A recent discovery that overclocking AMD’s latest chips blows a fuse to denote the chip has been overclocked has led to slightly misleading claims that it will automatically void the chips’ warranty for any type of failure. However, AMD clarified to Tom’s Hardware that overclocking AMD’s Ryzen Threadripper Pro 7000 (Storm Peak) and non-Pro lineup, among the best workstation CPUs, doesn’t automatically void the processor’s warranty.
↫ Zhiye Liu at Tom’s Hardware
Something about these fuses in processors doesn’t sit right with me.
I on the other hand love it. It is a strike for second hand users. I should be able to detect if the chip was overclocked, and AMD made DAMN sure that ite blown fuse does not hinder you from using it at stock speed OR overcloking yourself. This is a step to satisfy the regime in CCP so that used compromized CPU’s does not end up in critical systems and we should all applaud that tiny security measure.
This.
You never know what people does with their chips, and event though AMD almost always been kinda lax with overclocking, it’s a nice touch if any user can have a feedback about this.
My understanding is that it may not be the case that Joe Blogg’s will be able to check the CPU fuse though so how does it help?
Wait for the next update of https://www.cpuid.com/softwares/cpu-z.html
This isn’t a bad thing for secondhand users, but we should be clear that while AMD are just testing this today, these fuses will be used to deny warranty claims in the future.
I don’t typically want to overclock CPUs. Heck sometimes I even want to underclock them. But something else that needs to be understood is that enabling your DDR’s native XMP speed, which nearly everyone does because of course they do, may be considered “overclocking”. For example, in intel’s warranty process they do ask if you have XMP enabled and if so it’s grounds for them to not honor their CPU warranty. Previously there would not be evidence for this, so you could play dumb, but if these fuses become standard, the vast majority of customers could have their warranties voided.
I looked it up and this is AMD’s policy as well.
https://www.pcworld.com/article/394250/why-xmp-and-memory-overclocking-are-ok-even-if-they-void-your-warranty.html
Without evidence, the enforcement may have been lax in the past, but if fuses can provide this evidence, I get a strong feeling enforcing CPUs warranties is exactly the reason these fuses are being added. It’s not for the benefit of the second hand market.
Every CPU since software based overclocking was invented has certainly had fuses for this… this is non news. Even microcontrollers at the very bottom price tiers of all CPUs have fuses for various features, including fuses you can blow to prevent reprogramming or reading etc etc…
Fuses are NOT new. Neither is it likely that they just added them on Ryzen 7000… they’ve been there for like 2 decades.
cb88.
Is there evidence for this? This is what the article says…
I am familiar with those. Using fuses to configure features & binning versus using them to tattle on owners are very different use cases though. I’d appreciate a link if you’ve got any more information about manufacturers using them for the latter.
Evidence? you mean other than 40-50+ years of chips having fuses?
Fuses are also probably how they used to disable cores before they started lasering them off… and 4 core chips sold as 3 cores were the fuses failing to burn in.
cb88,
Evidence? you mean other than 40-50+ years of chips having fuses?
Fuses are also probably how they used to disable cores before they started lasering them off… and 4 core chips sold as 3 cores were the fuses failing to burn in.
Evidence as in the context of the article and discussion. The use of “fuses” to reconfigure chips after testing & binning isn’t new or controversial, but that’s not what we were talking about here. Is there any evidence that fuses have been used in the way they’re discussed in the article? The article suggests this use case could be new, but I’d be curious if you could provide evidence that it isn’t.
Alfman:
https://free60.org/Hardware/Fusesets/
https://en.wikipedia.org/wiki/EFuse
neuechristian,
Nobody was questioning the existence of efuses. I’ll requote the assertion where I asked for evidence…
Both myself and cb88 were explicitly talking about the use of fuses to record overclocking. Then for whatever reason the discussion pivoted to different point that doesn’t back the original assertion about fuses being used “for this” (ie the article). Since no evidence is being provided, it still seems that nobody here knows if fuses were ever used to record overclocking before.
Alfman this is a little more complex.
For AMD knowing if a Chip has been overclocked is important for more than warranties. Overclocking can and does make a chip age faster. So getting back a chip that been overclocked that has died may be a good idea to replace under warranty so you get them back in future. Yes that overclocked CPU failing early can be canary in the coal mine that you are going to have a stack of CPU need replacement due to some hidden defect. So allowing company to have stockpile of CPU to deal future failures under Warranty in a timely fashion.
Like if you overclock CPU and do something really stupid like pushing the CPU so far outside specification that it was going to fry no matter what of course AMD and Intel are not going to replace it. If what you did was just accelerate the CPU normal ageing so showing a hidden defect that AMD/Intel did not know about by most countries laws AMD/Intel have to replace this chip under warranty or be illegal.
Its amd/Intel that has to prove that the overclock did the damage not to warranty replace. Knowing if a chip has been overclocked lets AMD/Intel check of mass over-voltage/over amp-age and other stupidity.caused by overclocking on a motherboard that power supply system cannot take the overclock load. If that what killed the chip then it not AMD/Intel fault.
Manufacture defects are Manufacture defects. So a chip that would have failed it natural life span(some countries this 10 years for a CPU) due to some defect that has failed in 1 due to user overclocking(as in 9 year early) by most countries laws AMD/Intel are basically on the hook to replace it. Now if the defect would only have caused the chip to fail at 11 years as in out side the countries time frame under non overclocked state and it failed in 1 because you over clocked technically by law AMD/Intel is off the hook to provide you with replacement. Remember if it ever gets proven that AMD/Intel did not replace a chip they should of some countries will hit them for multi million dollar fines per chip not replaced.
AMD/Intel generally air on don’t ask don’t tell and replace it even if we know it been overclocked legally safer that way. The big thing for AMD/Intel is to look at the failed overclocked chips to work out if they need more replacement inventory of particular models or work out if a chip need a substitute because there is a defect that going to cause them too many returns in future.
Its the basic problem with warranties having evidence that user happened to use device X way does not make it instant that you can void warranty. You have to be able to prove absolutely as device maker that using the device X forbid way has caused some failure that would not have happened in the expected lifespan of the device not using X forbid way or you are on the wrong side of consumer protection laws that can hit really hard.
Warranties are lot harder to void that most think. It quite a regular thing around the world for countries being prosecuted and fined for voiding warranties because they have failed to absolute prove that the failure was the customers fault instead of a hardware defect that would happened in future anyhow.
Horrible right customer abuse of a device is not valid grounds by itself not to replace a device under warranty. Lots of cases doing up the required report legally required to be able to void warranty is more than what the device is worth. Cheaper 99% of the time just to ignore what the user has done to the device unless it some internally useful information like cpu like overclocked cpu failing early as a sign that non overclocked form a particular batch will fail so we need spares for when they do.
Yes a formal report proving something like a Ryzen Threadripper 7000 failed due to being overclocked not due to some design defect is going to run in 10 to 15 thousand USD and the reports outcome that there was no design defect could be wrong, Horrible problem right. Its just not worth Intel or AMD to fight over people overclocking cpus and breaking them unless it something so badly wrong that the report is simple todo and it clear. Like massively too much voltage/amp provide by motherboard.
oiaohm,
The only time amd will get those chips back to check the fuses is when someone tries to get service under warranty. The fuses provide very little benefit to them If not for this warranty interaction,
I agree with this in principal, however in practice I’ve dealt with enough warranty departments to know that many instruct employees to use any and every excuse they can to automatically deny warranty. To be clear, I’ve never had to go through this with AMD or intel I’ve had 100% success rate with CPUs either because I am lucky or they are extremely reliable. But I’ve had warranty issues with other things like disks, monitors, power supplies, etc and I’d say a good portion of those do fight their owner customers to wrongly deny warranties. It’s not right. Consumers even have the law on their side (Magnuson-Moss Warranty Act). But by the time you hire lawyers, serve legal documents, taking time off, travel out of state for court, etc, the reality is that Joe regular consumer is very likely to be even more in the hole.
Alfman
“The only time amd will get those chips back to check the fuses is when someone tries to get service under warranty. The fuses provide very little benefit to them If not for this warranty interaction,”
This is your mistake you are thinking the fuse provided no benefit other than for warranty. How do you check the fuses. Inserting a suspect CPU into socket to check fuses could end up frying you test board.
How the AMD/Intel RMA process mostly works is user has submitted the right form of report that the chip is dud. When AMD/Intel receives the dud chip they send on replacement straight way then put the dud chip in the chip design depart chips to disable and research. Please note the dud chip does not get powered on until after it been disassembled checked for shorts and reassembled to prevent destroying more hardware. The time between AMD/Intel receiving a dud chip and they finally get around to the full inspection can be 6 months so yes 6 months after they have send out the replacement chip can be when they find out the fuse that the chip was overclocked was broken.
Now if you admit that you have overclocked you chip before sending it in then you can hit a staff problem.
Please note this is not the first AMD cpu with fuse that blows when you overclock and historically they have never refused RMA on a chip for just a blow fuse that shows how the user was using the chip.
AMD/Intel normally start questioning and slowing down the RMA process if you are doing 2 or more CPUs in a row and will look a bit closer. (yes it very low odds normally that a person will have two broken CPUs in under 6 months without doing something stupid to them).
AMD rule normally is if the chip was clearly over-votage/over-amp damage or run hard without a cooler yes this is where you can look at CPU without powering it on and see that the user has done something really stupid due to visible physical damage. Exceptions have been made when the motherboard ODM is to blame or AMD firmware provided to motherboard ODM is to blame but the first person to make this RMA claim normally has an argument with AMD..
You don’t seem to understand what I’m saying. Nothing you say will happen until the user returns the faulty CPU. The only time they get CPUs back is under warranty.
Has anyone here, ever warrantee returned a CPU? My sample size is very small, but I’ve never heard of anyone doing it. I think. I have had entire motherboards replaced, which I think also ended up swapping with a new CPU and memory. Maybe it was the CPU and not the motherboard/RAM? Don’t know.
Bill Shooter of Bul,
Same. Of all the computer failures, it’s always been something else. Obviously I see how they could be damaged through either abuse or manufacturing defects, but CPUs have been extremely reliable for me.
I had an AMD cpu die after a raidmax power supply exposed it to mains voltage, but that’s to be expected 🙁
Believe it or not raidmax didn’t want to honor their warranty…WTF! That’s what convinced me not to buy cheap power supplies any more.
I have done an RMA with AMD twice, and AMD replaced the chip both times. The first time was with a Ryzen 1700X where it had some strange errors in Linux, ran the Ryzen test/kill Ryzen and it was faulty. Not sure if I can embed a link but if you search for Ryzen 1700 RMA Kill Ryzen you’ll see the github/reddit threads about it. Separately I had a 5800X that ran extremely hot, no overlock and with overkill cooling. I submitted it to AMD with an explanation and some records showing what I was experiencing and they tested it and sent a replacement.
Happy with AMD customer service, but also must be incredibly unlucky.
Not directly, but I did have a CPU replaced through Dell when I worked as a tech at a University. It was a early P4 and the internal cache was bad.
I also could have RMA’d my defective AMD Ryzen 1700 as it was manufactured early (launch chip). Those had a known defect that caused problems with a few things due to bad instructions. One thing that will consistently trigger it is the IPFW firewall in FreeBSD 11 or higher. Start changing rules while taking traffic and full system crash. I also saw it with some virtualization workloads. I ended up just throwing a 2700 in instead because I didn’t want to deal with the hassle.
So in using computers since the mid 90s, I’ve only seen 2 bad CPUs. It’s anecdotal but it’s still pretty low compared to other types of components.
When you have a 96 cores and 192 processing threads and Still need to overclock… you are doing some serious work!!
If you paid for a 96 core … wouldn’t you want it to go as fast as possible also?
Fair question, but I but honestly, when you are paying that much for a chip you assume it’s being used in a professional scenario.
I’d want it to therefore be as stable as possible, and exposing it to potential instability and/or potential damage feels risky.
A home CPU, costing a fraction of that, is a different matter of! But I do make the judgement call of “can I afford for this to go wrong” before I over clock anything…
I miss an edit button so much..
Met oo!
cpcf,
I see what you id there.
Actually no, if I’m spending several thousands for a 96 core chip, as Adurbe said I want it to be as stable as possible while using as many cores as possible, which means running at its officially rated speed, and therefore within its guaranteed thermal envelope. You don’t buy a 96 core CPU to get better framerates in games, they are for massively parallel workloads that can actually take advantage of all those cores, as well as in VM servers where you can assign one to four cores per VM. As much as I’d love to play around with such a high core count CPU, it’s not a toy or a curiosity, it’s a professional tool and should be treated as such.
Same. I could see it being used in desperation to squeeze a bit more juice out and the death of the cpu doesn’t mean much for reasons I’m not sure of, maybe academic research modeling or quants where someone else pays for the equipment?
Yea, enough work that its worth potentially damaging the cpu. Not my line of work, but hats off to those who do. With cloud taking over everything its odd to hear people using hardware to their limit.
Who is the fool to jeopardize a $13k CPU ? If you can afford one, you’d better get the 100% of a second one instead to just painfully milk 15% more of your first one.
Answering my own question as well here. Quants make the most sense. If you have a complex model that is massively parallel but can only run on CPU, not GPU and you need to keep it on prem and you have five back ups on hand at all time, yeah let the thing fry as fast as it can go. 13 k is a small cost in this case. Or maybe as part of a rendering farm for special effects? I wouldn’t imagine those to be CPU bound, but I don’t know. If you have a budget for developing new special effects in the millions a 13k chip dying in an effort to reduce the feed back loop between director and animators isn’t a bad price. I don’t think any mainstream crypto is worth enough to be cpu mined, so not them.