Discussion Samsung Exynos SoC thread

Page 8 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

adroc_thurston

Diamond Member
Jul 2, 2023
7,574
10,345
106
Let's wait. I don't be surprised if it appears on the Exynos 1680 instead. Yeah, going to the mid range instead of top one.
Again, no amount of effort would make Samsung IP competitive with RDNA in any dimension.
 

jdubs03

Golden Member
Oct 1, 2013
1,312
914
136
What’s with all these really high posts on Twitter? Show single core 4127 and multi core almost 13482., and another at 3960/12121
From what I can tell they’re all fake. It’s not listed in the search results.

Correct me if I’m wrong but the highest is this:
Single core: 3455
Multi core: 11621

Still great scores, though, right near the 8EG5.

For reference:
1762240066096.png
 
  • Like
Reactions: marees

Nothingness

Diamond Member
Jul 3, 2013
3,328
2,405
136
Still sounds fake, no info about clock rates... and even I think that the current results were made in a refrigerator.
Expecting the sad truth once the unit gets released at public:
Single core: 27XX ish
Multi core: 85XX ish
Do you expect Samsung to underperform that much other SoCs based on the same CPU, such as Vivo X300 Pro running Dimensity 9500?
 
  • Like
Reactions: hemedans

DZero

Golden Member
Jun 20, 2024
1,747
669
96
Do you expect Samsung to underperform that much other SoCs based on the same CPU, such as Vivo X300 Pro running Dimensity 9500?
Knowing Samsung? Sadly yes. Unless they releases an e version which ironically has better thermals than the normal one.

Even the Exynos 2400e is interesting enough more competent than the 2400 normal. But with vapor chambers could change the situation.
 
  • Like
Reactions: techjunkie123

hemedans

Senior member
Jan 31, 2015
292
164
116
Still sounds fake, no info about clock rates... and even I think that the current results were made in a refrigerator.
Expecting the sad truth once the unit gets released at public:
Single core: 27XX ish
Multi core: 85XX ish
Not sure about single core but multicore should be much better than that, it's 10 core configuration without small cores, 11-12K score is doable, and Geekbench measure burst speed, even current soc use 20W doesn't mean your phone can sustain that for few seconds.
 

CouncilorIrissa

Senior member
Jul 28, 2023
754
2,744
106
What’s with all these really high posts on Twitter? Show single core 4127 and multi core almost 13482., and another at 3960/12121
From what I can tell they’re all fake. It’s not listed in the search results.

Correct me if I’m wrong but the highest is this:
Single core: 3455
Multi core: 11621

Still great scores, though, right near the 8EG5.

For reference:
View attachment 133152
It's just a fake. Unless we're talking about an Apple device (the scores of those are typically hidden in the database before announcement), if the result does not show up in the browser, then chances are it does not exist.
You can check for yourself by searching by a motherboard name (s5e9965 in this case).
 
  • Like
Reactions: marees

jdubs03

Golden Member
Oct 1, 2013
1,312
914
136
It's just a fake. Unless we're talking about an Apple device (the scores of those are typically hidden in the database before announcement), if the result does not show up in the browser, then chances are it does not exist.
You can check for yourself by searching by a motherboard name (s5e9965 in this case).
Yeah that’s where I got that result I posted.

What annoys me is how fast the bullshit spreads.
And it doesn’t help when outlets like phonearena and others catch on and distribute it on a wider basis with no checking.

There is too much wish casting if it can even be called that. I don’t believe for one second what the guy said about internal testing either. I highly doubt even with liquid nitrogen they are attaining that single core reality.

It’s a shame too because that score I posted there, single core of 3455, is reported at 3.8 GHz. That score/frequency would be quite competitive with the A19 Pro. Taking best score vs. best score, it’s only 3% behind.

That should be the story, not some made up crap.
 
Last edited:

jdubs03

Golden Member
Oct 1, 2013
1,312
914
136
Just caught this. Figured it was relevant and it refers to the score I posted. Don’t shoot the messenger.
Test 5: Ready for Certification and Galaxy Software:
  • 3.92 GHz / 2.82 GHz
  • 3,959 Single Core / 12,010 Multi Core
  • 98% Sustained Performance
  • 78°C Temperature
——————————————————

Still taking with a mountain of salt. And I mean Mount Everest.

But even if this is 85% true, Samsung can extricate themselves from Qualcomm’s CPU grasp. Maybe their iPad Pro alternative could actually be competitive.
 

marees

Golden Member
Apr 28, 2024
1,962
2,588
96
Just caught this. Figured it was relevant and it refers to the score I posted. Don’t shoot the messenger.
Test 5: Ready for Certification and Galaxy Software:
  • 3.92 GHz / 2.82 GHz
  • 3,959 Single Core / 12,010 Multi Core
  • 98% Sustained Performance
  • 78°C Temperature
——————————————————

Still taking with a mountain of salt. And I mean Mount Everest.

But even if this is 85% true, Samsung can extricate themselves from Qualcomm’s CPU grasp. Maybe their iPad Pro alternative could actually be competitive.
We don't know 2 things
  1. Yield
  2. Cooling needs
The above 2 will determine if Samsung goes with Exynos or sticks to Qualcomm
 
  • Like
Reactions: jdubs03

Doug S

Diamond Member
Feb 8, 2020
3,693
6,528
136
Still taking with a mountain of salt. And I mean Mount Everest.

But even if this is 85% true, Samsung can extricate themselves from Qualcomm’s CPU grasp. Maybe their iPad Pro alternative could actually be competitive.

Samsung's biggest need is to stop buying Qualcomm SoCs and use their own. With those prices going up and up the difference between "fab your own" and "buy at Qualcomm's huge markup" gets larger and larger. If we assume Exynos 2600 is 120mm^2 (I have no clue just picking it because its a round number) that's around 500 die candidates per wafer. Let's call the "price" of each wafer (in quotes since Samsung is buying from itself) $25,000, then each candidate chip is $50. If they have a pretty bad yield of only 50% that's $100 per working chip which is a lot better than the ~$250 Qualcomm is charging for their SoCs. A savings of $150 per phone on that one part is massive!

So it sounds great but there are two obstacles to Samsung declaring freedom from Qualcomm like Apple will next September. One, as you say, that mountain of salt regarding all these unsupported performance rumors. They don't have to beat Qualcomm SoCs but they have to at least be in the same ballpark, so I agree with you if the rumors are 85% true that's good enough.

Two, the Exynos modem. Even when Samsung's SoCs were quite competitive with Qualcomm's they always limited the regions where phones using Exynos SoCs were sold. Why? I think it is reasonable to guess the modem is why, and all the complaints from Pixel owners about Google using Exynos modems serve as the example. Samsung probably needs to address whatever shortcomings are responsible for that. Apple managed to make a modem that's quite competitive (better in some ways, actually) than Qualcomm's, and Samsung has been doing their own modems for far longer. Samsung has no excuse not to fix that, especially if it becomes the only thing left standing between them and cutting Qualcomm loose.
 

marees

Golden Member
Apr 28, 2024
1,962
2,588
96

Galaxy S26 Ultra sticks with Snapdragon, Exynos comeback delayed to S27 Ultra​


An industry source who requested anonymity said,

“Samsung Electronics will only include the Exynos 2600 in the base and Plus models, and will equip the Ultra model with Qualcomm’s mobile Application Processor (AP).”

Exynos in Galaxy S Ultra is a difficult decision as it’s the highest-selling model of the lineup. Samsung can’t risk it to save its in-house mobile chipset as it will be a long-term harm to the segment’s reputation.

Samsung has made great progress in its 2nm GAA node, which manufactures the Exynos 2600. As it’s the first technological leap, it’s difficult for Samsung to make it to the Galaxy S26 Ultra; hence, the Ultra will use Snapdragon only.

 
  • Like
Reactions: jdubs03

DZero

Golden Member
Jun 20, 2024
1,747
669
96

Galaxy S26 Ultra sticks with Snapdragon, Exynos comeback delayed to S27 Ultra​


An industry source who requested anonymity said,

“Samsung Electronics will only include the Exynos 2600 in the base and Plus models, and will equip the Ultra model with Qualcomm’s mobile Application Processor (AP).”

Exynos in Galaxy S Ultra is a difficult decision as it’s the highest-selling model of the lineup. Samsung can’t risk it to save its in-house mobile chipset as it will be a long-term harm to the segment’s reputation.

Samsung has made great progress in its 2nm GAA node, which manufactures the Exynos 2600. As it’s the first technological leap, it’s difficult for Samsung to make it to the Galaxy S26 Ultra; hence, the Ultra will use Snapdragon only.

So... Ultra will be noticeably more expensive compared to the others? Expected.
 

Doug S

Diamond Member
Feb 8, 2020
3,693
6,528
136
“Samsung Electronics will only include the Exynos 2600 in the base and Plus models, and will equip the Ultra model with Qualcomm’s mobile Application Processor (AP).”

Well that makes it pretty clear those benchmarks showing GB6 of 4000+ were all BS. If it that was good they'd be using it in their top product not the other way around lol
 
  • Like
Reactions: Geddagod

Geddagod

Golden Member
Dec 28, 2021
1,582
1,648
106
Samsung's biggest need is to stop buying Qualcomm SoCs and use their own. With those prices going up and up the difference between "fab your own" and "buy at Qualcomm's huge markup" gets larger and larger.
Do you think Exynos at TSMC would still be worth it vs continuing to use Qcomm? Cuz realistically that's the only way I would ever see an Exynos being competitive with whatever Qcomm has.
Anyone know how good the physical design/implementations of ARM cores are from Samsung LSI vs other companies such as Mediatek (afaik only ever used TSMC) or even Qcomm?
So far all I know is that Xiaomi seems very good, Mediatek questionable, and heard Google is just bad.
 

jdubs03

Golden Member
Oct 1, 2013
1,312
914
136
Well that makes it pretty clear those benchmarks showing GB6 of 4000+ were all BS. If it that was good they'd be using it in their top product not the other way around lol
The excuse is that Samsung’s cellular modem feature set is what is holding them back. CPU wise though it looks quite competitive performance wise (average GB ST of 3500-3600) with better efficiency.

Time will tell; will be good to see if Geekerwan can get his hands on it.

Looks like 75% of units will be Qualcomm:
 
  • Like
Reactions: marees

Doug S

Diamond Member
Feb 8, 2020
3,693
6,528
136
Do you think Exynos at TSMC would still be worth it vs continuing to use Qcomm? Cuz realistically that's the only way I would ever see an Exynos being competitive with whatever Qcomm has.
Anyone know how good the physical design/implementations of ARM cores are from Samsung LSI vs other companies such as Mediatek (afaik only ever used TSMC) or even Qcomm?
So far all I know is that Xiaomi seems very good, Mediatek questionable, and heard Google is just bad.

You're assuming that Samsung foundry can never catch up to TSMC. Maybe they can't, but it doesn't look like Samsung is ready to throw in the towel on that anytime soon.

I think Exynos is CLEARLY worth it, just based on the ridiculous prices Qualcomm is now rumored to be charging for Elite 5. Just imagine what it will cost when they go to N2! But like I said it has to be in the ballpark of Qualcomm performance wise - a little slower is OK since I don't think performance matters to the average person nearly as much as posters on boards obsessing over benchmark results would like to believe.

I still think their modem is likely a bigger issue, and will continue to believe that until they prove it isn't by shipping a single model of Galaxy that uses Exynos worldwide instead of regionally. They could/should have fixed that years ago, while they were messing around with broken processes that forced them to pay much higher prices for Qualcomm's SoCs.