Intel says their 13th/14th Gen instabilities are from 'elevated operating voltages stemming from a microcode algorithm'

[Xiee]

On 8/5/2024 at 3:57 PM, Mark Kaine said:

10 years ago i was rocking a Athlon 64x2, dodged a bullet there!

Same. I actually had 2 of those. One which was a personal PC and another which I used as a file server for our internet cafe.

[Bitter]

2 hours ago, Vishera said:

Anyway the single core performance was terrible and you would have been better off with a 2600K at a significantly lower price.

That was always the story with AMD until about Zen 2.

[Dabombinable]

5 hours ago, Vishera said:

The thing is that the FX 9590 was just an overclocked FX 8300 and only 1% of the motherboards in the market had the VRMs to drive that power hog.

Anyway the single core performance was terrible and you would have been better off with a 2600K at a significantly lower price.

That may be, but that didn't stop the price from being absurdly high.

[Mark Kaine]

So what do we think will the outcome of all this be? i mean i know this is about the technical side, but it seems unfixable for most part. so huge lost in trust, government intervention with trillions to "save intel", the usual?  

 

AMD raising prices...? oof.

 

or will it just blow over and people will continue to buy "i7s and i9s" (are they even still called like that lol)?

 

 

[starsmine]

1 minute ago, Mark Kaine said:

So what do we think will the outcome of all this be? i mean i know this is about the technical side, but it seems unfixable for most part. so huge lost in trust, government intervention with trillions to "save intel", the usual?  

 

AMD raising prices...? oof.

 

or will it just blow over and people will continue to buy "i7s and i9s" (are they even still called like that lol)?

 

 

I dont know what you mean fixable. or even needs fixing. its not a problem in meteor lake, it shouldnt be a problem in lunar or arrow lake. 
post fix, we can go back to recomending new 13/14th gen, but yea the used market it trashed. 

[Mark Kaine]

10 hours ago, Bitter said:

That was always the story with AMD until about Zen 2.

generally i think this has always been over exaggerated, just because intel was faster in comparison didn't mean amd was actually slow , and as we know intel speeds always came at a cost (non existant security checks , power draw, etc)

 

yeah yeah, i know people always complain about amd too , but my dual core athlon was fine really,  didn't know im missing something! 

[Mark Kaine]

8 minutes ago, starsmine said:

its not a problem in meteor lake, it shouldnt be a problem in lunar or arrow lake. 

oh yeah, you say that now, but will people believe it...?

 

on other hand, as said,  it might be an EA situation where people say they "boycott" something but actually never go through with it... gotta get that next "super hot" intel for sure (i just think that trust is broken, to me they're profoundly in EA territory now - haven't bought an EA game in 20+ years...)

 

ps: besides the cpus themselves what needs "fixing" is trust... people now know intel most likely isn't fulfilling warranties, why take the risk, amd just works (until now)

[starsmine]

4 minutes ago, Mark Kaine said:

oh yeah, you say that now, but will people believe it...?

Because the microcode to fix it is out and the processes are now in the books written in blood. 
Any problem they have with those chips will be new unique problems.

4 minutes ago, Mark Kaine said:

on other hand, as said,  it might be an EA situation where people say they "boycott" something but actually never go through with it... gotta get that next "super hot" intel for sure (i just think that trust is broken, to me they're profoundly in EA territory now - haven't bought an EA game in 20+ years...)

I dont think anyone is calling for a boycott of intel here. Its just not recommending people buy the parts until a fix is in.

[ImorallySourcedElectrons]

3 hours ago, Mark Kaine said:

So what do we think will the outcome of all this be?

Intel expanding their pre-production testing sample size and adjusting the associated procedures. Looking at the incidence rate, this looks like something that you could miss in pre-production testing.

 

Like I know we sure did miss a few things like this in our products at work when the financial thumb screws were tightened on our testing budget and management was exerting pressure to release yesterday. The end result was that statistics took a backseat and some idiots on the team just went along with whatever middle management decided.

 

So I'm genuinely interested in hearing what the internal story is about this one once the dust settles.

[Kisai]

17 hours ago, Bitter said:

That was always the story with AMD until about Zen 2.

Not quite. AMD had to shed a terrible track record of utterly crap chips between the 486 era and the Slot A Athlon. Then they stumbled a few times in trying to catch up to Intel. The Phenom and FX processors were all terrible, either in design, performance or cost.

 

Like no offence to AMD die-hards, but the only time I even considered an AMD CPU was with the Zen 2, and even then, Microsoft and OEM MB manufacturers had to screw that one up.

 

Heck my mom has an Intel 7th gen desktop and a 2nd gen desktop, and I have a 7th and 9th gen laptop and that TPM requirement absolutely blindsided people still using Windows 10 on perfectly usable hardware. It makes it risky to pick an AMD part unless you are aiming for a mid or high end part, because you might just end up with the MB OEM cheaping out out the chipset features or BIOS support.

 

[Monkey Dust]

8 hours ago, Mark Kaine said:

So what do we think will the outcome of all this be? i mean i know this is about the technical side, but it seems unfixable for most part. so huge lost in trust, government intervention with trillions to "save intel", the usual?  

 

AMD raising prices...? oof.

 

or will it just blow over and people will continue to buy "i7s and i9s" (are they even still called like that lol)?

 

 

AMD have already announced prices for the new 9000 chips, they are lower than the 7000 were at launch. I guess they may have locked in prices with the retail chain before Intel's issues were as well understood.

 

I think they might smell blood, and rather than push prices up, go for market share. Prices may go up in the long run, but with ARM getting established in the Windows ecosystem, turning it into a 3 horse race, maybe not.

 

For Intel's part, I think this will be mostly forgotten by all but the tech community in a few months. The big risk is that with all the job loses, something like this is more likely to happen again.

[IseeFractals]

On 8/6/2024 at 11:37 AM, porina said:

How are you measuring this? One measure of company size is market cap, or what the stock market thinks a company is worth. Not saying it is a good measure, but it is one none the less. AMD are almost double Intel currently at 218B vs 115B. And this isn't only due to the recent troubles. AMD have been ahead of Intel on that measure for a long time. Then you get nvidia blowing both out of the water at 2.5T.

 

Other methods could be used, revenue, profits, head count?

Headcount, a comparison i was making based off of a video i recently saw in which it stated intel has 130,000 employees while AMD has just shy of 11,000. Intel count was correct, but AMD is actually around 26,000, so my bad, it was late and i was too lazy to look it up. Even so that's a massive disparity in workforce given that AMD currently has double the market cap of Intel.

[ImorallySourcedElectrons]

11 minutes ago, IseeFractals said:

Headcount, a comparison i was making based off of a video i recently saw in which it stated intel has 130,000 employees while AMD has just shy of 11,000. Intel count was correct, but AMD is actually around 26,000, so my bad, it was late and i was too lazy to look it up. Even so that's a massive disparity in workforce given that AMD currently has double the market cap of Intel.

Intel also runs a massive fab operation, which requires tens of thousands of employees, which AMD outsources. If this is the right choice is a discussion in and of itself, but overall I'm tending towards keeping fabs in-house being better as it allows the different parts of the company to support each other. So you'd have to compare the headcount with TSMC and Samsung Electronics, both of which are monster-sized companies (respectively near 100k and 300k if my memory serves right). And even then, Intel also produces way more than just CPUs and GPUs. For example, Intel is a major manufacturer of networking chipsets, something that AMD does not really have.

 

So that employee disparity makes a whole lot of sense, actual manufacturing operations usually take quite a bit of staff. To illustrate, you want that fab running 24/7? Congrats, you just earned yourself the need to hire the amount of workers you need to staff it times four (3 shifts per day + weekend coverage) with an additional 10 to 20% overhead to account for organisational and administrative tasks, and the need to cover folks who are sick and on holidays. For example, Intel cites they have about 4000 people working in their fabs in Israel. But if you then consider the fact that they probably have globalized their support roles and run 15 fabs in total, you can probably say that around 70k of those folks are working in the manufacturing leg of the company.

 

So before you pull out your boardroom executive stock valuation math, realise that it doesn't really explain why a company might have such a large headcount.

[IseeFractals]

On 8/6/2024 at 9:56 AM, BlueChinchillaEatingDorito said:

And here we thought 12th gen was something they could build on... 

 

Ehh AMD's period of mediocrity was also over a decade. They thought throwing more cores at the problem with the entire Phenom, Phenom II, and FX line would close the gap but more cores don't help when each core was unimpressive. Even overclocking when gains were significant back then didn't tip the scale in their favour. Sort of like how we got 8 core chips on Android phones early on yet the experience was still as clunky. 

That's just not true, that's just the narrative that intel's marketing and misinformation department painted. AMD had great x64 performance, intel...did not, and part of the pain point for AMD was that despite Vista being built around AMD cpu's, with the intent of it being exclusively 64bit....intel through a hissy fit because they didn't have anything that met the minimum specs (nor did Nvidia have a DX10 card capable of handling Aero) so microsoft was quite literally forced into releasing a cut down 32bit version of Vista....32bit vista sucked as a whole, but 32bit vista paired with x64 AMD chips sucked more....but the simple action of taking the same AMD hardware, and running 64bit vista brought about a 50% performance increase. While intel focused on prolonging 32bit, AMD was pushing forward, but unfortunately consumers stuck with what they knew and intel spent a shitton of money on convincing them to continue doing so. 

The Phenom ii quadcore's were still competitive when they launched, and i know that because i was the first person to leak their performance numbers way back when. AMD didn't truly begin releasing "medicore" products until bulldozer, which launched in late 2011 (and even those weren't as bad as everyone makes them out to be in terms of performance, though power consumption was absolutely terrible) First gen ryzen launched in March 2017, by which point intel was 3 years deep in their 14nm++++(++++).  Bad sales don't necessarily mean bad products, and good sales don't mean good products, just look at how well pentium 4  sold.3

[IseeFractals]

3 minutes ago, ImorallySourcedElectrons said:

Intel also runs a massive fab operation, which requires tens of thousands of employees, which AMD outsources. If this is the right choice is a discussion in and of itself, but overall I'm tending towards keeping fabs in-house being better as it allows the different parts of the company to support each other. So you'd have to compare the headcount with TSMC and Samsung Electronics, both of which are monster-sized companies (respectively near 100k and 300k if my memory serves right). And even then, Intel also produces way more than just CPUs and GPUs. For example, Intel is a major manufacturer of networking chipsets, something that AMD does not really have.

 

So that employee disparity makes a whole lot of sense, actual manufacturing operations usually take quite a bit of staff. To illustrate, you want that fab running 24/7? Congrats, you just earned yourself the need to hire the amount of workers you need to staff it times four (3 shifts per day + weekend coverage) with an additional 10 to 20% overhead to account for organisational and administrative tasks, and the need to cover folks who are sick and on holidays. For example, Intel cites they have about 4000 people working in their fabs in Israel. But if you then consider the fact that they probably have globalized their support roles and run 15 fabs in total, you can probably say that around 70k of those folks are working in the manufacturing leg of the company.

 

So before you pull out your boardroom executive stock valuation math, realise that it doesn't really explain why a company might have such a large headcount.

Intel's Fabs account for less than half their workforce, so even if you want to make that argument they're still three times the size of AMD, and the employee count of TSMC and samsungs fabs aren't really relevant, when they're producing chips for....pretty much everyone producing anything that uses a semiconductor. 

But since you bring up fabs...intel hasn't really been doing to great on that front for....about a decade, have they? Sure they can churn out network adapters without issue, but when it comes to things like CPU's and over the past few years, GPU's, those fabs haven't been producing great results. 14nm ++++++++, GPU's they had to outsource production of, and of course the fact that every chip they've produced for the past few years is at imminent risk of catastrophic failure. 

What's their solution to that? Cut the advertising and marketing departments? No, of course not.....let gut R&D! Because it's not about the product you're selling....but how you sell the product!  

 

[ImorallySourcedElectrons]

9 minutes ago, IseeFractals said:

Intel's Fabs account for less than half their workforce,

Citation required, because if that's true, Intel is way below the average number of people you require to run a fab of those magnitudes. Also most of those figures only count direct workers and don't account for the massive amount of support staff you need to run a fab. I don't think you quite understand how much folks you need to keep those gears running, you're easily looking at 5000 - 10000 people per fab to keep a major fab like that running if the numbers presented at industry conferences are anything to go by.

32 minutes ago, IseeFractals said:

 and the employee count of TSMC and samsungs fabs aren't really relevant, when they're producing chips for....pretty much everyone producing anything that uses a semiconductor. 

Hilariously wrong take on this one, the reason why I put them up against Samsung and TSMC is that they're the only ones creating the same category of product. Otherwise I'd be comparing to the likes of SMIC, Micron, Kioxia, SK Hynix, UMC, etc. While Samsung and TSMC are absolute industry juggernauts, they have plenty of competitors to deal with for less challenging logic chips.

50 minutes ago, IseeFractals said:

But since you bring up fabs...intel hasn't really been doing to great on that front for....about a decade, have they? Sure they can churn out network adapters without issue, but when it comes to things like CPU's and over the past few years, GPU's, those fabs haven't been producing great results. 14nm ++++++++, 

As someone who actually has been involved in chip tape-outs over the last years, I strongly disagree with you. Want me to be blunt with you? All the major suppliers are pretty much par for the course for most practical chip designs, you're confounding system architecture of a chip (which you seem to be fanboying for based on some of your previous posts) with process node technology. At several points over the last decade, Intel was ahead of Samsung and TSMC for number of transistors they could squeeze into a given unit of surface area, but the actual density you can achieve for a practical design is driven by entirely different factors (actual layout, thermals, achievable "wiring" densities, etc.) The electronics industry operates on razor thin margins, you just go for whoever's the cheapest.

 

Another example of how the internet and investor crowd tends to oversimplify the comparison is the yield aspect, it's not nearly as black and white as a lot of the armchair analysts make it out to be. If I can run my older steppers at far higher throughputs, the WAN show-based tax write-off logic aside, I can actually accept a lower yield and write off a significant portion of the wafers because I didn't have to fork over a quantity of money at risk of a Disney duck infestation to upgrade my production equipment.

 

And we could go on for a while, but a lot of "zomg TSMC is so far ahead of everyone else" is quite misconstrued. And if we wish to talk CPU architecture, that's an entirely different conversation, but there both sides are doubling down on a set of design decisions that might leave them cornered later on.

1 hour ago, IseeFractals said:

 GPU's they had to outsource production of, and of course the fact that every chip they've produced for the past few years is at imminent risk of catastrophic failure. 

Outsourcing production is common, everyone in the industry does it - even if they can do it themselves, Intel didn't because they just wanted to keep their internal order books of their fabs filled. They're now making different economic assessments most likely, if TSMC can do the same thing cheaper than you can yourself, and you can then fill your order book with a customer who pays full sticker price, what would you do?

1 hour ago, IseeFractals said:

What's their solution to that? Cut the advertising and marketing departments? No, of course not.....let gut R&D! Because it's not about the product you're selling....but how you sell the product!  

Financial ghouls gonna ghoul, no argument there.

[leadeater]

49 minutes ago, ImorallySourcedElectrons said:

At several points over the last decade, Intel was ahead of Samsung and TSMC for number of transistors they could squeeze into a given unit of surface area, but the actual density you can achieve for a practical design is driven by entirely different factors (actual layout, thermals, achievable "wiring" densities, etc.)

It's only recently TSMC has been seen and talked about as the industry benchmark for silicon fabrication. Otherwise it has always been Intel for all intents and purposes, only time not was when we were literally discovering how to make silicon logic chips.

 

Even then density wise Intel was ahead of TSMC even when people were talking about it the other way around, node names became completely meaningless in a very short period of time to the benefit of TSMC and only TSMC.

 

As you are quite aware the things that matter is transistor density and types i.e. logic, memory/cache, I/O etc and even that is still surface level comparisons. You have all sorts of factors like leakage characteristics and voltage characteristics, wire accuracy and precision (effectively yields), DUV or EUV, how many EUV layers. Lots of stuff way out of my depth to evaluate between all the market players, I just know for decades if you wanted to look at the best then you were looking at Intel.

 

Also actual fabricated dies of actual designs are not done at maximum possible density of the process node either and it's not uniform across the die.

 

The entire industry is dependent on ASML anyway and what they can do and their capabilities, it's a lot like F1 rules around engines. Who can do the best within the same "rules and regulations" or the same equipment.

[Kisai]

4 hours ago, IseeFractals said:

What's their solution to that? Cut the advertising and marketing departments? No, of course not.....let gut R&D! Because it's not about the product you're selling....but how you sell the product!  
 

Chances are they gutted mostly software engineers in the "AI" department, along with marketing.

Like scratching my head to figure out where they would let go of people, unless they are going to abandon their Israel fabs(22nm,14nm,20nm)/R&D for political reasons, it doesn't really make sense to just go "alright supervisors, decide which 1/8th of your team is the weakest link, and give them the pink slip"

 

You don't automatically get more more work out of existing staff when you let people go unless those people were literately doing nothing and the work was being carried by the existing staff anyway. But then you end up with people doing the jobs of two and three people who then just want to quit ASAP because of the miserable working conditions.

 

You know how petty these cuts are? They cut free fruit.

 

I'd love to see the CEO and board take a 15% paycut to "make some sacrifices" too. Why do CEO's get to keep their paychecks when they let go of so many jobs, they should be cutting their own pay by the same amount of headcount lost.

[ImorallySourcedElectrons]

1 hour ago, leadeater said:

It's only recently TSMC has been seen and talked about as the industry benchmark for silicon fabrication. Otherwise it has always been Intel for all intents and purposes, only time not was when we were literally discovering how to make silicon logic chips.

 

Even then density wise Intel was ahead of TSMC even when people were talking about it the other way around, node names became completely meaningless in a very short period of time to the benefit of TSMC and only TSMC.

 

As you are quite aware the things that matter is transistor density and types i.e. logic, memory/cache, I/O etc and even that is still surface level comparisons. You have all sorts of factors like leakage characteristics and voltage characteristics, wire accuracy and precision (effectively yields), DUV or EUV, how many EUV layers. Lots of stuff way out of my depth to evaluate between all the market players, I just know for decades if you wanted to look at the best then you were looking at Intel.

 

Also actual fabricated dies of actual designs are not done at maximum possible density of the process node either and it's not uniform across the die.

Yeah, the entire discussion is moot when it comes to naming, and most folks also forget the peak density numbers are for highly repetitive memory structures, or for some types of simple highly repetitive circuit (e.g., oscillators and delay lines) used for technology characterization. And even within a process node as the computer enthusiasts might recognise, there might be sub-divisions that they're not necessarily aware of because you rarely see full documentation released. For example, check what STM offered at one point (I think it's somewhat out of date though):

(Source: https://europractice-ic.com/technologies/asics/stmicroelectronics/ - an overview from a MPW intermediate)

 

So I totally agree, I think it's way out of anyone's depth to really compare the general purpose digital logic technologies at this point, and all of them are somewhat related to each other anyway. I suppose there might be a few folks left at places like imec, Fraunhofer, some of the foundries, and the typical MPW intermediates who have enough technical know-how and access to actually compare technologies, but I know we sure as hell didn't. Our main worry was hitting the necessary qualification check boxes, and that's one of the areas where TSMC really shines. And the costs involved can also be really mindboggling in odd ways, like I know of a case where more money was spent on Gel Paks (a type of packaging used for bare dies in some instances) than on the actual foundry costs.

 

So for us online to go shouting that TSMC is reigning supreme, I ain't touching that one with a 40 ft barge pole. Especially given Samsung's willingness to occasionally throw mindboggling amounts of money at making problems go away if they feel their market position is threatened in any way. If my memory serves right, they were spending a couple of billion from their cash reserves a month earlier this year.

 

2 hours ago, leadeater said:

The entire industry is dependent on ASML anyway and what they can do and their capabilities, it's a lot like F1 rules around engines. Who can do the best within the same "rules and regulations" or the same equipment.

ASML's dominance is a weird thing, and I wonder part of Intel's hesitance was partially intended to challenge that? They did definitely prove that it's possible without going for EUV, I'm just not sure if it was the best move of them given that they built up some technical debt by doing so. At this point I just primarily hope they don't buckle under investor pressure to split the company, because that would probably be a major loss to the electronics industry. I also wouldn't be surprised if a lot of the current media narrative being spun up around TSMC's dominance is a strategic attempt to pressure Intel, traders have definitely done such things in the past with major electronics manufacturers.

 

Simultaneously, I would love to see someone actually step up and compete with ASML. You occasionally heard crazy e-beam-based proposals float around in the past, but I don't think those went anywhere. There's of course also the proposed soft x-ray lithography methods, but that's such a blurry line with EUV, and a whole different can of worms.

 

 

 

But to get back to the topic at hand, we actually had an interesting conversation regarding this topic at work today while speccing new systems for [redacted  ]. The price for systems with the affected Intel products seems to have gone down somewhat, so we were literally in a case where we were considering getting the affected desktop chips over Xeons because we could get a better price on them. The Xeons won out due to a few other reasons, but it did make me wonder if any large organisations are going to take the gamble. Are integrators cutting their losses and just selling them at cost? If so, AMD might see their sales affected significantly anyway.

[DutchFakeTuber]

Apparently MSI and ASUS already released a new BIOS update for the newly released microcode update.

Jayztwocents made a quick video of it already, but I am really curious of what GN will say about it.

[porina]

Really looking forward to seeing the testing for the new microcode. I guess the biggest two questions are: does it impact performance, and is there a visible change in the voltages?

[Lurick]

2 hours ago, porina said:

Really looking forward to seeing the testing for the new microcode. I guess the biggest two questions are: does it impact performance, and is there a visible change in the voltages?

Jay at least saw a slight change in Time Spy Extreme testing of about 700 points but otherwise nothing major but time will tell.

[porina]

10 minutes ago, Lurick said:

Jay at least saw a slight change in Time Spy Extreme testing of about 700 points but otherwise nothing major but time will tell.

Skimmed the charts and worked out the % change in reported tests:

 

R23 MT +0.2%
R23 ST +0.4%
R24 MT -0.6%
R24 ST no change
Geekbench MT -0.7%
Geekbench ST -0.4%
Handbrake +1.7%
Blender Classroom +0.7%
Blender Junkshop -1.1%
Blender Monster +0.1%
TimeSpy EX CPU -5.7%
SOTTR -0.3%
CP2077 avg -3.0%
CP2077 min -2.1%

 

I don't know his methodology e.g. repeat testing to make sure there isn't an outlier but most of it is within margin of error. TSE would be worthy of follow up testing to rule out a glitched run for example, and is repeatable. Note I didn't watch the whole video so I don't know if that was addressed at all. CP2077 also shows a bit of a delta so similar could be said.

 

The time charts also don't seem to show any obvious difference. He seems to be looking at reported voltages. Think it was Buildzoid who did voltage probes previously so that might be more insightful when he gets around to repeating it with new microcode.

 

I got some "what if" scenarios forming in my head. Wont share them as speculation, but if they happen it could cause the community to blow up. Again.

[porina]

https://community.intel.com/t5/Processors/Microcode-0x129-Update-for-Intel-Core-13th-and-14th-Gen-Desktop/td-p/1622129

Intel's statement on the new microcode. In short, caps voltage requests to 1.55v but you can still override this for manual OC if you want to. Will be distributed by BIOS only, no OS pushes. Most tasks are not performance impacted but some are claimed to have a "moderate" impact. So expect the usual places to be testing that as I write.

 

Intel are still looking at minimum working voltage (Vmin) shifts on affected CPUs and will issue a further update later this month. Does this imply they might try to make unstable CPUs stable again by applying more voltage? But with the cap still in place this might lead to less boost potential.

[Glazarus]

I don’t care about a slight performance loss, as long as I’m getting stability and reliability. Since these are beta bios updates, what are the risks here? How long until we have a proper bios update that isn’t beta that has the updated microcode (Based on previous cycles)?

 

Everyone is jumping straight to testing the performance loss, but is that really important at the moment? Shouldn’t we be testing and looking into how stable this update is, especially since we’re getting only beta updates. 
 

Sitting on the creator board from Asus, the bios update just dropped, but I’m trying to get word from anyone who’s updated whether or not it’s safe to update or not. My 13th gen hasn’t shown any instabilities as far as I know, so I’m keeping my PC shut off until I know for certain we have a stable bios with new microcode.