Snapdragon 8 Elite Review: How Efficient is The New Flagship?

Posted by fatso486@reddit | hardware | View on Reddit | 16 comments

[-]

Wh1teSnak@reddit

Best single thread efficiency: A18pro (Big margin)

Best multi thread efficiency: 8 Elite (Good margin)

Best GPU efficiency: 9400 (Small margin)

Honestly all great SoC's. This year we got like 2 generation's worth of improvements on the android side. Makes stuff more exciting compared to PC!

[-]

TwelveSilverSwords@reddit

This affirms what we saw in the dieshots; Oryon-M is a completely new E-core, designed from the ground up!

Even Oryon-L has some minor differences compared to Oryon-M. Also it's notable that the new Adreno GPU doesn't have Tile Based Rendering.

Very stellar SoC overall. No wonder that ARM is feeling threatened and wants to sabotage Qualcomm's custom CPU efforts.

[-]

based_and_upvoted@reddit

Maybe Qualcomm should've paid ARM what they owe for the license rights to use Orion in non server environments.

But what do I know, let the courts decide which billionaire company should pay which.

[-]

There is the court of law and the court of public opinion. It seems ARM fears that they'll lose in the court of law, so they are trying to win the court of public opinion to get Qualcomm to settle early.

[-]

based_and_upvoted@reddit

Why are you quoting a person from Qualcomm, who is obviously not going to be impartial? Unless I missed your point.

I don't know whether or not ARM can outright revoke Qualcomm's license like that, but it is almost certain ARM is in the right regarding the nuvia purchase license not being valid for non server products.

It's good Qualcomm is tasting a bit of its own poison, too.

[-]

Vince789@reddit

Why are you quoting a person from Qualcomm

We need to look at both sides, not just Arm's

Qualcomm seems extremely confident they're in the clear legally, hence have consistently said cya in court

Whereas Arm seems to be desperately trying to force an early settlement before the court case starts

but it is almost certain ARM is in the right regarding the nuvia purchase license not being valid for non server products

Why should Arm have any right over who acquires Nuvia and how Nuvia's IP is used??

Arm's essentially arguing they have ownership right to any custom architecture based on Arm ISA. That's an extremely dangerous precedent for the industry with major implications beyond Qualcomm

[-]

based_and_upvoted@reddit

Why should Arm have any right over who acquires Nuvia and how Nuvia's IP is used??

Because from the information that has been made public, Nuvia had a license to make server CPUs, Qualcomm purchased them and is using the IP designed for servers to be used in consumer. I bet that if Qualcomm had used the IP for servers only like Nuvia had in their contract with ARM, that nothing would've happened.

[-]

Vince789@reddit

Qualcomm claims they already had their own ALA which covers custom cores for phones to laptops to servers (which Arm has essentially confirmed with the cancellation notice)

Hence the question becomes does Arm have ownership of Nuvia's IP?

It could be possible, we don't know Nuvia's ALA terms, that's what Arm will have to prove in court

[-]

BabySnipes@reddit

ARMs crab mentality is gonna screw the consumer over.

[-]

McSnoo@reddit

Pay up the fee. WaNt to be Apple yet doesn't want to pay the price like apple. Crab mentality my aSS.

[-]

theQuandary@reddit

I think the L-core changes are more than minor.

Moving from 192kb to 128kb indicates an entire L1 cache redesign which is a really big change.

120 -> 157 is a massive 24% jump in integer scheduling. 192 -> 209 for float scheduler is smaller, but still an 8% increase. Both of these involve pretty big changes not just in the scheduler, but in the rest of the frontend to keep them fed.

Likewise, the move from 650 ->679 ROB isn't massive, but is notable because Apple's chip actually reduced the ROB size for M2 because they couldn't use it effectively. This seems to indicate that Qualcomm's team is actually using what they have and wanting more.

These explain the increase in Geekbench/GHz of about 7%. If that holds true across the system, that's a bigger IPC jump than M3 -> M4.

[-]

DerpSenpai@reddit

in 3 years QC was able to outcompete ARM completely in CPU design and the worst is yet to come

[-]

kyralfie@reddit

Very stellar SoC overall. No wonder that ARM is feeling threatened and wants to sabotage Qualcomm's custom CPU efforts.

Yeah, now we know why. Oryon sure lit some fire under ARM. I wonder how much x925's design was affected by early Nuvia presentations and claims.

[-]

Noble00_@reddit

Reference Design vs OnePlus 13 https://imgur.com/a/WKujqUN

Do Keep in mind OP13's RAM is \~11% slower.

What's noticeable is Spec2017 and GB16 don't show similar trends when compared to Apple's SOCs, which isn't too surprising. OP13 is slower than the reference model partly due to the slower RAM. This is especially noticeable in the Nomad benchmark where bandwidth is important.

Interestingly enough, Genshin has a lower internal resolution in OP13 than the Reference Design, -15%. Even with a higher resolution on the Reference Design, performance and power consumption is similar. That said, high performance is mode is interesting. For 2x the FPS (120), 23% higher res, it's a 26% higher power consumption.

The second game, Star Rail(?), internal res has a very slight difference, <2%. It is running with 5% less power than the reference design, but frame time isn't as consistent. That said, still very much more consistent than the iPhone 16 Pro Max. If we compare OP13 to iP16PM resolution its around a \~7% difference (reference was \~5%). High performance mode is slightly less exciting than the previous game, 2x FPS for 19% higher pwr consumption (no change in res),

The third game is the worst showing for Android in general. Negligible res difference between OP13 and ref and similar perf. 18% lower res than the iP16PM, but frame times are better where the iPhone trails to \~56FPS. Wattage is similar.