• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Discussion Qualcomm Snapdragon Thread

Page 131 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
That's interesting. Gerard Williams of Qualcomm is on stage at Hot Chips and says the Oryon CPU inside the Snapdragon X Elite has three clusters of cores (we knew this) but says one cluster is "operated in an efficient manner" and the other two are "fully performant."
Indeed, this is why Snapdragon X Elite identifies itself as 8P+4E in software such as Geekbench.
Qualcomm PR says that Williams "misspoke in the moment."

"As you know, our performance cores are more efficient than many efficiency cores from the competition; as we scale performance on our cores, we optimize greatly for power efficiency as well
That's true lol. You know Intel has a problem when there "E-cores" pull more than 10W.

Still, having Oryon E-cores would be beneficial for Snapdragon. We'll have to see how fares the Snapdragon 8 Gen 4.
 
Chips&Cheese coverage of Hot Chips 2024's Qualcomm Oryon presentation!

There's some new slides here which Qualcomm didn't reveal during their first Snapdragon X architecture disclosure.

I think Qualcomm’s characterization of the L2 cache is accurate. Intel’s E-Cores also share a L2 across a quad core cluster, but use 2 to 4 MB of L2 for the whole cluster. Oryon’s L2 is high capacity in that respect. Crestmont cores in Meteor Lake also have 20 cycle latency to L2, making Oryon’s L2 quite impressive. It has 3-6 times as much capacity, yet maintains the same cycle time latency while being able to clock higher
That's the impressive thing about Qualcomm (and Apple's) cache hierarchy. It also eliminates the need for an L3.
 
The "efficiency" Intel is seeking with their E cores isn't the power efficiency Apple seeks for its E cores. Intel is seeking area efficiency to drive MT throughput with theirs. Different goals, different power profile.
It is for Lunarlake. They just used it to try to beat the competition at all costs, that's why it runs at an inefficient range.
 
That's the impressive thing about Qualcomm (and Apple's) cache hierarchy. It also eliminates the need for an L3.
Intel used to be absolute leaders in memory hierarchy with the densest, lowest latency caches and fast memory controllers supporting the latest standards.

The brain drain that occurred over the years mean they are working in other companies now.
 
That's the impressive thing about Qualcomm (and Apple's) cache hierarchy
It's literally the old pre-Nehalem Intel stuff (the defining part of IDC designs was that fat shared L2! Dothan had like 1M of L2 which for the time was hueg).
Passable for client parts but obviously won't fly in DC at all.
 
Davinci Resolve finally arrives to Windows On ARM
Finally is a bit harsh given it's a pretty sizable suite of software to port.

It's a video editor, a color grader, a sound editor, a compositor and can handle plenty of animation style stuff.

Essentially it's equivalent to multiple different Adobe suites in one.

Unfortunately the technical debt owed to combining all that functionality together and adding more and more features does mean it's typically been pretty unstable.

Interesting how PC World couches the release:


I wonder if it going native first will eat a chunk of Adobe's business.
 
Finally is a bit harsh given it's a pretty sizable suite of software to port.

It's a video editor, a color grader, a sound editor, a compositor and can handle plenty of animation style stuff.

Essentially it's equivalent to multiple different Adobe suites in one.

Unfortunately the technical debt owed to combining all that functionality together and adding more and more features does mean it's typically been pretty unstable.

Interesting how PC World couches the release:


I wonder if it going native first will eat a chunk of Adobe's business.
I think Qualcomm needs a better iGPU if want this platform to used by pros.

Resolve uses the GPU a lot.
 
Why wait for Mediatek?
"As little as $700" is not going to fly for a lot of people.

Qualcomm are greedy bugs and always have been.

This was exactly my concern over their attempts to take significant market share.

Even their lower end SoCs are expensive.
 
"As little as $700" is not going to fly for a lot of people.

Qualcomm are greedy bugs and always have been.

This was exactly my concern over their attempts to take significant market share.

Even their lower end SoCs are expensive.
This is just myth though, for years now Qualcomm is producing better and competitive low end soc while Mediatek is recycling old tech, in lowend Mtk still using A75 and A76 while Qualcomm at minimum you get cortex A78 even in sub $100 phones.

In midrange Qualcomm is using only new cores, 7s gen 3 has A720, 8s gen 3 has both X4 and A720 same to 7+ gen 3. What does mtk have? Nothing to compete at best you get A715 which isn't much better than A78, heck mtk has some soc with just 2 big cores like Dimensity 7200 competing in $300 price point.
 
This is just myth though, for years now Qualcomm is producing better and competitive low end soc while Mediatek is recycling old tech, in lowend Mtk still using A75 and A76 while Qualcomm at minimum you get cortex A78 even in sub $100 phones.

In midrange Qualcomm is using only new cores, 7s gen 3 has A720, 8s gen 3 has both X4 and A720 same to 7+ gen 3. What does mtk have? Nothing to compete at best you get A715 which isn't much better than A78, heck mtk has some soc with just 2 big cores like Dimensity 7200 competing in $300 price point.
I wish I could say I'm surprised, but I'm not.

I think that ARM have gone nuts on licensing costs for post v8-A cores.

The fact that all the lower end SoC ODM's seem to be something like 4-6 generations behind state of the art IP in the lower end tells a nasty tale.

Amlogic's A76 SoC came even later than RK3588 and on a cheaper node to boot.

Despite how late RK3588 was still not a peep about any future successor.
 
Back
Top