. Then Intel fought back with with their Core 2 Quads, AMD with Thunderbirds, back to intel with their higher i-series, up until about 2-3 years ago and now AMD’s Ryzen offering the better performance/$ again. It’s too bad intel seems to be unable to keep costs competitive and maintain quality. I’ve never had a CPU quit on me yet (knock on wood). Motherboards, RAM, PSUs, sure. I used to partial upgrade every 2 years or so, but the golden era of PC building is gone. The high prices of GPU’s alone really killed the momentum we had from say ‘05-‘15.
Intel’s strategy seems to be just chugging power into the CPU and hoping for the best.
It feels kinda like there’s a race and one person’s breathing hard and sweating bullets only to have another runner breeze past them like it’s nothing.
I know this is sort of still doable with aliexpress kits, but I miss the days of being able to make “weird” builds. My first build was an Athlon XP-M 2500+. It was a mobile chip that was just a binned desktop chip. It used the same socket as desktop, had no IHS, and ran at a “lower voltage” thanks to the binning. Overclockers DREAM in back in like 2005.
It was a sign for me when Amazon sent me two AMD chips. Thank you!