Tuesday, August 5th 2025

AMD's Next-Gen AM6 Socket Expected to Boost Pin Count to 2100 While Matching AM5 Size

AMD's next AM6 socket will have a much higher pin count while still working with current cooling systems. New patents filed by AMD (US 20250149248) and reports from Bits&Chips cited by Wccftech show that the socket will have 2100 pins up from AM5's 1718 pins. This 22% jump in pins could lead to better performance. More pins might allow for power delivery above 200 watts beating AM5's 170-watt limit. Extra pins could also mean more data lanes and faster input/output supporting PCIe 6.0 in the future. However, we might not see widespread use of PCIe 6.0 until about 2030. Keep in mind that while patent diagrams suggest increased pin density, some details remain speculative since certain illustrations may reference server-grade SP5 sockets rather than mainstream consumer platforms. The actual implementation may differ from current projections, particularly regarding memory channel configurations.

AMD seems to keep AM6 the same size as AM5, so it will work with existing AM5 and AM4 coolers. Still, third-party companies might develop new cooler versions that work best with Zen 7 chip layouts and heat needs. The switch to the new socket will take time since AM5 will still support the upcoming Zen 6 chips giving the platform about three more years of life. We expect to see AM6 launch with the Zen 7 design marking AMD's next big socket change.
Sources: Wccftech, Bits&Chips
Add your own comment

49 Comments on AMD's Next-Gen AM6 Socket Expected to Boost Pin Count to 2100 While Matching AM5 Size

#26
usiname
Tek-CheckSadly? Nonsense. If one really needs more than two channels for specific workloads, one can pony up and buy HEDT system. Also, there are Strix Halo systems for anyone who needs four channels.
So if I want the same CPU with 4 instead 2 memory channels I can "just" pay 3x the price of the 2 channels CPU for the same compute power, good to know. About Strix Halo, where I can plug the dGPU if I am not happy with the RTX 4060 level gpu or additional expandable cards?
Posted on Reply
#27
jnv11
freeagentYou lost me at ADHD

I have 5600X, 5900X, and 5800X3D. I was always switching back and forth between my 5900X and X3D. I prefer my 9900X over all of them..

Seems ok to me. Also, fast as fuck boi.

:confused: :confused: :confused:
I am still seeing new cases of dead 9800X3D CPUs being posted to Reddit from time to time. However, these new cases all seem to take place on ASRock motherboards and not other brands of motherboards. This looks like an ASRock problem to me.
Posted on Reply
#28
freeagent
jnv11I am still seeing new cases of dead 9800X3D CPUs being posted to Reddit from time to time. However, these new cases all seem to take place on ASRock motherboards and not other brands of motherboards. This looks like an ASRock problem to me.
I run Asus.. zero problems.

I built this system last winter, she hasn't had an easy life..

Posted on Reply
#29
Tek-Check
usinameSo if I want the same CPU with 4 instead 2 memory channels I can "just" pay 3x the price of the 2 channels CPU for the same compute power, good to know.
Indeed. There are more things that are good to know in order to avoid acting naively about HEDT CPUs. It's a niche item, it requires significantly more design and production effort, more silicon and more complex platform in terms of IO and moving parts. Pay for it.

You want it cheaper? Sure. Ask Intel to be competitive and release new Xeons. You want it even more cheap? Absolutely. Ask Nvidia, Qualcomm and Apple to make their first ever DIY products with four channels.

There are good reasons why even some of the richest companies dare not venture into DIY desktop space in the first place, let alone offer diverse CPUs with varied memory channels. Think about it.
usinameAbout Strix Halo, where I can plug the dGPU if I am not happy with the RTX 4060 level gpu or additional expandable cards?
You can use eGPU enclosure if you need more.
Posted on Reply
#30
ghazi
I can't believe we're already looking at DDR6 (after all it would have to be AM5+ with DDR5). DDR5 looks to be the first generation of memory I skip entirely.
Posted on Reply
#31
Tek-Check
ghaziI can't believe we're already looking at DDR6 (after all it would have to be AM5+ with DDR5). DDR5 looks to be the first generation of memory I skip entirely.
Zen6 is expected to launch with 8,000 MT/s RAM speed, so the introduction of DDR6 in Zen7 might start around 9,000-10,000 MT/s.
Posted on Reply
#32
usiname
Tek-CheckIndeed. There are more things that are good to know in order to avoid acting naively about HEDT CPUs. It's a niche item, it requires significantly more design and production effort, more silicon and more complex platform in terms of IO and moving parts. Pay for it.

You want it cheaper? Sure. Ask Intel to be competitive and release new Xeons. You want it even more cheap? Absolutely. Ask Nvidia, Qualcomm and Apple to make their first ever DIY products with four channels.

There are good reasons why even some of the richest companies dare not venture into DIY desktop space in the first place, let alone offer diverse CPUs with varied memory channels. Think about it.

You can use eGPU enclosure if you need more.
To have lower price nor Intel, nor AMD, nor Nvidia should do something, just people like you should not exist and the market will fix itself
Enjoy crippled eGPU and 3x prices, we are here thanks to you
Posted on Reply
#33
Tek-Check
usinameTo have lower price nor Intel, nor AMD, nor Nvidia should do something, just people like you should not exist and the market will fix itself
Enjoy crippled eGPU and 3x prices, we are here thanks to you
What kind of braindead nonsense is this? My existence is irrelevant for the market.

If you want lower prices, competitors need to be ready to invest into developing and offering equivalent DIY platforms to what AMD offers. You know this basic market rule, so don't pretend. And spare me from personal comments. Reported to mods.
Posted on Reply
#34
usiname
Tek-CheckWhat kind of braindead nonsense is this? My existence is irrelevant for the market.

If you want lower prices, competitors need to be ready to invest into developing and offering equivalent DIY platforms to what AMD offers. You know this basic market rule, so don't pretend. And spare me from personal comments. Reported to mods.
Sorry for hurting your feelings, but people like you who fight tooth and nail to defend the overpriced nonsense cause even higher prices. Don't be disappointed next time when you go to shopping for hardware and there is no discount for the warrior who fights for the honor of multi-billion dollar companies.
Posted on Reply
#35
Tek-Check
usinameSorry for hurting your feelings, but people like you who fight tooth and nail to defend the overpriced nonsense cause even higher prices. Don't be disappointed next time when you go to shopping for hardware and there is no discount for the warrior who fights for the honor of multi-billion dollar companies.
You don't seem to differentiate between levels of explanation. I didn't defend prices. I merely explained why prices are the way they are. The same applies to halo products that do not have competition, such as 5090. If I explain to you why 5090 is expensive, I hope you will not attack me for defending Nvidia. So, get your ducks in order and stop personal comments when someone tries to explain to you something pretty simple.
Posted on Reply
#36
Kursah
usinameSorry for hurting your feelings, but people like you who fight tooth and nail to defend the overpriced nonsense cause even higher prices. Don't be disappointed next time when you go to shopping for hardware and there is no discount for the warrior who fights for the honor of multi-billion dollar companies.
There is no need for this here than going after folks directly. Find a better way to get your opinions across if you must post them. Check the forum rules & guidelines, link in sig.

To the rest of you, please carry on in a constructive fashion. Thanks! :toast:
Posted on Reply
#37
Tek-Check
XajelDDR6 is rumoured to have a 4x 24bit per DIMM, so If I understood the rumours correctly (which I might not), a single DIMM will have a quad channel, a dual DIMM will be Octal Channel. The CPUs will have 2x IMCs, each IMC will handle a single DIMM (4x 24bit).

You might have noticed that 2x 32bit in DDR5 is 64bits per DIMM, just like how DDR4 and lower was. But DDR6 having 4x 24bits means a total of 96bit per DIMM, which is an increase in bus width as well.

Separating the bus into sub channels reduces the challenges with wider bus in higher frequencies. In DDR6, they reducing the subchannels from 32bit to 24bits, and increased the number of subchannels to 4. This allows them to further increase the speed more. No wonder DDR6 requires a new slot design, it's not just the T topology of current DIMM slots limiting the frequency, but the increased bus width requires more contacts too, so they're considering CAMM2 sockets.
It should be noted that each one of the four 24-bits sub-channels on a single DDR6 DIMM has 8-bit section for error correction, thus totalling 64-bit (4x16-bit) for data and 32-bit for error correction per DIMM. Zen6 CPUs are indeed runoured to feature two IMCs in silicon, one for each 96-bit (64-bit + 32-bit) memory logic.
resources.altium.com/p/ddr5-vs-ddr6-heres-what-expect-ram-modules
Posted on Reply
#38
ZeDestructor
XajelI don't see it the same size as AM5, it certainly looks bigger, but its still within the boundaries of the older mounting holes that it could basically keep them in place.


This mean that "some" AM5 coolers might not be compatible with AM6, not because of the surface design, but because of the mounting mechanism. They could just have an AM6 specific mount and get with it, but the larger surface of AM6 will probably means a new cooler design.


Edit: The FIG3 pictures clearly shows more than 2 DIMM channels, this socket could be SP6's successor.





Yes and no.

DDR5 now has a 2x32bit bus per DIMM, so it's technically a 2x DIMM setup like most of us know as Dual Channel is actually a Quad Channel, though the technical term I prefer is each DIMM have 2 sub channels, a 2x DIMM (dual channel) system have a total of 4x subchannels.

DDR6 is rumoured to have a 4x 24bit per DIMM, so If I understood the rumours correctly (which I might not), a single DIMM will have a quad channel, a dual DIMM will be Octal Channel. The CPUs will have 2x IMCs, each IMC will handle a single DIMM (4x 24bit).

You might have noticed that 2x 32bit in DDR5 is 64bits per DIMM, just like how DDR4 and lower was. But DDR6 having 4x 24bits means a total of 96bit per DIMM, which is an increase in bus width as well.

Separating the bus into sub channels reduces the challenges with wider bus in higher frequencies. In DDR6, they reducing the subchannels from 32bit to 24bits, and increased the number of subchannels to 4. This allows them to further increase the speed more. No wonder DDR6 requires a new slot design, it's not just the T topology of current DIMM slots limiting the frequency, but the increased bus width requires more contacts too, so they're considering CAMM2 sockets.

The two main concerns I have regarding CAMM2 sockets are:
1. Desktop platform will probably have a single CAMM2 socket as it can replace 2x DIMMs (Natively Dual Channels). Upgrading RAM will mean replacing the old module with a new one, unlike now where you can add extra DIMMs for more capacity (assuming the guy doesn't care about RAM speed).
2. The perpendicular design of the CAMM2 module means larger area on the motherboard, which will be more challenging with SFFs, especially the mini ITX designs, some might opt for backside replacement, but not all cases will support this, and heat management might be an issue.
I'd be a bit careful when talking about the 4×24b subchannels on DDR6 - on DDR5, it's not 2×32b subchannels, it's 2×40b (32b data + 8b ECC). if JEDEC is still needing the 8b per subchannel for ECC, then we could be looking at 4×16b data + 4×8b ECC, for a total of 64b data + 32b ECC. Based on what I see public online, I would expect the 64b data + 32b ECC for DDR6 setup to be what we do get in the end. Would love to be wrong though.

LPDDR6 is indeed a 24b channel composed of 2 12b subchannels (though no ECC at all), so there is a rather thigher-than-normal chance that we would be seeing 24b data subchannels on DDR6 proper. Still, I'm not getting my hopes up. Much prefer getting my hopes up for CXL suportcoming to the consumer socket and dropping an fat 5.0×8 (or ideally 6.0×8) memory expansion card in there.
jnv11I am still seeing new cases of dead 9800X3D CPUs being posted to Reddit from time to time. However, these new cases all seem to take place on ASRock motherboards and not other brands of motherboards. This looks like an ASRock problem to me.
Sounds to me like said redditors did not install updated firmware, because ASRock has already publicly admitted fault (they shipped too high voltages) and have already released fixed firmware.

Personally I've got a 9800X3D on an Asus X670E board and aside from being a bit temperamental with memory training (despite the super-basic JEDEC 5600 setup with ECC enabled.. or ECC is the part causing the temperamentalness? 🤔). That said, every firmware update incrementally made it batter, to the point that these days it's pretty dang reliable, though I will say that there is still room for improvement compared to how utterly bulletproof my X99, X79 and Z77 systems were with similarly lax memory setups (minus the suicidal Asus X99, ofc).
Tek-CheckIt should be noted that each one of the four 24-bits sub-channels on a single DDR6 DIMM has 8-bit section for error correction, thus totalling 64-bit (4x16-bit) for data and 32-bit for error correction per DIMM. Zen6 CPUs are indeed runoured to feature two IMCs in silicon, one for each 96-bit (64-bit + 32-bit) memory logic.
resources.altium.com/p/ddr5-vs-ddr6-heres-what-expect-ram-modules
Dang, you beat me to it 🤣
Posted on Reply
#39
jnv11
ZeDestructorI'd be a bit careful when talking about the 4×24b subchannels on DDR6 - on DDR5, it's not 2×32b subchannels, it's 2×40b (32b data + 8b ECC). if JEDEC is still needing the 8b per subchannel for ECC, then we could be looking at 4×16b data + 4×8b ECC, for a total of 64b data + 32b ECC. Based on what I see public online, I would expect the 64b data + 32b ECC for DDR6 setup to be what we do get in the end. Would love to be wrong though.

LPDDR6 is indeed a 24b channel composed of 2 12b subchannels (though no ECC at all), so there is a rather thigher-than-normal chance that we would be seeing 24b data subchannels on DDR6 proper. Still, I'm not getting my hopes up. Much prefer getting my hopes up for CXL suportcoming to the consumer socket and dropping an fat 5.0×8 (or ideally 6.0×8) memory expansion card in there.



Sounds to me like said redditors did not install updated firmware, because ASRock has already publicly admitted fault (they shipped too high voltages) and have already released fixed firmware.

Personally I've got a 9800X3D on an Asus X670E board and aside from being a bit temperamental with memory training (despite the super-basic JEDEC 5600 setup with ECC enabled.. or ECC is the part causing the temperamentalness? 🤔). That said, every firmware update incrementally made it batter, to the point that these days it's pretty dang reliable, though I will say that there is still room for improvement compared to how utterly bulletproof my X99, X79 and Z77 systems were with similarly lax memory setups (minus the suicidal Asus X99, ofc).



Dang, you beat me to it 🤣
I have found posts on Reddit complaining about dead 9800X3Ds with the latest firmware. I am not sure that ASRock has found the root cause of this. However, it is also possible that the CPUs got damaged before the firmware update and that they were dead CPUs walking until they finally failed as a result. I doubt that ECC has any connection to these failures as most consumers are using non-ECC.

Since you are running ECC, maybe you should look into logging ECC corrections and errors if you can figure out how to do that. I am wondering if the ECC correction and error rates could be a possible early warning indicators.
Posted on Reply
#40
Xajel
ZeDestructorI'd be a bit careful when talking about the 4×24b subchannels on DDR6 - on DDR5, it's not 2×32b subchannels, it's 2×40b (32b data + 8b ECC). if JEDEC is still needing the 8b per subchannel for ECC, then we could be looking at 4×16b data + 4×8b ECC, for a total of 64b data + 32b ECC. Based on what I see public online, I would expect the 64b data + 32b ECC for DDR6 setup to be what we do get in the end. Would love to be wrong though.

LPDDR6 is indeed a 24b channel composed of 2 12b subchannels (though no ECC at all), so there is a rather thigher-than-normal chance that we would be seeing 24b data subchannels on DDR6 proper. Still, I'm not getting my hopes up. Much prefer getting my hopes up for CXL suportcoming to the consumer socket and dropping an fat 5.0×8 (or ideally 6.0×8) memory expansion card in there.
I thought about the ECC but I don't know I didn't thought about while writing my reply, you're totally correct, the 4x24b subchannels might relate to the server market then as they include ECC bits usually.

I personally hope ECC will become standard on desktops, or at least more adoption that the current measly situation. With more higher clocked kits for desktop platforms (Registered, Unbuffered), I don't mean very high clocks, but something good above the JEDEC standard.
Posted on Reply
#41
ZeDestructor
jnv11I have found posts on Reddit complaining about dead 9800X3Ds with the latest firmware. I am not sure that ASRock has found the root cause of this. However, it is also possible that the CPUs got damaged before the firmware update and that they were dead CPUs walking until they finally failed as a result. I doubt that ECC has any connection to these failures as most consumers are using non-ECC.

Since you are running ECC, maybe you should look into logging ECC corrections and errors if you can figure out how to do that. I am wondering if the ECC correction and error rates could be a possible early warning indicators.
Could well be ASRock still having faulty firmware, but considering the precipitous dropoff in reports after they released the updated firmwares, I'm rather more suspicious of the reporters - it doesn't take much to damage a chip, so it could well happen on first boot if the users didn't use flashback to update the UEFI without a CPU installed and instead updated on first boot from the "GUI" flasher.

ECC definitely has no bearing on the failures, I just added it for context on what my history with UEFI updates has been with my board - namely that things got incrementally better with each update.
XajelI thought about the ECC but I don't know I didn't thought about while writing my reply, you're totally correct, the 4x24b subchannels might relate to the server market then as they include ECC bits usually.

I personally hope ECC will become standard on desktops, or at least more adoption that the current measly situation. With more higher clocked kits for desktop platforms (Registered, Unbuffered), I don't mean very high clocks, but something good above the JEDEC standard.
It's specifically during POST/pre-POST memory link training, so well before ECC functionality would be a factor. What might be a factor is that EC4 ECC UDIMMs are using 36-bit wide subchannels rather than the usual 32-bit wide subchannels of non-ECC memory, so it would be a bit tougher to train, hence my speculation.
Posted on Reply
#42
5900xDdr4AM4Lyfe
freeagentYour loss, AM4 is slow.

Rookie numbers.

SLOW is an absolutely incorrect and misrepresented word to be speaking about am4 and its relative "speed" to AM5.... 5900x pbox x570 xtreme rev 1.2 f39i bios win11 pro(24h2 and still blocking 25h2). Tug gaming 3080TI 12GB OC, 32 2X16 G.SKILL TRID NEO 3800MHZ cl16. 32" 240hz (260oc) 8, 10, 12 bit color bitrate 1.4 1.4a b, 2.4? 2.1 hdmi 4k120hz hdmi... 1ms 32GQ850-B hdr 600 nano IPS PANEL + ACER PREDATOR 144 HZ 24 INCH 1P80 NATIVE 160 HZ Oc basic colors (full or limited rgb 444 or 424 422. 8bit) .....Also I can run iracing high to max settings and multiple overlay/side apps and can render or run code with hypervisors using vs code or just code on a seperate virtual session..... iracing uses a direct drive wheel motor, 3 pedals with fast af bit rates at high dpi and resolution keyboard seperate and mouse form main desk still running and 3x 165hz curved plus 1 flat on desk, and AM4 HAS NO ISSUES. so... slow is absolutely not the right word... even relatively. It's not slow. It's amazing that am4 still can hang with the best while keeping consumption down and temps in the sub 85c....
Posted on Reply
#43
freeagent
5900xDdr4AM4Lyfeslow is absolutely not the right word...
Compared to AM5, yeah it is.

In case you missed this part:
freeagentI have 5600X, 5900X, and 5800X3D. I was always switching back and forth between my 5900X and X3D. I prefer my 9900X over all of them..
I have 2 AM4 systems running right now, and a third in boxes.
Posted on Reply
#44
5900xDdr4AM4Lyfe
freeagentCompared to AM5, yeah it is.

In case you missed this part:


I have 2 AM4 systems running right now, and a third in boxes.
I have used 12 ge, 13,14, all 900k f, etc and the ryzen 9's from the zen3 am4, am5 9000 9900x, 7800x3d, and maybe more for extended peridions on multiple applications. am4 is asolutely not "Slow" lol. throwing money and building many differnt or useing diffferent configurations of AM4 vs a few other AM5 is not a check box that you filld somehow all of the sudden making you the expert to declare the truth. Truth is..... Its not slow and the AM5 is obviously faster, but am4 still handles literally 90 percent of the possible workloads available to it unless you're using hash rate and folding and other non classified as benchmarking normal use cased work loads. the difference is not negligable y any meens, but its still competetive with AM5 in a mid/igh tier vs am4's highest tier zen 3 5900 950 chips.. That is am4 reardless, is amazingly competetive givent the amount of time they have ben actively relavant andi honestly wouldnt be suprised if AMD came out with some software instrutionset breakthroughs available to the am4 that would make them even more competetive is they took the time rather than r&D for new tech and h/w dev.
Posted on Reply
#45
Essaudio
Why do you need more pins for PCIe6 - asking seriously. It’s still the same number of lanes (pins) just higher clocks, right? Is it grounding / power ir something else?
Posted on Reply
#46
Tomorrow
AM4 is still perfectly usable in 2025. Sure it's slowER than AM5, but it's not unusable crap everyone is desperate to get rid of like what happened to AM3 after AM4. Generally if you have 12th gen Intel or AM4 with Zen 3 you still have reasonably modern system.

HUB has tested CPU bottleneck and only with 4090-5090 do you start to get to the point when AM4 based X3D is holding back the GPU.
If you're using 5070 Ti/9070 XT or slower then AM4 with Zen 3 is still viable.

As 5800X3D owner i dont think even 9800X3D is fast enough to justify an upgrade. Im waiting for Zen 6 X3D on AM5. Likely 2027. Hopefully DDR5 prices have fallen back to some normalcy by then. Right now is a bad time to upgrade to anything unless you literally have no working PC.
Posted on Reply
#47
A Computer Guy
5900xDdr4AM4LyfeSLOW is an absolutely incorrect and misrepresented word to be speaking about am4 and its relative "speed" to AM5....
I still hold my 3950x/5950x dearly (and now 5975wx) but Zen3 cores are definitely slow compared to 7950x (Zen4) and memory operation is significantly faster in AM5. AM4 is dead long live AM4. Working in VM's the difference was very noticeable especially in daily use for me so I use a fast (7950x) and a slow (5975wx) machines for various purposes because of AM5 tradeoffs and my 5950x is just fine for 60fps 1080p gaming. No hard feelings I still love AM4 but I'm pining for that 9950X3DD or TR 9960X.
Posted on Reply
#48
freeagent
5900xDdr4AM4LyfeI have used 12 ge, 13,14, all 900k f, etc and the ryzen 9's from the zen3 am4, am5 9000 9900x, 7800x3d, and maybe more for extended peridions on multiple applications. am4 is asolutely not "Slow" lol. throwing money and building many differnt or useing diffferent configurations of AM4 vs a few other AM5 is not a check box that you filld somehow all of the sudden making you the expert to declare the truth. Truth is..... Its not slow and the AM5 is obviously faster, but am4 still handles literally 90 percent of the possible workloads available to it unless you're using hash rate and folding and other non classified as benchmarking normal use cased work loads. the difference is not negligable y any meens, but its still competetive with AM5 in a mid/igh tier vs am4's highest tier zen 3 5900 950 chips.. That is am4 reardless, is amazingly competetive givent the amount of time they have ben actively relavant andi honestly wouldnt be suprised if AMD came out with some software instrutionset breakthroughs available to the am4 that would make them even more competetive is they took the time rather than r&D for new tech and h/w dev.
That is just your opinion, not fact.

Lets look at your 5900X for instance. Chances are, you are not running it like how I was. 260/180/240 +200 -25AC. At full multicore PBO load, 5900X can do 4500MHz with Linpack. 9900X at 310/210/240 can do the same work at 5400MHz. The shear clock speed bump, then IPC is what helps make AM5 so fast. I do not hate AM4, it had a good run, heck I spent a lot of money on AM4. But.. AM5 is faster. Only 3-4 cores on my 5900X can do 5150, and max average effective clock is usually around 4600-4700MHz. My 9900X is 1GHz faster, for sure. And instead of only a few cores being able to hit top speed, almost the entire CCD0 hits 5850, and average effective is 5650.
Posted on Reply
#49
_roman_
It depends if you are multicore user.

when you argue with a 12 or 16 cor eryzen 5000 series cpu, you are a multi core user.

Gentoo toolchian wise my previous ryzen 5800x, a 8 core processor is nearly identical, but a bit slower as my ryzen 7600x, a 6 core without DRAM tuning. I did compare in the past a heavily tuned 5800x vs a out of the box experience Ryzen 7600x with a very bad motherboard choice.

Ryzen 5000 is not bad when you have a high core count, 12 or 16 core processor and you are using multi core software. Single Core Wise you may loose without checking 5800X3d vs 7600x.

I bought the cheapest only offer in may 2023, a ryzen 7600x. That placeholder cpu does a very good job.
Posted on Reply
Add your own comment
Nov 22nd, 2026 16:11 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts