News AMD's Strix Halo being tested with 128GB RAM — shipping records reveal more about extreme 120W APU

[eX_Arkangel]

I can forgive the desktop APU neglect and naming. But if the socket is nerfed by design, there's no fixing the situation for years. And for all we know, AM6 will be the same dual-channel affair, but with a faster DDR6 standard to "justify" it.

I think there might have been technical reasons for not launching Rembrandt desktop APUs on AM4. The big one being that it uses DDR5 memory. When the going gets tough, get it soldered to a mobo. Probably not done cheaply by ASRock but maybe Minisforum and other Chinese companies. That's what I hope to see with Strix Halo, although nobody should kid themselves about the price/perf beating traditional CPU+GPU combos.

thats an IMC change only, even Intel offered both DDR4 and 5 for their current lineup and AMD has done the same in the past, theres also Mendocino and steam deck APU, which used Z2 and RDNA 2 with DDR5 IMC (doing the opposite, upgrading Z2 IMC from D4 to D5), so the technical reason is doubtful at best.

As for the socket, then yes if there is not enough "extra" unused/unmapped pins enough for adding an extra channel then there's no way to fix it aside from on-package memory, AM6 needs to fix this from the get go, and i think they'll go to "at least" 3 ch just like Nehalem LGA 1333, as Intel is clearly catching up, and they (AMD) have lost the node advantage by using TSMC as Intel is now doing the same (even using newer nodes), so they need to go all out with AM6 if they want to keep the market and mindshare in desktop and mobile segments.

They can even split the tiers with the # of channels per chipset, eg: AB tier 2ch, B tier 2ch entry 3ch high, X tier 3ch entry 4ch high, X-E tier 4ch all. Just an example.
Also they need to start doing proper desktop APU's instead of using crippled ultra mobile leftovers, as not only the L3 cache is decimated but PCIe lanes/gen, ports, etc are all skimmed to the bare minimum, and Intel is going all out on APU for both mobile and desktops as it represents a way for them to enter the gpu market from the "backdoor" while managing to get a very wide adoption, specially in the OEM/corp channel.

As competition heats up, prices go down+ tech goes all out (hopefully) in every way possible.

 

[usertests]

AM6 needs to fix this from the get go, and i think they'll go to "at least" 3 ch just like Nehalem LGA 1333

That was an ancient decision for HEDT. I would be surprised to see anyone do 3-channel (max). They should probably aim for 4 channel, 1 DIMM per channel.

Then make a chipset with half the channels intended for Mini-ITX or cheap systems.

Also they need to start doing proper desktop APU's instead of using crippled ultra mobile leftovers, as not only the L3 cache is decimated but PCIe lanes/gen, ports, etc are all skimmed to the bare minimum

They just won't, I think. Not on the socket. You'll have to buy Strix Halo soldered for an amazing APU experience anytime soon.

I think the key to improving the cache in all of AMD's product lines will be to make X3D the default, as a cost-cutting measure. Get all L3 cache or most of it off the expensive node, 3D stack a cache chiplet using an older node. This should also be done with the mainstream mobile APUs.

Hope we get to see that happen as soon as Zen 6.

Intel is going all out on APU for both mobile and desktops as it represents a way for them to enter the gpu market from the "backdoor" while managing to get a very wide adoption, specially in the OEM/corp channel.

I believe Intel is still making the desktop iGPUs a third or half of the size of mobile, so it can't be said that they are going all out on desktop APUs. If Arrow Lake desktop for example includes 64 EUs of Xe1, that will trail far behind 128 EUs of Xe2. Though if they do double it to 64 EUs, they are taking the desktop iGPU more seriously than AMD, which is reusing the same I/O chiplet with 2 CUs RDNA2.

The hundreds of millions of Intel iGPUs in the wild before they started making dGPUs again didn't help. They just needed a lot of time to improve neglected drivers. Some say they are much better now, while others still report weird issues.

 

[thestryker]

thats an IMC change only, even Intel offered both DDR4 and 5 for their current lineup and AMD has done the same in the past, theres also Mendocino and steam deck APU, which used Z2 and RDNA 2 with DDR5 IMC (doing the opposite, upgrading Z2 IMC from D4 to D5), so the technical reason is doubtful at best.

The desktop APUs use literally the same die as mobile, and Rembrandt only lists DDR5/LPDDR5 support. So unless AMD omitted DDR4 support on their public specifications that would mean they couldn't use it on AM4.

 

[eX_Arkangel]

The desktop APUs use literally the same die as mobile, and Rembrandt only lists DDR5/LPDDR5 support. So unless AMD omitted DDR4 support on their public specifications that would mean they couldn't use it on AM4.

Z2 both dt and mobile only used D4, yet they were able to "update" the IMC to D5 for mendocino/SteamDeck SOC, Z3 OG IMC is also D4 both for DT and Mobile, only Rembrandt was "updated" to D5 IMC, but is still a Z3 on N6, so it works both ways, in fact i dare say is easier to go back to original IMC than to update to a new gen on and old product designed with previous gen in mind (Z3-D4).
The reason again is not clear at all, but i still think it has nothing to do with technical reasons.

 

[eX_Arkangel]

That was an ancient decision for HEDT. I would be surprised to see anyone do 3-channel (max). They should probably aim for 4 channel, 1 DIMM per channel.

Then make a chipset with half the channels intended for Mini-ITX or cheap systems.

They just won't, I think. Not on the socket. You'll have to buy Strix Halo soldered for an amazing APU experience anytime soon.

I think the key to improving the cache in all of AMD's product lines will be to make X3D the default, as a cost-cutting measure. Get all L3 cache or most of it off the expensive node, 3D stack a cache chiplet using an older node. This should also be done with the mainstream mobile APUs.

Hope we get to see that happen as soon as Zen 6.

I believe Intel is still making the desktop iGPUs a third or half of the size of mobile, so it can't be said that they are going all out on desktop APUs. If Arrow Lake desktop for example includes 64 EUs of Xe1, that will trail far behind 128 EUs of Xe2. Though if they do double it to 64 EUs, they are taking the desktop iGPU more seriously than AMD, which is reusing the same I/O chiplet with 2 CUs RDNA2.

The hundreds of millions of Intel iGPUs in the wild before they started making dGPUs again didn't help. They just needed a lot of time to improve neglected drivers. Some say they are much better now, while others still report weird issues.

LGA 1333 wasn't really HEDT (at least no workstation level like we see today), as it topped out at 6 cores, with normal i7 920 at 4C, also it launched earlier than lower end nehalem (i5-3 700 series). But aside from that i agree that 3ch would be a cheapskate choice from both of them is they top out there, 4ch is long overdue in desktop, is an ancient tech that for some reason they refuse to implement on DT at least (up til Zen2 there was no real need, but when 16C became mainstream desktop the need was evident) specially with the uprising of heavy multicore CPU's and maybe in some future powerful APU's.

Thats my example of split the ch per chipset in the previous comments 2ch for low end AB- series chipset, 2ch entry B series 3ch top B, 3ch entry X series 4ch top, and 4ch X-E chipset. Mini ITX is weird as some use it with APU's to go really small, and it could benefit from the extra ch, maybe slots in the back of the MB?.

Yeah i thought intel was going harder on DT iGPU's, well see, but so far their offering still beats AMD ones by far in anything but graphics, and thats gonna change with ARL apparently.

Intel iGPU's so far werent meant to be dGPU's so their were basically overestimated display engines with basic drivers etc, after ARC launched and intel's bid for a discrete lineup, iGPU's can be used as a backdoor for mass adoption/ beta testing/polishing new uarch etc, at least in the gaming part, as enc/dec and display engine they are doing fine IMO. So far the perf increases they have gotten trough drivers updates/optimizations are promising, and could put them on a similar ground to AMD with BMG, particularly in the low to mid end and iGPU

 

[thestryker]

Z2 both dt and mobile only used D4, yet they were able to "update" the IMC to D5 for mendocino/SteamDeck SOC, Z3 OG IMC is also D4 both for DT and Mobile, only Rembrandt was "updated" to D5 IMC, but is still a Z3 on N6, so it works both ways, in fact i dare say is easier to go back to original IMC than to update to a new gen on and old product designed with previous gen in mind (Z3-D4).
The reason again is not clear at all, but i still think it has nothing to do with technical reasons.

Okay the point I'm making went right over your head apparently. Rembrandt only has a DDR5 controller, and AMD APUs are literally the same die as the mobile ones. That means if AMD were to use Rembrandt on desktop they would have to do one of two things: have DDR5 support on AM4 or not use the mobile die but rather a new one. Neither of these two things were ever going to happen which is why Rembrandt never came to desktop.

 

[eX_Arkangel]

Okay the point I'm making went right over your head apparently. Rembrandt only has a DDR5 controller, and AMD APUs are literally the same die as the mobile ones. That means if AMD were to use Rembrandt on desktop they would have to do one of two things: have DDR5 support on AM4 or not use the mobile die but rather a new one. Neither of these two things were ever going to happen which is why Rembrandt never came to desktop.

same back to you, completely ignored my point, Z2 only had D4 IMC, but somehow they manage to use D5 IMC for mendocino/Steamdeck, so.... get a glove as a hat to see if you can catch the point im making before it blazes past your head this time.

 

[thestryker]

same back to you, completely ignored my point, Z2 only had D4 IMC, but somehow they manage to use D5 IMC for mendocino/Steamdeck, so.... get a glove as a hat to see if you can catch the point im making before it blazes past your head this time.

You do understand that the IMC is integrated into the die for the mobile/APU parts right?

They didn't "somehow" manage to use a new IMC they just designed the die with a new IMC.

 

[eX_Arkangel]

exactly,

You do understand that the IMC is integrated into the die for the mobile/APU parts right?

They didn't "somehow" manage to use a new IMC they just designed the die with a new IMC.

so glad you finally understood that

 

[usertests]

LGA 1333 wasn't really HEDT (at least no workstation level like we see today), as it topped out at 6 cores, with normal i7 920 at 4C, also it launched earlier than lower end nehalem (i5-3 700 series).

I forgot that it was a separate "enthusiast desktop" tier that was basically eliminated later. But they were around $885 to $999 for those 6 cores.

I'll say this about Intel desktop iGPUs. AMD is reusing the same graphics in Ryzen 7000/9000 because it's there for basic functionality and they don't care. Intel is apparently pushing theirs ahead. Even if they remain slower than AMD's desktop APUs, these are the kinds of chips I'll end up picking up in cheap refurbished office PCs. So when they double/triple the UHD 730/770 performance, that's great for me.