Question Geekbench 6 released and calibrated against Core i7-12700

I

igor_kavinski

Lifer

Jul 27, 2020

20,903

14,489

146

• Nov 1, 2024

• #751

Doug S said:

The competition is literally made by the same fab, if Apple was having yield issues so would AMD and Intel, unless you think Apple's designers are incompetent and are sending TSMC designs that inherently yield less well than their x86 brethren.

Click to expand...

So Qualcomm's binning issues with Oryon point to the Nuvia team being incompetent?

 

Upvote 0 Downvote

M

MS_AT

Senior member

Jul 15, 2024

365

798

96

• Nov 1, 2024

• #752

Doug S said:

Apple is not having any issues with yields. Other than the (major) hiccup that was N3B, TSMC has been firing on all cylinders. The competition is literally made by the same fab, if Apple was having yield issues so would AMD and Intel, unless you think Apple's designers are incompetent and are sending TSMC designs that inherently yield less well than their x86 brethren.

Click to expand...

Shouldn't they be more likely to run into yield issues thanks to the simple fact that M4 pro and Max are I guess physically bigger than Zen5 chiplets?

 

Reactions: Tlh97 and igor_kavinski

Upvote 0 Downvote

G

gdansk

Diamond Member

Feb 8, 2011

3,276

5,186

136

• Nov 1, 2024

• #753

MS_AT said:

Shouldn't they be more likely to run into yield issues thanks to the simple fact that M4 pro and Max are I guess physically bigger than Zen5 chiplets?

Click to expand...

GPUs often have some duplicated structures for yield purposes. And since most the die area is GPU who knows.

 

Upvote 0 Downvote

D

Doug S

Platinum Member

Feb 8, 2020

2,888

4,912

136

• Nov 1, 2024

• #754

igor_kavinski said:

So Qualcomm's binning issues with Oryon point to the Nuvia team being incompetent?

Click to expand...

Apple doesn't bin on speed or power, they bin to a limited extent on functional blocks - though at the scale they do so most of the "lesser" versions are likely to be fully functional.

That means Apple has to have a fairly conservative pass/fail frequency that basically all of the fully functional chips will quality at. If Apple binned like Qualcomm/Intel/AMD they'd be able to offer faster versions of their chips but one thing Apple hates is having a lot of SKUs. That was true even when they were buying chips from Intel and they had access to a much wider variety of frequency/power/core counts.

Another problem with binning on frequency is that if your demand models are off you have a shortage of higher bins, whether the "higher" bin is high frequency or low TDP. We've seen that sort of thing from Intel, even in their prime (i.e. not talking about paper launches like the 1.13 GHz P3) Were they incompetent then? No, they just guessed wrong about demand and there were more buyers than they had chips in those bin(s) coming out of the fab.

 

Upvote -1 Downvote

D

Doug S

Platinum Member

Feb 8, 2020

2,888

4,912

136

• Nov 1, 2024

• #755

MS_AT said:

Shouldn't they be more likely to run into yield issues thanks to the simple fact that M4 pro and Max are I guess physically bigger than Zen5 chiplets?

Click to expand...

You always have lower yield at larger sizes, though as gdansk mentions in regular structures like caches (other than L1) and GPU/NPU you can build in some redundancy. Whether or to what extent Apple or other companies do I have no idea, you'll have to ask someone smarter.

If for example you have 90% yield with a 100 mm^2 chip you'll have 81% yield at 200 mm^2 and 66% at 400 mm^2. Given that Apple is binning a lot more functional units in M4 Max (P cores, GPU cores, memory controllers, maybe SLC I'm not sure) they're able to compensate for that by recovering many of the defective 400 mm^2 (dunno the actual size but safe to say it is in the ballpark) M4 Max dies. If 66% are fully functional that's more than enough to satisfy demand, and maybe you recover another 10-15% of "bad" dies for the lower bin, and your yield is 80%.

So my handy internet calculator says you'd get 140 400 mm^2 dies on a wafer, if you get 100 that you can sell (that's 71% yield) and it costs $25K per wafer that's $250 per chip. Obviously there's test/packaging etc. but even with that pessimistic yield estimate that's a cost of at most $300 per M4 Max (assuming 400 mm^2 which is just a wild guess) which is more than competitive with what OEMs are paying for Intel/AMD's top bin laptop chips. Apple's costs are fine.

 

Upvote 0 Downvote

I

igor_kavinski

Lifer

Jul 27, 2020

20,903

14,489

146

• Nov 3, 2024

• #756

285K with DDR5-9600: https://browser.geekbench.com/v6/cpu/8629471

Not sure if LN2 is involved.

 

Upvote 0 Downvote

I

igor_kavinski

Lifer

Jul 27, 2020

20,903

14,489

146

• Nov 3, 2024

• #757

Colorful Technology And Development Co.,LTD CVN B860I FROZEN WIFI V20 vs Gigabyte Technology Co., Ltd. X670E AORUS MASTER - Geekbench

MAXSUN MS-iCraft Z890 Pacific vs ASUS System Product Name - Geekbench

GB5: 7 wins for 285K in ST, 14 wins in MT.
GB6: 8 285K wins in ST, 11 wins in MT.

No idea about the memory speed in GB5 but in GB6 both are running at DDR5-6400.

 

Upvote 0 Downvote

poke01

Platinum Member

Mar 8, 2022

2,581

3,409

106

• Nov 14, 2024

• #758

Gigabyte Technology Co., Ltd. B650M GAMING X AX - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. B650M GAMING X AX with an AMD Ryzen 7 7700X processor.

browser.geekbench.com

Windows 11 IoT LTSC is the GOAT Windows. I never achieved over 2900 until I used this version. Its almost on par with CachyOS

 

Reactions: Tlh97, lightmanek and igor_kavinski

Upvote 0 Downvote

I

igor_kavinski

Lifer

Jul 27, 2020

20,903

14,489

146

• Nov 14, 2024

• #759

poke01 said:

Windows 11 IoT LTSC is the GOAT Windows.

Click to expand...

Licensed and activated? Are the keys available to buy easily?

 

Upvote 0 Downvote

poke01

Platinum Member

Mar 8, 2022

2,581

3,409

106

• Nov 14, 2024

• #760

igor_kavinski said:

Licensed and activated? Are the keys available to buy easily?

Click to expand...

Yep, I used a workaround for the activation. The ISO is from Microsoft

 

Reactions: igor_kavinski

Upvote 0 Downvote

Det0x

Golden Member

Sep 11, 2014

1,346

4,545

136

• Nov 20, 2024

• #761

Tuned 9800X3D

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 7 9800X3D processor.

browser.geekbench.com

Not too shabby for a 8core X3D chip 🤣

Also did a Geekbench AI run

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 7 9800X3D processor.

browser.geekbench.com

 

Reactions: igor_kavinski, Tlh97, lightmanek and 1 other person

Upvote 0 Downvote

lopri

Elite Member

Jul 27, 2002

13,306

685

126

• Monday at 7:53 AM

• #762

@Det0x Wow so amazing. You should try a Linux run, it'll get you at least 3% more points. XD

 

Upvote 0 Downvote

lopri

Elite Member

Jul 27, 2002

13,306

685

126

• Monday at 7:57 AM

• #763

Why don't they update the "Processor Benchmark" chart? Still not showing the Ultras and Zen 5's.

 

Upvote 0 Downvote

Det0x

Golden Member

Sep 11, 2014

1,346

4,545

136

• Monday at 4:42 PM

• #764

lopri said:

@Det0x Wow so amazing. You should try a Linux run, it'll get you at least 3% more points. XD

Click to expand...

Cant submit linux runs to hwbot, and i'm on a mission

Only the LN2 guys thats untouchable 🤣

 

Reactions: igor_kavinski

Upvote 0 Downvote