AMD Ryzen 7 5800U "Cezanne" Based on "Zen 3," Geekbenched

[Kapone33]

Yes, it is. That's translated x86 code on the M1 keeping up with native x86 code on Cezanne. That's impressive.

I was going to buy a Zen 3, but I'll probably buy an M-based Mac now. AMD clearly can't compete against a company like Apple.

If you think for a minute that the current iteration of AMD gives a damn about what Apple wants to do you are more mistaken than your posts seem to imply. BTW this is not a site for fanboyism.

 

[Steevo]

The apple phallus is deep inside that one.

I want availability, I had to buy 3xxx series CPUs and multiple laptops with older series due to availability.

Give us the chips.

 

[dyonoctis]

CZN's MT scores at 6.21 GB/s while ST scores have 5.58 GB/s is strange. AES should be able to scale with core count on CZN's side.

AES-XTS is pointless for Blender3D type workloads.

I went and made a comparison with the 4800u, (and even zen 2 vs zen3) the results are really weird. I don't know why, but zen2 is faster than zen3 in MT AES-XTS, Navigation and machine learning are also points were zen2 often takes the leads in MT, or is just as fast, even though they lost to zen 3 in ST. There is either something wrong with the scheduler, or geekbench is full of shit

 

[Chrispy_]

I went and made a comparison with the 4800u, (and even zen 2 vs zen3) the results are really weird. I don't know why, but zen2 is faster than zen3 in MT AES-XTS, Navigation and machine learning are also points were zen2 often takes the leads in MT, or is just as fast, even though they lost to zen 3 in ST. There is either something wrong with the scheduler, or geekbench is full of shit

You really can't compare zen2 laptops vs zen3 laptops unless they are configured to use the same cooling, the same cDTP, and the same RAM timings.

By altering the cDTP and cooling of my Zen1 laptop from 12-25W I can create a range of results that's three times faster at the high end than it is at the low end. Unless laptop test results are stipulated, in detail, that they are running on otherwise identical configs, the results are meaningless - and Geekbench is inconsistent run-to-run and version-to-version too, for added random noise.....

 

[dyonoctis]

You really can't compare zen2 laptops vs zen3 laptops unless they are configured to use the same cooling, the same cDTP, and the same RAM timings.

By altering the cDTP and cooling of my Zen1 laptop from 12-25W I can create a range of results that's three times faster at the high end than it is at the low end. Unless laptop test results are stipulated, in detail, that they are running on otherwise identical configs, the results are meaningless - and Geekbench is inconsistent run-to-run and version-to-version too, for added random noise.....

yes, but those results are in line with the desktop results. for some reason in geekbench zen 2 is better at specific task than zen 3 when multithreating is involved... (3600x vs 5600x) idk there's too many variable with that benchmark and the online results

 

[Chrispy_]

yes, but those results are in line with the desktop results. for some reason in geekbench zen 2 is better at specific task than zen 3 when multithreating is involved... (3600x vs 5600x) idk there's too many variable with that benchmark and the online results

The problem with Geekbench is that it's modified to constantly tune new architectures between versions. It's less of a benchmark and more of a test of how much effort the devs have put into tuning it to run on a specific CPU family.

 

[michwoz]

Biggest question yet - is it using RDNA or not?

What would be the point of putting RDNA on DDR4 platform, without on package memory?

 

[zlobby]

What would be the point of putting RDNA on DDR4 platform, without on package memory?

Eh, these babies support 4266MT/s in dual channel mode. With bank and row interleaving the total memory bandwidth is quite nice. IMO, the iGPU should benefit significantly.

@W1zzard and crew, can you guys please make analysis how iGPU performance scales with memory speed on modern iGPU, e.g. Xe, Vega?

 

[michwoz]

It will benefit, sure, but it will still suck. The benefit is not that great imo. iGPUs hit memory performance limit at Vega 11/ Vega 8 (7Nm) / Intel Xe 96EU level some time ego (way below 2 Tflop FP32). To reach levels significantly better than that, a SoC with integrated memory is a required (consoles, Kaby Lake-G, Apple M1). "Traditional" APUs won't cut it.