Qualcomm 8 gen 4.

[Jimmyjames]

It’s the time of year where leaks purport to show the performance of upcoming devices. Today’s leak is for the 8 gen 4 soc. The leak shows a single core score over 3200 and multi-core over 10,000. A 40% single core jump seems unlikely to me, but we’ll see.

OnePlus PJZ110 - Geekbench

Benchmark results for an OnePlus PJZ110 with a Qualcomm ARMv8 processor.

browser.geekbench.com

 

[Cmaier]

possible, i guess.

 

[Yoused]

SC is always the biggest core, which, on an 8, is usually one X-series core. In ths case, it might be one of their in-house Elite cores rather than a stock ARM X core. In either case, big cores and the 500 series E cores are v9 only (no 32-bit) and only the older mid-power 700 series cores support 32-bit (yes, there is more to v9 than just being 64-bit only, but that is a good indicator).

 

[Jimmyjames]

I’m curious how they managed to get more performance out of their cores on a phone, than they did in their laptops. It does make me suspicious of this result.

 

[Cmaier]

I’m curious how they managed to get more performance out of their cores on a phone, than they did in their laptops. It does make me suspicious of this result.

Through the magic of hiring the best engineers from Apple (because whenever any engineer leaves apple, it’s always the best engineer, which is why Apple is doomed).

 

[Jimmyjames]

Through the magic of hiring the best engineers from Apple (because whenever any engineer leaves apple, it’s always the best engineer, which is why Apple is doomed).

Understood. I will immediately throw my Apple products away.

 

[leman]

Leaked documents suggest that 8 Gen 4 is a 3nm chip with Oryon cores, so I don’t find that performance surprising. The power draw must be quite high however, at those clocks.

 

[Jimmyjames]

Leaked documents suggest that 8 Gen 4 is a 3nm chip with Oryon cores, so I don’t find that performance surprising. The power draw must be quite high however, at those clocks.

So what was preventing them reaching that performance with their X Elite chips?

 

[Yoused]

So what was preventing them reaching that performance with their X Elite chips?

X Elite was on N4.

 

[Jimmyjames]

X Elite was on N4.

Ahh interesting. Thanks.

 

[dada_dave]

Leaked documents suggest that 8 Gen 4 is a 3nm chip with Oryon cores, so I don’t find that performance surprising. The power draw must be quite high however, at those clocks.

So what was preventing them reaching that performance with their X Elite chips?

X Elite was on N4.

I believe leaks also suggested that they’ll be the second generation Oryon cores, as in a new microarchitecture rather than the same one ported to a new node. But I don’t think that’s known for sure. As @leman said, power draw must be high at those clocks!

 

[Cmaier]

I believe leaks also suggested that they’ll be the second generation Oryon cores, as in a new microarchitecture rather than the same one ported to a new node. But I don’t think that’s known for sure. As @leman said, power draw must be high at those clocks!

given how long it took them to get the first gen done, I’d be surprised if they are already shipping the second microarchitecture (unless it’s a very minor difference from the first)

 

[dada_dave]

given how long it took them to get the first gen done, I’d be surprised if they are already shipping the second microarchitecture (unless it’s a very minor difference from the first)

Aye, unless as has been hinted there was a substantial delay in getting the first out the door - especially relative to when the silicon design was ready. We don’t know for certain that there was mind you but the leaked Dell documents indicated they thought their Apple comparisons would still be the M1, maybe the M2. TSMC N4 started volume production at the end of 2022 and we know the software side of Windows on Arm still has issues so it’s possible the silicon has been sitting in the can for awhile while they tried to get the rest of it usable. And I remember @Artemis said the ARM lawsuit delayed the initial silicon design as well.

Obviously this is all just slightly informed speculation on my part. But I think it’s possible they’ve got a brand new core. These products aren’t set to launch for awhile anyway - Gen 3 phones I think didn’t launch until January of this year after being announced in October 2023*. So if these benchmark leaks are accurate, then either they’re very early leaks or Qualcomm has shifted its timetable.

Also entirely possible it’s just Oryon V1 ported to N3.

*Edit: Ah, apparently the Xiaomi 14 launched in China shortly after the snapdragon announcement, but global launch wasn’t until 2024 and most products didn’t launch globally until 2024 as far as I can tell. So not quite right.

 

[Artemis]

Yes the lawsuit and other delays impacted things, I don’t think this is indicative of the team or Qualcomm’s productivity more broadly, but could be wrong, though it was a reasonably extenuating circumstance. Fitting the cores back over to QC fabric etc.

It should *NOT* be a significantly new micro architecture, there might be design and power delivery/fabric tweaks but the IPC lines up with X Elite stuff mostly.

What is new is A) almost certainly the physical design with N3E and B) more SLC is for sure going to be seen here because it’s a phone chip

C) E Cores (custom)

Also of note: with N4/N4P and the first laptop chip, QC really has 3.4GHz as their baseline most common frequency.

With this part it’s looking like 4GHz, and 4.3GHz is the “Plus” model or whatever that they usually do with some extra frequency. I suspect power on that isn’t fantastic, but this is still a phone profile so even with gaming phones or boosting, that 4.34GHz & 3200 GB6 ST is doubtful going to be more than 8-10W. Point being, huge average ST speed and power improvement with a similar micro architecture, probably due to the node allowing higher peak speeds and some other cache or Qualcomm power delivery tweaks.

And the standard baseline chip should be at 4GHz & 2750-2900 GB6 ST. It’s rumored at about 4.5-6W which is normal and fine and makes sense.

Overall the fact that the baseline N3 Phone chip is at 4GHz for all parts on roughly the same micro architecture and doing so in (again, remains to be seen for sure but probably?) phone profile is a good sign for the X Elite 2 on N3 in 2025/2026. Also likely shows they tweaked a few things, because node alone wouldn’t take the Oryon core down to 5-6W. I could be wrong and the baseline 8 Gen 4 is like 8W for 3.7GHz and throttling terribly lol, but I doubt it. The Plus model for gamers and shit sure.

Anyway, I think 4+GHz for the XE2 will be standard along with a sizable IPC increase, power reigned in a bit and new E cores will probably buoy the battery life of perf/$ (unlike Intel, Qualcomm hasn’t even played that card yet lol).

Arguably the most interesting thing about the 8 Gen 4 is what it will say about how much room to run Qualcomm has for laptops, much like Lunar Lake imho showed this by basically doing less with more, save the GPU.

 

[Artemis]

So what was preventing them reaching that performance with their X Elite chips?

N4P like he said. Now in theory N3 buys you like + 10% more speed at the same power than N4P, which is one thing, but I suspect it bought a bit more at the fMax top end of what’s possible with good yield, and then of course you increase power and it’s not “free” but that’s all else equal. They may well have taken the max good yield freq they could get for a given physical design, and tweaked other things that were messed up in the original Oryon (power delivery and fabric mess was mentioned) or added more SLC which is also huge.

But yeah it decidedly shows more than a mere + 10% was left on the table from N3, because 4GHz should be mass market standard here, vs 3.4 with the N4 laptop Oryon stuff.

 

[dada_dave]

N4P like he said. Now in theory N3 buys you like + 10% more speed at the same power than N4P, which is one thing, but I suspect it bought a bit more at the fMax top end of what’s possible with good yield, and then of course you increase power and it’s not “free” but that’s all else equal. They may well have taken the max good yield freq they could get for a given physical design, and tweaked other things that were messed up in the original Oryon (power delivery and fabric mess was mentioned) or added more SLC which is also huge.

But yeah it decidedly shows more than a mere + 10% was left on the table from N3, because 4GHz should be mass market standard here, vs 3.4 with the N4 laptop Oryon stuff.

I think the Elite was even N4 not N4P, no? ... which accentuates your point that they should be getting a small performance/efficiency boost from the switch.

 

[Yoused]

4GHz should be mass market standard here, vs 3.4 with the N4 laptop Oryon stuff

4GHz is like full-on SC, though. It should drop as more cores start playing.

 

[leman]

Something very interesting about these new N3 Qualcomm chips - assuming that the leaked scores are accurate, is that they allow us to exactly quantify the micro-architectural difference between Qualcomm and Apple. It’s the same process and almost identical performance (A18 only gets ahead in SME-enabled tests), but the A18 is clicked 5-7% lower. That’s the IPC difference.

 

[dada_dave]

Something very interesting about these new N3 Qualcomm chips - assuming that the leaked scores are accurate, is that they allow us to exactly quantify the micro-architectural difference between Qualcomm and Apple. It’s the same process and almost identical performance (A18 only gets ahead in SME-enabled tests), but the A18 is clicked 5-7% lower. That’s the IPC difference.

For what it’s worth these cores are also a little better in iso-clock performance than the Snapdragon Elite though of course GB’s variance comes into play so hard to say for sure. Around 6% better than the Elite 84 by my estimate so it could be a new core or at the very least as Avalanche was to Firestorm.

Edit: I used this as a point of comparison:

SAMSUNG ELECTRONICS CO., LTD. Galaxy Book4 Edge - Geekbench

Benchmark results for a SAMSUNG ELECTRONICS CO., LTD. Galaxy Book4 Edge with a Snapdragon X Elite - X1E84100 - Qualcomm Oryon processor.

browser.geekbench.com

 

[Artemis]

4GHz is like full-on SC, though. It should drop as more cores start playing.

That’s true but peripheral to my point. That’s the case in virtually any system with lateral dissipation of heat.

My point is just that they can yield dual-core for 4GHz easily on N3E, such that the peak ST performance (and similarly ST even when taken down a notch with other cores playing) is higher on a phone chip with N3E and their second chip (even if the microarch is the same roughly) — and a lot higher — we’re looking at 4-4.3GHz baseline vs 3.4GHz. Much of that about N3E, some of that probably physical design independent of it and/or other details.

That said I want to see what power looks like. Like at 4-6.5W, what’s going on? It will be super disappointing if the 2400 GB6 range is like a 5.7W performance lol


the 4.3 thing makes no sense though, even with tweaks and N3E that’s going to be crazy.

if the 2850/4GHz score is around 5W +- 1 Watt measured from Mobo, I consider that pretty good. If it’s not, that kinda sucks