Having more cores per cluster is better?Sounds like >99% of laptop, non HEDT desktop use case to me. The rest 1% are from AMD/Intel marketing, running ideally scaling benchmarks like CB23?
I think their MT score is held back by "SoC" level limitations, like too small L3 cache, IMC getting swamped by too many request. 4C cluster size is not helping either, pretty much means inter cluster core comms take a huge hit and impact performance.
Still, for 1st generation, 3250'ish is GREAT SC GB6 score to be at, MT scaling being not optimal is pretty much expected from a chip with TOTAL cache budget of 42MB ?
According to AT, each CPU core should have 3MB L2 cache total 36MB. Plus 6MB L3 cache; that's slightly less than M2's 8MB.
Ooo. Still ARM v8.7?The isa seems to be stuck at ARM 8.7 though:
Understandable, considering the need to get it out of the door, but still slightly dissapointing.
I'd also prefer even a microcoded implementation of SVE to no SVE at all, especially for a 2024 product
Then what are we going to be using for benchmarking?Geekbench is excellent for ST, not so much MT
Yeah there are plenty of ways to screw this upWhat a breath of fresh air. Even if it's just competitive with Apple by next year, it's still a massive step in the right direction. The rest of the world was put on notice, they better have very good output in years to come.
Hmm. Wouldn't clusters with less cores be better for efficiency? As you can power down an entire cluster if it's not in use.Having more cores per cluster is better?
Hopefully my next PC is gonna be powered by Snapdragon🔥Yeah there are plenty of ways to screw this up
- Lack of OS / application support
- Stupid company policies, artificially limiting products, HW- or software development, collaborations, etc
But there should be plenty of "low hanging fruits" to iterate upon (already apparent of what little we know). Really reminds me of Zen 1
Hmm. Wouldn't clusters with less cores be better for efficiency? So you can power down an entire cluster if it's not in use.
Well, if NVIDIA will indeed start churning out ARM CPUs for PCs that have excellent performance, things might change rather quickly. Maybe this is the snow ball that starts the avalanche.Yeah there are plenty of ways to screw this up
- Lack of OS / application support
- Stupid company policies, artificially limiting products, HW- or software development, collaborations, etc
But there should be plenty of "low hanging fruits" to iterate upon (already apparent of what little we know). Really reminds me of Zen 1
They did mention it, and it seems…. Right in line with (actually beats but at more power likely) Apple and Intel both. 30% more less power than Apple’s ST iso-perf (tho it’s the M2 Max where that’s true).If they don't mention it there's probably a reason.
They are clearly designing to win on MT given their high end having 12 big cores and no little ones.
lol no. x86 isn’t going anywhere.I hope these numbers come out as mentioned across the board. Would either kill x86 or force Intel/AMD to innovate even faster. Anyway this is releasing mid next year and we will have different x86 models to compare. May be AMD will release Zen 5 across the board and Intel will pull in ARL/LNL.
Slower 1T than top of the line desktop silicon from AMD/Intel. The benchmark numbers were very intentional and cherry picked. This chip is not as fast/efficient as they are making it out to be.Yes and no.
Reading between the lines I think that they are very competitive on 1T perf and at good power.
The fact that they didn't specify which laptop SoC's they were comparing against with nT makes me think that they may have been SoC's with majority Intel E/mont cores - which is a valid comparison, but it also makes me think that they don't want to compare to certain AMD configurations (Dragon Range) at similar TDP, plus we have Strix Point coming up next year which isn't exactly going to be a slouch either.
I may be wrong, but the lack of transparency being shown here is out of place with the rest of their claims given this is a first impression they are trying to give with a brand new platform launch (as Elite X is effectively rebooting that segment for them).
Geekbench 6 is not a good benchmark since it doesn’t scale like Version 5 does.This chip has 50% better performance than an M2 in multi-threaded tasks, according to the slide.
Didn't they mean M2 Pro/Max?
Because if their chip scores 3227 in Geekbench single-core and have 12 performance cores, I would expect it to be much more than 50% faster than an M2, which has only 4 performance core scoring ~2800 Geekbench points each.
EDIT: I see the M2 Pro/Max are only about 50% faster than the M2 in Geekbench 6, which appears to be the tool Qualcomm use for their comparisons. Which makes their new chip about as fast as an M2 Pro/Max in multicore tasks, at 30% less power consumption than an M2 Max. Not bad, but we'll see what Apple has to offer with the M3.
Meteor Lake is not listed on the chart for a reason (unreleased) but if it were, it would make this qualcomm offering look “meh”.I think Intel is in big trouble. They don't have efficient laptop chips.
MTL ain't it.
GB5 is a good benchmark. When run properly it is just as good as SPEC. GB6 did away with multicore scaling, so it is useless as a benchmark.Then what are we going to be using for benchmarking?
Is Geekbench no longer the silver standard of benchmarking?
*with SPEC being the Gold standard.
1) True, but no one cares bout desktop nuclear reactor ST. Mobile is where the market is, and that’s what this chip is targeting. It is… 14% faster on ST than an M2 Max and even slightly faster than a 13980HX (mobile and monolithic albeit beefy, 5.6GHZ) chip. 30% less platform power than an M2 Max on ST iso-perf. AMD’s Phoenix lineup is not going to beat QC on peak ST or the curve.My Projections: slower 1T performance than AMD’s best. Perf/watt similar (or slightly worse) to AMD’s best. nT depends on how fast 1T perf is, but will certainly be behind AMD’s best. There is a reason they compared the chip to Intel Raptor Lake, it is one of the least efficient architectures out there when pushed to the limit.
I do think it is a great first attempt, but as usual Qualcomm is over promising. can’t wait to see reviews.
Not really, no.If you make this analysis from geekerwan video it's not a good assumption to make.
GB5 is a good benchmark. When run properly it is just as good as SPEC. GB6 did away with multicore scaling, so it is useless as a benchmark.
Yep Intel is going to in trouble and I doubt MTL is going to fix it.Livestream ongoing:
Qualcomm Snapdragon Tech Summit Live Blog: Compute Spotlight
www.anandtech.com
This is cinebench btw:
And it's also 2x faster than M2 in cinebench)if i heard her correctly) meaning at least there it's faster than M2 Pro.
Btw the native DaVinci Resolve demo on 64bit WARM was impressive.
Some task there were 1.7x faster than on 12 core Intel chips, with npu it was 3x faster
100% agree. GB6 ST is much more valuable and more accurate.Nonsense. GB5 is useless benchmark and vendors loved cheating in it by padding scores with irrelevant stuff like vector AES instructions cause surely every user is going to stream AES at the speed of multiple gigabytes stright to /dev/null. Some other parts were not great either.
The only virtue it had was good scaling with core count, HT and was loved by "my threadripper is larger than your xeon" crowd even if said threadripper had no memory controller in half of CCXs and mediocre performance everywhere except in cb23 style tasks.
I am valuing GB6 score much higher for desktop and laptop use than GB5 cause it is much harder to cheat. But surely i can understand why all that "we have heterogenous chip with bunch of useless arm cores with triple number names" crowd hates it - cause GB6 exposes just how useless and incompetent they are vs proper chips. Theyd scale soooo good in gb5.
I think people will be surprised with how close Zen 5 is in perf/watt to ARM SOCs. Especially - standard Strix Point APU.Intel/AMD now have to work twice as hard to thwart the threat of WinARM.
The next two years could change everything.
Has skylake sp taught you nothing?Intel/AMD now have to work twice as hard to thwart the threat of WinARM.
The next two years could change everything.