News Next-gen Snapdragon X2 chips for PCs to boost core count from 12 to 18, says report

Let's be real, only good gpus (and low prices) will give ARM a "significant" boost on windows. Usage of latest nodes and core gens by Intel and AMD pretty much narrowed and leveled most scenarios. Means ARM stays uninteresting for "most" customers, with minor exceptions. However, suppliers would love to have ARM onboard, putting pressure on ASPs in general.
 
  • Like
Reactions: usertests
On-package DRAM is good news, but putting NAND on there is just evil. NAND doesn't like high temperatures and stands to gain nothing from such tight integration, due to having read latencies like 2+ orders of magnitude higher than DRAM. Worse yet, this makes it non-upgradable/replaceable, rendering all laptops with it that much more disposable. Yuck.

I sure hope these cores at least have SVE!
 
Last edited:
Usage of latest nodes and core gens by Intel and AMD pretty much narrowed and leveled most scenarios.
Apple still has a pretty solid lead over even the latest x86 laptop CPUs. That shows x86 has not closed the ISA gap.

An even better comparison is with the old M3's, because those are made on the exact same node as Intel is using for Lunar Lake and Arrow Lake. This page only has the M3 base model, but it still smashes all of the x86 contenders on single-threaded performance and efficiency, in spite of the fact that it doesn't have SVE:

Snapdragon X used an older process node than the past few Apple SoCs (equivalent to AMD's Zen 5 node) and is also a first gen Oryon core. I'm sure there were things they wanted to do in those cores/SoCs that had to get dropped from the schedule, which makes me very excited to see how these V3 cores look (note that the V2 cores were exclusive to phone SoCs).
 
Last edited:
I too fail to see the point of on package NAND and it doesn't seem like something which would make sense for Qualcomm either. This would serve to raise the packaging cost and limit capacity while providing no tangible benefit. While I don't particularly like on package DRAM there are benefits and so long as it doesn't limit them to two packages should be fine usability wise.

As far as the core designs are concerned it'll be interesting to see how much benefit comes from IPC vs clock speed. A supposed core count of 18 also makes me wonder what the clusters, or if there are any, will look like since current are 4.
 
As far as the core designs are concerned it'll be interesting to see how much benefit comes from IPC vs clock speed. A supposed core count of 18 also makes me wonder what the clusters, or if there are any, will look like since current are 4.
Do we have any word on whether they now have a true E core? That could also cast the core counts into an interesting light.
 
I hope they increase the memory bandwidth and improve GPU performance this time around.
192-bit (triple channel) or better, please.

Apple's M-series GPU is meh, but the UMA sure does wonders with a 256 ~ 512-bit memory bus on the Pro and Max models.
I don't expect 410GB/s of memory bandwidth, but please hit at least 200GB/s.

Otherwise another swing and miss from Qualcomm, and I'll look forward to what NvidiARM and ARMD has to offer from their skunk works.
 
  • Like
Reactions: bit_user
Do we have any word on whether they now have a true E core? That could also cast the core counts into an interesting light.
I've seen no concrete evidence one way or the other on the design, but there haven't been any rumors about different core architectures. It's possible they're just following in the footsteps of phone where they are using a 2 "prime" performance core and 6 performance core cluster design.

I hadn't looked into the phone chip design as I didn't recall seeing anything at the more tech oriented sites, but it's not as straightforward as I expected. The 2 prime cores actually have more L1 cache (192KB vs 128KB) and L2 (both clusters have 12MB total) per core on top of the higher boost clock speeds available.
 
I can't read German, but the WinFuture article uses the phrase "den Speicherchips der SSD". Some googling indicates that Speicherchip ("memory chip") is a general term covering RAM, flash, EEPROMs, maybe even a mask ROM.

The surrounding paragraph also seems pretty speculative, i.e. it interprets "SIP" to mean "system-in-package", then lists RAM and secondary storage as things a SIP tends to include. The bit about a 1 TB NVMe SSD is in an earlier paragraph describing the test system.

But let's assume Qualcomm is putting Speicherchips on-package. My interpretation is one or both of the following:
  1. A bootloader or "restore to factory settings" ROM. Some flash memory (such as automotive EEPROMs) is rated for high temperatures, e.g. up to 200°C. They're not available in capacities anywhere near 1TB, but enough for a bootloader & firmware.
  2. SSD memory controller and/or DRAM cache. Apple's M-series chips include the former, not sure about the latter.
 
I hope they increase the memory bandwidth and improve GPU performance this time around.
192-bit (triple channel) or better, please.

Apple's M-series GPU is meh, but the UMA sure does wonders with a 256 ~ 512-bit memory bus on the Pro and Max models.
I don't expect 410GB/s of memory bandwidth, but please hit at least 200GB/s.

Otherwise another swing and miss from Qualcomm, and I'll look forward to what NvidiARM and ARMD has to offer from their skunk works.
I read a few months ago this third gen will be massively improving GPU performance by like 200% or so. Yes it's technically third gen as 2nd gen is for phones/tablets.
 
  • Like
Reactions: Notton