yfn_ratchetCan't wait to see the 385 in a handheld with unified memory. Would absolutely kill it if those specs managed to make it under the 15W line.

Standard Strix Point with 12 cpu cores (only 4 of which are Zen 5) and 16CU's runs at 15-54W TDP. So good luck with that one.

GhostRyderCool, I can't wait to see some iGPU tests on these!

I still think LPDDR5 8533X will still be insufficient to feed the 40CU's even with 256 bit bus. Hopefully these systems start at 32GB and can be configured with 48/64/96GB.

Gooigi's Exan 8 core with 32cu, UNDER 15 watts is a tall order, but with 24 CUs and with lots of bandwidth...it's doable

Halo is full Zen 5 cores isn't it, not Zen 5c cores. I find it impossible to believe even lowest tier Halo with 32CU's and 8 Zen 5 cores will be able to come in U class 15W TDP range. 28W will be a miracle as the iGPU will be clocking so low it'll probably get whooped by regular Strix Point.

Minus InfinityI still think LPDDR5 8533X will still be insufficient to feed the 40CU's even with 256 bit bus. Hopefully these systems start at 32GB and can be configured with 48/64/96GB.

Has leak from around 3 months ago for Strix Halo test with 128GB RAM.

I wonder how much power these would use, so whether they could fit handheld gaming devices (Steam Deck 2, perhaps?), or if they're laptop only parts.

AusWolfI wonder how much power these would use, so whether they could fit handheld gaming devices (Steam Deck 2, perhaps?), or if they're laptop only parts.

Around 130W is what i've read.

Evildead666Around 130W is what i've read.

That rules out handhelds, unfortunately. :( Still impressive specs, though.

Some more info:

Looking at the SKUs, the AMD Ryzen AI Max "Strix Halo" family will feature the Ryzen AI Max+ 395 as the flagship offering, and that will be equipped with 16 Cores based on the Zen 5 architecture & 40 Compute Units based on the RDNA 3.5 GPU architecture. This chip is going to be the fastest one in the stack and will feature TDPs of up to 130W as revealed in earlier leaks. The chip will also come with four-channel LPDDR5X memory support.

So 4 channel memory support…very interesting.

It’s targeting high end gaming laptops and who know, maybe Valve uses this as a new SteamBox. Steam Machine

Source.

Also, early Geekbench result from a HP laptop:

wccftech.com/amd-strix-halo-ryzen-ai-max-12-core-apu-spotted-hp-zbook-ultra-mobile-workstation/

Neo_MorpheusSome more info:


So 4 channel memory support…very interesting.

It’s targeting high end gaming laptops and who know, maybe Valve uses this as a new SteamBox.

Source.

Also, early Geekbench result from a HP laptop:

wccftech.com/amd-strix-halo-ryzen-ai-max-12-core-apu-spotted-hp-zbook-ultra-mobile-workstation/

4-channel memory makes sense with so much oomph. I'd say it's pretty much necessary for such a beast of an iGPU.

HisDivineOrderBut the minipc market is going to blow up with some incredibly performing devices with these chips either way in the near term. I can't wait to see what they can do.

I sure hope so!

ymdhisHere's hoping these will make it to AM5 socket, even if it means having to pair them with slower memory.

Probably will not be significantly faster than Strix Point if it does. That is not just slower memory, but only half as many channels as well. That is despite AMD only utilizing ~66-75% of the maximum theoretical memory bandwidth of the socket, due to IF bottlenecks.

A hypothetical new quad-channel(or octa-channel DDR5) socket would be nice to hope on. Won't be cheap and probably will cannibalize non-Pro Threadrippers, but one could hope.

yfn_ratchetCan't wait to see the 385 in a handheld with unified memory. Would absolutely kill it if those specs managed to make it under the 15W line.

In another two or three nodes, hopefully, if it ever scales to that.

Detractors are reminded to recall how far integrated graphics had come since things like Intel GMA or Nvidia nForce, or even relatively modern things like Tiger Lake or Zen 3 APUs.

Darmok N JaladCrazy how Apple’s chip naming scheme has been adopted by AMD and Intel. I know terms like Ultra and Max aren’t exactly unique, but neither have been used in a long time. Now Intel is in Ultra mode and AMD is selling a Max chip.

Anyway, this could be really fun in a Mac mini-like encolsure. RAM will have to be soldered, I'd think, but the storage could be expandable. I'd give one a serious look for the right price.

I seem to recall how AMD used to do equivalent-frequency naming schemes like the Athlon 64 3000+. Things used to be much simpler.

A Mac Mini-like won't have a much wider thermal envelope than a handheld. A proper laptop or a SFF appliance with proper cooling, 96+GB of memory, and a more reasonable price than Pro-line Macs would probably do it for me.

EDIT: I was thinking a Macbook Mini for some reason.:oops:

Neo_MorpheusIt’s targeting high end gaming laptops and who know, maybe Valve uses this as a new SteamBox.

I was thinking that myself. Why not a new Steam Machine? I guess it will depend on pricing and availability more than anything.

Minus InfinityHalo is full Zen 5 cores isn't it, not Zen 5c cores. I find it impossible to believe even lowest tier Halo with 32CU's and 8 Zen 5 cores will be able to come in U class 15W TDP range. 28W will be a miracle as the iGPU will be clocking so low it'll probably get whooped by regular Strix Point.

This will be a proper 45+W chip that will make workstation laptop great again (pardon the pun). One issue that plagues current laptop design is the insistence on using a dedicated GPU on most machines even if they wouldn't need one. You can't get an AMD Phoenix Point laptop without a discrete GPU included, just like you can't get an Intel HX laptop without one. To makes matters worse hybrid graphics was interesting when it was invented but windows quickly made it such a bad idea, I'm yet to have a laptop where the system works correctly

With such a big graphics component maybe, just maybe, we'll finally get good laptop designs with the best CPUs on the market and with no discrete GPU attached to fulfill sales quotas.

Neo_MorpheusSo 4 channel memory support…very interesting.

Yes and no, DDR5 made things confusing. Is it 4x 64bits regular channels (thus 8x 32bits channels with how DDR5 works) or it's 4x32bits channels which is the same as 2x64bits we always had with DDR4?

4x32bits channels is faster than 2x64bits channels, but claiming 4 channel memory support is misleading, they're just 2 channels each with 2 subchannels, just like claiming ECC is now standard because DDR5 added basic onchip error correction.

edit: it's LPDDR5X so not even 4x32bits, it will be 4x16bits. Nothing to be excited about, quad channel has become meaningless marketing :(
edit2: they do mention 256-bit LPDDR5X-8000, that would really be 4 channel 4x64 memory and fucking great!

ZareekI was thinking that myself. Why not a new Steam Machine?

No coffee makes you forget things, thanks for reminding me of the correct name.

I think they are still scared of doing that due to failure of the first batch.

trsttteedit2: they do mention 256-bit LPDDR5X-8000, that would really be 4 channel 4x64 memory and fucking great!

Was thinking about your first comment but your edit makes the news even better.

JWNoctisProbably will not be significantly faster than Strix Point if it does. That is not just slower memory, but only half as many channels as well. That is despite AMD only utilizing ~66-75% of the maximum theoretical memory bandwidth of the socket, due to IF bottlenecks.

A hypothetical new quad-channel(or octa-channel DDR5) socket would be nice to hope on. Won't be cheap and probably will cannibalize non-Pro Threadrippers, but one could hope.


In another two or three nodes, hopefully, if it ever scales to that.

Detractors are reminded to recall how far integrated graphics had come since things like Intel GMA or Nvidia nForce, or even relatively modern things like Tiger Lake or Zen 3 APUs.


I seem to recall how AMD used to do equivalent-frequency naming schemes like the Athlon 64 3000+. Things used to be much simpler.

A Mac Mini-like won't have a much wider thermal envelope than a handheld. A proper laptop or a SFF appliance with proper cooling, 96+GB of memory, and a more reasonable price than Pro-line Macs would probably do it for me.

EDIT: I was thinking a Macbook Mini for some reason.:oops:

I guess more Mac Studio, which has the footprint of the current mini, but with added height for more cooling capacity. Really, something similar to what cools the current gen of consoles I'm sure will be sufficient.

Is it chiplet design? If so, they have always had high idle power consumption so far.

I wonder how AMD is going to price this chip, and where we'll find it available.

Its idle power consumption is going to be bad since it's a chiplet design, its power draw is way higher than Apple's Max chips, and even with a 256-bit bus and LPDDR5X it's still going to have way less memory bandwidth than a Max chip, and on the level of a Pro chip, which is a 30~40W chip, against Strix Halo's up to 120W.
Even the M3 Max is a 70W chip with way more memory bandwidth.

CPU-wise it may trounce over them (with 16 fat Zen 5 cores at 100W+ you better do so anyway), but the GPU may still be a bit lacking.

For the folks hoping for an AM5 version, it's a no go, the IOD (which has the GPU) is only compatible with 256-bit LPDDR5X. It's also a bad pick for handhelds due to the bad power consumption, but we shall see if any manufacturer tries to do one anyway.

It's going to be a great CPU for a mini-PC tho.

EDIT: adding some info for comparison sake:


Strix Halo with LPDDR5X-8533 and its 256-bit memory bus should be doing around 273GB/s. This puts it somewhere in between the desktop RX7600 and the RX7700, and really close to the mobile RX7700 (a bit less bandwidth, but 8 more CUs).

igormpIts idle power consumption is going to be bad since it's a chiplet design

It's a monolithic chip IIRC.

R0H1TIt's a monolithic chip IIRC.

I believe most of the mobile parts/APU's (not 7000/9000* series with basic IGP) are monolithic, this is also why they can acheive much higher memory speeds compared to their desktop counterparts

R0H1TIt's a monolithic chip IIRC.

So far most rumours point to this not being the case. Even in the OP that's the mention:

Nomad76The Strix Halo APU is expected to use two CCDs (Core Complex Dies) with 8 cores each. One variant is rumored to have just a single CCD

And CCDs allude to it being chiplet based.

From another rumour site:
wccftech.com/amd-strix-halo-ryzen-ai-max-apu-skus-leak-ryzen-ai-max-395-16-cpu-cores-40-gpu-cores/

AMD Ryzen AI HX Strix Halo Expected Features:

• Zen 5 Chiplet Design
• Up To 16 Cores
• 64 MB of Shared L3 cache
• 40 RDNA 3+ Compute Units
• 32 MB MALL Cache (for iGPU)
• 256-bit LPDDR5X-8000 Memory Controller
• XDNA 2 Engine Integrated
• Up To 60 AI TOPS
• 16 PCIe Gen4 Lanes
• 2H 2024 Launch (Expected)
• FP11 Platform (55W-130W)

At first I also had hoped for it to be monolhitic, but that doesn't seem to be the case (and it'd be a hecking big chip to begin with).

Marcus LI believe most of the mobile parts/APU's (not 7000/9000* series with basic IGP) are monolithic, this is also why they can acheive much higher memory speeds compared to their desktop counterparts

Yeah, most are monolithic, but their high-performance models are chiplet-based. As an example, Dragon Range (7000 HX mobile lineup) is chiplet-based, being basically the same as the desktop chips, with the same IO die even (DDR5 and no LPDDR5, only 2 CUs for the iGPU):

en.wikipedia.org/wiki/List_of_AMD_Ryzen_processors#Dragon_Range_(7045_series,_Zen_4/RDNA2_based)

Strix Halo seems to follow a similar part (using the desktop CCDs), but with a WAAAYYY better IOD this time (256-bit LPDDR5X + 40CU iGPU).

igormpEven in the OP that's the mention:

It doesn't mention anything like that. By "chiplet" design, WTFtech probably meant a single chiplet, or they don't know what they're talking about!

igormpAs an example, Dragon Range (7000 HX mobile lineup) is chiplet-based, being basically the same as the desktop chips

That's because they're literally desktop chips. You remember the 2700 (non-X) in some laptops about 6~7 years ago?

igormpStrix Halo seems to follow a similar part (using the desktop CCDs), but with a WAAAYYY better IOD this time (256-bit LPDDR5X + 40CU iGPU).

They'd have to design an entire new I/O die for that and a new CPU die/chiplet as well.

As with Zen 2, each CCD connects to the IO die through an Infinity Fabric link. On desktop, this link is 32 bytes per cycle in the read direction and 16 bytes per cycle in the write direction. That differs from AMD’s mobile parts, where the Infinity Fabric link from a core cluster can do 32 bytes per cycle in both directions. Infinity Fabric runs at 2 GHz on both setups, just as it did on desktop Zen 4. That’s not a surprise, since AMD has re-used Zen 4’s IO die for Zen 5. At that clock speed, each cluster has 64 GB/s of read bandwidth and 32 GB/s of write bandwidth to the rest of the system.

chipsandcheese.com/2024/08/14/amds-ryzen-9950x-zen-5-on-desktop

I have high hopes for this chip, can't wait to see 40 CU and more memory bandwidth in action, there is so much promise here.

The names though? My lord.

Ryzen AI Max+ 395
Core Ultra 9 285K

Can y'all that decide these names get a damn grip?

R0H1TBy "chiplet" design, WTFtech probably meant a single chiplet, or they don't know what they're talking about!

I honestly doubt it. Anyhow, do you have any source saying it's a monolithic design, other than wishful thinking?
Here are other sources reinforcing that rumor:

The "Strix Halo" silicon is a chiplet-based processor [...] "Strix Halo," on the other hand, use the same one or two "Zen 5" CCDs, but with a large SoC die featuring an oversized iGPU, and 256-bit LPDDR5X memory controllers not found on the cIOD.

www.techpowerup.com/321693/amd-strix-halo-zen-5-mobile-processor-pictured-chiplet-based-uses-256-bit-lpddr5x

The two CPU dies are known as CCDs (core chiplet dies) and each contains eight of the company’s Zen 5 CPU cores – thus the maximum 16-core configuration. These CCDs apparently have a relatively modest size of 7.327 x 9.055mm or 66.3mm2 but the GPU is something else entirely.

The Strix Halo GPU purportedly measures a massive 16.02 x 19.18mm or 307mm2[...]

www.pcgamesn.com/amd/strix-halo-guide

videocardz.com/newz/amd-strix-halo-zen5-apu-leak-shows-307mm%C2%B2-die-with-rdna3-5-graphics-compared-to-rtx-4070-80w

R0H1TThey'd have to design an entire new I/O die for that and a new CPU die/chiplet as well.

That's the point, strix halo's main selling point is a new IOD (which has the 40CU GPU and the 256-bit memory controller). The CPU dies, however, are your good old 8 cores Zen 5 CCD found on the desktop chips. There's no need to redesign the CCDs themselves.

Strix Halo is... not enough "Halo" in CPU performance.
www.tomshardware.com/pc-components/cpus/amds-strix-halo-zen-5-apu-tested-in-geekbench-ai-benchmark-ryzen-ai-max-390-sample-falls-behind-ryzen-7-7840hs

igormpThe CPU dies, however, are your good old 8 cores Zen 5 CCD found on the desktop chips. There's no need to redesign the CCDs themselves.

There's no NPU on desktop zen5 IIRC, how are they going to sell them for Windows laptops without a dedicated NPU missing?

This is assuming they're at least targeting high end laptops, which may not be true of course.

R0H1TThere's no NPU on desktop zen5 IIRC, how are they going to sell them for Windows laptops without a dedicated NPU missing?

This is assuming they're at least targeting high end laptops, which may not be true of course.

The NPU goes in the IO die, which, as I said before, is totally new.

TumbleGeorgeStrix Halo is... not enough "Halo" in CPU performance.
www.tomshardware.com/pc-components/cpus/amds-strix-halo-zen-5-apu-tested-in-geekbench-ai-benchmark-ryzen-ai-max-390-sample-falls-behind-ryzen-7-7840hs

...in AI performance, running on the CPU only. I dunno, but if I had a SOC with an NPE AND a massive IGP, the last place I'd elect to run an AI workload is the CPU. How CPU-only AI performance translates to other applications is certainly up for debate, but it seems outside the main purpose of such a product.