Possible AMD Prototype Processor with DDR5 Memory Hits BAPCo CrossMark Database

[btarunr]

Quite possibly the first sighting of a next-generation AMD processor with DDR5 memory surfaced on the web. A BAPCo CrossMark Database entry references a prototype processor with the name-string "AMD Eng Sample: 100-000000560-40_Y," running on a platform titled "ASUSTeK COMPUTER INC. M3402RA." The chip has 16 GB of memory across 2 memory channels, and a memory frequency of 4800 MHz DDR. The platform here could be either a desktop motherboard, or a notebook. 4800 MHz is an unusual memory speed for a mobile platform, unless it's a single stick of DDR5-4800 SO-DIMM, with two 40-bit channels.

The first notebooks with DDR5 memory make landfall early next year, when Intel launches mobile variants of its 12th Gen Core "Alder Lake" processors. This would mean that DDR5 SO-DIMMs are already in circulation with OEMs. If the theory of this being a mobile chip holds true, it could very well be the "Rembrandt" APU that combines "Zen 3+" CPU cores with an iGPU based on the RDNA2 graphics architecture. If however the platform is a prototype Socket AM5 motherboard, it could be one of the first sightings of a next-generation "Raphael" desktop processor with "Zen 4" CPU cores, and a combination of DDR5 memory and PCI-Express Gen 5.



View at TechPowerUp Main Site

 

[ncrs]

4800 MHz is an unusual memory speed for a mobile platform, unless it's a single stick of DDR5-4800 SO-DIMM, with two 40-bit channels.

How so? Alder Lake shows as 2 channels as well despite being "quad channel DDR5" (2x2x32-bit).
And as expected the DDR5 channels saga continues to be confusing for everyone

 

[usiname]

How so? Alder Lake shows as 2 channels as well despite being "quad channel DDR5" (2x2x32-bit).
And as expected the DDR5 channels saga continues to be confusing for everyone

32+8bit for ECC

 

[ncrs]

32+8bit for ECC

Yeah, I know that, but ECC is not really that common in laptops.
I was questioning the "single stick of DDR5-4800" because on Alder Lake dual sticks is also shown as "2 Channel(s)" instead of 4 in that database.

 

[usiname]

Yeah, I know that, but ECC is not really that common in laptops.
I was questioning the "single stick of DDR5-4800" because on Alder Lake dual sticks is also shown as "2 Channel(s)" instead of 4 in that database.

Ah I didn't get you the first time. Its written "Desktop" in the benchmark page, most likely is 2x8gb DDR5, but I don't know, its so strange to see so many single channel 12900k systems

 

[Valantar]

How so? Alder Lake shows as 2 channels as well despite being "quad channel DDR5" (2x2x32-bit).
And as expected the DDR5 channels saga continues to be confusing for everyone

As this is how Windows, Intel, and everyone else reports things, I assume BapCo to do the same - ironically, to avoid confusion

@btarunr This seems pretty universal at this point (check Task Manager on any system with 32-bit channels, whether DDR5 or LPDDR4X: "Channels" when not speaking on a deep technical level equals "64-bit-ish channels or aggregate channels", which in DDR5 terms means a single DIMM with 2x32-bit (+2x8 for ECC if applicable) has one "channel" in these descriptions. So, "dual channel" in this scenario likely means 2 "channels" of 2x32-bit DDR5.

 

[Caring1]

As I mentioned on another site when I first read this leak about 12 hours ago at least, I doubt it is an APU as it uses the Microsoft basic display adapter driver, not an AMD driver for graphics.

 

[ncrs]

As I mentioned on another site when I first read this leak about 12 hours ago at least, I doubt it is an APU as it uses the Microsoft basic display adapter driver, not an AMD driver for graphics.

If it's an unreleased APU it wouldn't get drivers automatically from Windows Update which would result in using the basic adapter driver. If it's not an APU then it most likely has some sort of other video adapter present and chances are it's one of the released ones, which would pull drivers automatically and not be identified as a basic adapter. Of course that's assuming default Windows configuration, because automatic driver installation can be turned off.
However, this is all speculation

 

[silentbogo]

If the theory of this being a mobile chip holds true, it could very well be the "Rembrandt" APU that combines "Zen 3+" CPU cores with an iGPU based on the RDNA2 graphics architecture.

It's an asus Vivobook 14 OLED. Screen resolution is also a dead giveaway. Previous model had board ID M3401.
There are also many speculations that it'll have Navi2 graphics.

 

[Mussels]

It's an asus Vivobook 14 OLED. Screen resolution is also a dead giveaway. Previous model had board ID M3401.
There are also many speculations that it'll have Navi2 graphics.

Good info, detective bogo

 

[IceShroom]

It's an asus Vivobook 14 OLED. Screen resolution is also a dead giveaway. Previous model had board ID M3401.
There are also many speculations that it'll have Navi2 graphics.

It will be waste of RDNA2 design if it is like the current Vivobook.

 

[Valantar]

It will be waste of RDNA2 design if it is like the current Vivobook.

How so? The Vivobook OLED seems to have been well received from what I can tell.

 

[IceShroom]

How so? The Vivobook OLED seems to have been well received from what I can tell.

Dedicated gpu, this how is it waste of RDNA2 design. If OEM going to use dedicated gpu then 2-4CU Vega is better for die size, yield and cost.

 

[Valantar]

Dedicated gpu, this how is it waste of RDNA2 design. If OEM going to use dedicated gpu then 2-4CU Vega is better for die size, yield and cost.

Ah. Yeah, I agree on that point - unless it's a much more powerful dGPU (which I wouldn't expect in a 14" thin-and-light) I'd want to stick with the iGPU for sure.

 

[silentbogo]

It will be waste of RDNA2 design if it is like the current Vivobook.

I wouldn't go as far as to call it a "waste". I don't think there will be any super-powerful Navi iGPUs either, so it'll do just as good (or as bad) as anything we have today, only smaller and more power-efficient.
Just something to get you through YT or Spider Solitaire, while 3050 is taking a smoke break.

 

[Valantar]

I wouldn't go as far as to call it a "waste". I don't think there will be any super-powerful Navi iGPUs either, so it'll do just as good (or as bad) as anything we have today, only smaller and more power-efficient.
Just something to get you through YT or Spider Solitaire, while 3050 is taking a smoke break.

What makes you think it will be more power efficient and not faster? Within any given power envelope in computing, efficiency is mainly used to improve performance. It is exceedingly rare that something has its power budget cut because it's more efficient than its predecessor.

Given that current Vega APUs are plenty usable for gaming I would expect RDNA2 APUs (especially with DDR5) to deliver a significant performance increase, especially as RDNA2 massively improves performance/TFLOP over Vega. At the same clocks and CU count and assuming the RAM can feed both equally, any RDNA2 GPU should outperform a Vega GPU by ~50%.

 

[silentbogo]

What makes you think it will be more power efficient and not faster? Within any given power envelope in computing, efficiency is mainly used to improve performance. It is exceedingly rare that something has its power budget cut because it's more efficient than its predecessor.

My theory is old and boring AF - there's no competition. Intel Xe is good, but still fell short. UHD770 is underwhelming, and newer iterations of Xe didn't promise any significant improvements [yet]. Also, at its current state DDR5 is only marginally faster than DDR4, which won't help with iGPU gaming performance.
Another not-so-crazy thought, is that it'll cut into their low-end dGPU segment. They've already "leaked" few low-end desktop cards(6400/6500), and Navi 24 mobile cards aren't on the market either.

 

[Valantar]

My theory is old and boring AF - there's no competition. Intel Xe is good, but still fell short. UHD770 is underwhelming, and newer iterations of Xe didn't promise any significant improvements [yet]. Also, at its current state DDR5 is only marginally faster than DDR4, which won't help with iGPU gaming performance.
Another not-so-crazy thought, is that it'll cut into their low-end dGPU segment. They've already "leaked" few low-end desktop cards(6400/6500), and Navi 24 mobile cards aren't on the market either.

Yeah, I don't think I subscribe to that. Both because 96EU Xe is purely held back by drivers (it beats 5000-series Vega 8 in anything where its drivers are even half-decent), and because the increases from DDR5 are larger than you describe. Sure, DDR4 >4000 is widely available, but very few people actually have that, especially paired with an APU - RAM that fast is expensive. And while newer Ryzen APUs are great for RAM OCing, that is a tiny niche activity.

A more valid comparison would be at JEDEC specs, and the highest end JEDEC DDR4 - 3200 - is beaten in bandwith by 50% by even relatively low-end DDR5-4800 (it's not the lowest clocked DDR5 there is, but it's a mid-range speed in the first generation of DDR5, and JEDEC is projecting standards up to DDR5-8400). Even accounting for common XMP kits, something like DDR4-3600, 4800 is a 33% increase. Meaning that for the vast majority of APU builds, DDR5 is a major improvement.

Remember, desktop APUs are relatlively unimportant compared to mobile, and even LPDDR4X-4266 is trounced by first-gen LPDDR5-5500 like in the Steam Deck. For non-LP laptops, again JEDEC DDR4-3200 is going up against much higher speed JEDEC DDR5, likely all of it 4000+.