AMD asks mobo makers to remove OC options for 5800X3D

d3athf1sh

[H]ard|Gawd
Joined
Dec 16, 2015
Messages
1,216
"TechPowerUp has verified a rumor posted over on VideoCardz that is quite puzzling, as AMD has asked motherboard makers to remove support for overclocking in the UEFI/BIOS for the Ryzen 7 5800X3D. When we asked for a reason as to why this was the case, we were told that AMD was keeping that information to themselves for the time being."

https://www.techpowerup.com/292782/...move-overclocking-options-for-ryzen-7-5800x3d

not sure what to say/think about this. it may only be temporary and they could have kinks worked out by launch. or maybe performance is TOO good and puts the rest of the line up to shame? who knows right now?
 
Last edited:
IBM and a number of chip manufacturers have struggled with 3D stacked chips for a long time and they always get caught up in cooling the stacks, most manufacturers have to include micro heartbreaks in-between (lots of patents from lots of companies on that topic), but those always add latency, height, and complexity, something AMD can't add in this case because they needed to keep the vertical dimensions basically identical and adding latency to the cache would basically defeat the purpose of having it and complexity adds cost. I am guessing OC'ing the chip in any significant capacity simply stresses it too much as the thermal density in those areas is just too much to handle and it damages connecting areas of the chips.
 
Kind of silly to give it the X in the name if that's the case though.
To be fair the x at no point signified overclockable. It's always been simply the higher performing part at stock speeds. Something that remains true in this case.

Hopefully if it is a heat issue and they aren't able to adequately protect the chip by other means, they explain that.
Who else here knows someone that smoked a 1ghz tbird because they didn't install the cooling properly or tried to overclock too high?
 
Try to avoid heat issue would be my guess, same has not going up to the 12-16 core with it.
 
  • Like
Reactions: pavel
like this
To be fair the x at no point signified overclockable. It's always been simply the higher performing part at stock speeds. Something that remains true in this case.

Hopefully if it is a heat issue and they aren't able to adequately protect the chip by other means, they explain that.
Who else here knows someone that smoked a 1ghz tbird because they didn't install the cooling properly or tried to overclock too high?

Except non x parts do not have the oc features unlocked, and it’s up to the board.

X parts have always had full overclock features. But I’m sure sure you were aware, but I don’t know if everyone seeing this would be.

It’s stands to reason the non x parts are not the unlocked overclockable parts.
 
iirc, -X models support PBO(2?), but all AMD CPUs are overclockable...well, except maybe this one. Whether you have bclk adjustment or just multipliers depends on the motherboard, and multipliers may be locked on certain CPUs.
 
hey for all we know they may just be removing "certain" features? like say bclk which would probably mess with the cache speed as well as the cpu. because they're obviously still gonna have boost enabled?

like when i first got my tuf x570 it didn't have bclk adjustment and a couple bios' later when they finally enabled it, it was worthless anyway because even adjusting it 1mhz would cause the system to literally take like 3-4 min to boot. but really to tell you the truth, on my 3800X, i get the best performance leaving all the cpu settings and pbo stuff at stock and just tweak the memory. and i've heard other people with high core count cpu's say the same thing
 
This chip can probably burn up pretty easily is my guess. AMD probably torched a couple trays of engineering samples to come to the conclusion.
 
  • Like
Reactions: pavel
like this
As long as it works with AMDs auto features properly who cares.... does anyone seriously manually overclock AMD chips anymore, I mean what is the point ? Other then saying it can be done. Performance gains are small... and in some cases you end up with actual performance regression. To me overclocking AMD chips at this point... is pretty much akin to the people wanting to put Linux on their toasters and toothbrushes. Sure perhaps its a silly fun thing to do... but there isn't any real point to it.

I suspect as others have suggested its a heat issue... likely the chips temp probes are not granular enough to ensure cooling is happening where it needs to be with probably very delicate interconnects to the 3D cache. I also suspect that with just the auto overclock this chip is going to destroy everything else in their line up for all but the most demanding stuff that can really use 16 physical cores. More cache has always been an instant win in the majority of software.
 
As long as it works with AMDs auto features properly who cares.... does anyone seriously manually overclock AMD chips anymore, I mean what is the point ? Other then saying it can be done. Performance gains are small... and in some cases you end up with actual performance regression. To me overclocking AMD chips at this point... is pretty much akin to the people wanting to put Linux on their toasters and toothbrushes. Sure perhaps its a silly fun thing to do... but there isn't any real point to it.

I suspect as others have suggested its a heat issue... likely the chips temp probes are not granular enough to ensure cooling is happening where it needs to be with probably very delicate interconnects to the 3D cache. I also suspect that with just the auto overclock this chip is going to destroy everything else in their line up for all but the most demanding stuff that can really use 16 physical cores. More cache has always been an instant win in the majority of software.
The Auto features are what AMD is asking them to remove. It also has lower base and boost clocks. Back in January when AMD was doing interviews about the technology they did a good job of avoiding any questions about overclocking the chip.
 
If they take the function away, I hope it returns later on IF any quirks are worked out. I'll still probably end up buying the X3D anyway if I can manage to get my hands on one. It sounds pretty sweet
 
If they take the function away, I hope it returns later on IF any quirks are worked out. I'll still probably end up buying the X3D anyway if I can manage to get my hands on one. It sounds pretty sweet
It's not a "quirk" its a fundamental problem with how they have layered the cache together.

"AMD places its 3D V-Cache on top of Zen 3's 'native' L3 cache and then connects them using through silicon vias (TSVs). This creates a non-uniform surface of the die, complicating positioning the integrated heat spreader (IHS) on top of the chip. To make it even, AMD places structural silicon spacers on top of Zen 3's processing cores, which limits their ability to dissipate heat. Yes, silicon does serve as a good thermal conductor, but it still traps a non-zero amount of heat between the die, structural silicon, and IHS."
https://www.tomshardware.com/news/amd-ryzen-7-5800x3d-overclocking-support

There are better methods for dealing with this inherent problem with TSV's but they are not cheap, they will likely use them going forward for the Zen4 chips but that will add a few $$$ to each one as the packaging becomes more complex.
 
Hmm... given all this, I wonder if it would make more sense to put the 3D cache on the motherboard instead of the chip? Like in the 386/486 days. Maybe that would be called L4 cache. :)
 
No, the latency would be too high. RAM is already about as fast as you can get things anyways.
Didn't Intel refer to the Broadwell GPU cache colloquially as Layer 4 cache or was that just Internet conjecture?
 
Hmm... given all this, I wonder if it would make more sense to put the 3D cache on the motherboard instead of the chip? Like in the 386/486 days. Maybe that would be called L4 cache. :)
I don't think motherboard cache makes sense when the memory controller is on chip. You'd either need a secondary memory controller on the board to manage the L4 + RAM and the chip memory controller would have to deal with inconsistent access latency on the memory bus, or you'd need to add a bunch more control/address/data lines for the cache.
 
Anyone who's run a 5800x knows it can run hot as heck fire. As has been discussed many times here, it has too many cores on too few chiplets with high vcore to hit those 4.8ghz speeds. A 5800X3D is going to need even more voltage to be stable (even accounting for it's dropped speeds). It will have little to no headroom, and AMD doesn't want enthusiasts complaining is my guess.
 
Hmm... given all this, I wonder if it would make more sense to put the 3D cache on the motherboard instead of the chip? Like in the 386/486 days. Maybe that would be called L4 cache. :)
cache latency would increase by 1ns every 6" of travel, that gets really slow really fast.
 
Probably the GPU doesn't like the heat or clock increase and gets unstable.
 
It's not a "quirk" its a fundamental problem with how they have layered the cache together.

"AMD places its 3D V-Cache on top of Zen 3's 'native' L3 cache and then connects them using through silicon vias (TSVs). This creates a non-uniform surface of the die, complicating positioning the integrated heat spreader (IHS) on top of the chip. To make it even, AMD places structural silicon spacers on top of Zen 3's processing cores, which limits their ability to dissipate heat. Yes, silicon does serve as a good thermal conductor, but it still traps a non-zero amount of heat between the die, structural silicon, and IHS."
https://www.tomshardware.com/news/amd-ryzen-7-5800x3d-overclocking-support

There are better methods for dealing with this inherent problem with TSV's but they are not cheap, they will likely use them going forward for the Zen4 chips but that will add a few $$$ to each one as the packaging becomes more complex.
Interesting I always thought cpus used a comparably thick amount of silicon on every produced chip. People were able to cut into the silicon surface on past cpus with no detrimental effects.

Makes me wonder if they are making the cache sections out of the same thickness of silicon and cant fine a efficient way to get away with thinner layers.
 
Interesting I always thought cpus used a comparably thick amount of silicon on every produced chip. People were able to cut into the silicon surface on past cpus with no detrimental effects.

Makes me wonder if they are making the cache sections out of the same thickness of silicon and cant fine a efficient way to get away with thinner layers.
I imagine if it's too thin, with vias, it could become (more) brittle. Not something you want under a heatspreader compressed with 100+ different aftermarket coolers of varying dubiousness.
 
Sounds like overlooking is not going to happen (cpu freq)

I, for one, am ok with that. Nobody knows how these things are going to survive in the wild. I would rather they disable oc up front than be forced to institute a recall because people are melting the 3d links
 
As long as it works with AMDs auto features properly who cares.... does anyone seriously manually overclock AMD chips anymore, I mean what is the point ? Other then saying it can be done. Performance gains are small... and in some cases you end up with actual performance regression. To me overclocking AMD chips at this point... is pretty much akin to the people wanting to put Linux on their toasters and toothbrushes. Sure perhaps its a silly fun thing to do... but there isn't any real point to it.

I suspect as others have suggested its a heat issue... likely the chips temp probes are not granular enough to ensure cooling is happening where it needs to be with probably very delicate interconnects to the 3D cache. I also suspect that with just the auto overclock this chip is going to destroy everything else in their line up for all but the most demanding stuff that can really use 16 physical cores. More cache has always been an instant win in the majority of software.
Indeed sometimes the performance “gains” are negative with manual overclocks…
 
Setting voltage to 1.2 and core clock to 4500 on my 5600x gets me the same cinebench score as using Ryzen master PBO while 10c-15c cooler.
 
I'm guessing that the cache acts like an electric blanket for the cores. It's possible that besides the overall temperature increase, there are micro hot spots during certain workloads that would be flat out dangerous if overclocking.
 
I'm guessing that the cache acts like an electric blanket for the cores. It's possible that besides the overall temperature increase, there are micro hot spots during certain workloads that would be flat out dangerous if overclocking.
That's why I wonder why amd didn't just make a new io die with more cache or put it off to the side or something.
Shame they killed off tr that massive heat spreader would have been a great thing for this type of stuff
 
Back
Top