Hacker News Reader

Comment by ben-schaaf | Hacker News Reader

The 6850U had a peak of 35W (https://www.phoronix.com/review/amd-ryzen7-6850u/8). Around the same as the M2 Max apparently.

View on HN

Sadly at about about half the Geekbench 5 score. M2 is single core 2061, multicore 15281. 6850U has single 1469 and multi 7365 scores. M2 is also ahead in Cinebench R23 and splits the single and multi in R20 with the 6850U.

To go from that to 70% ahead on multicore would be amazing because it would force Intel and Apple to respond.

The cynic in me, having watched Intel and AMD fight for the desktop crown post Ryzen launch though is skeptical that the scores are some combo pushing power limits to rather undesirable levels and extremely selective benchmarking (It's 70% better at this one thing).

But AMD doesnt package everything in one big die.

Is that directly relevant to the consumer? The M1 architecture seems to be fairly sustainable. Intel, not so much.

It is, because I can buy a framework laptop w/ 64GB ram, 2 TB SSD and these latest Ryzen chips for half the cost of a macbook pro with similar addons.

That 64GB RAM isn’t also VRAM though. Not exactly a direct comparison.

It is in a Framework or any other low-power system that doesn't use a discrete GPU.

My understanding is that the only reason dGPUs generally don’t use system RAM isn’t that they don’t have access, but that the supporting software doesn’t use it out of choice because the of the speed hit (iGPUs, of course, do generally use system RAM as VRAM.)

> My understanding is that the only reason dGPUs generally don’t use system RAM isn’t that they don’t have access, but that the supporting software doesn’t use it out of choice because the of the speed hit

It looks like one of the advantages of Apples put-everything-on-the-package strategy is that they can have a very wide bus to ram, which makes using system ram for the GPU much more palatable.

Isn't the main difference that Apple is building something like 4-16 channel memory controllers in a mobile device while you normally don't get more than 2-4 channels even on a desktop? That's a lot of transistors and (potentially) power usage but if you can get on the latest node and have a market willing to pay for giant chips it lets you get impressive amounts of bandwidth.

I don't think you need the RAM to be on the same package for that, it just makes the timings easier.

dGPUs do have access to system RAM. It's goes over the PCIe bus and because of that is too slow for direct usage.

half? not likely

Yes, I am not exaggerating. The Framework came to around $1.8k and the Macbook Pro came to ~ $3.9k. The only compromise with the Framework is that it is a 13" screen and the Macbook pro is a 14". But that's because Apple only offers up to 24GB ram on the 13".

The "Apple tax" isn't really a thing for midrange specs any more, but they will absolutely charge you out the nose to max things out. The top-end macs are targeted at people who won't even notice if the laptop's price is doubled.

We notice… but grumble and pay it because we’ve justified it to ourselves. There are of course outliers with heaps of money that don’t care about the cost, but few businesses are so cavalier with cashflow as to not notice and check why someone is buying a $7k laptop or $25k desktop

They charge you out the nose even to do small upgrades. $200 for bumping the SSD from 256GB to 512GB and $200 to go to 16GB RAM. You can easily find brand name good PCIe Gen 4 2TB NVMe drives for less than $200 now.

I think we can’t make claims about “fairly sustainable” until it’s been around for more than two releases.

It is relevant to the benchmarks which are used in ads

Don’t forget they just announced Z1 for Steam Deck like machines (and Aerith for steam deck). They’re certainly not not in the game for low TDP. AMDs little GPUs are also much more capable at gaming because they get the decades of AMD experience and software in that space. M1 can maybe run a few games

There's no way AMD is going to make up that much ground in such a short amount of time. You will end up being correct in that AMD will be faster in one or two cherry-picked benchmarks. I will add these benchmarks will have zero relevance to anything regarding actual CPU power, and AMD will be greatly lacking in real performance.

> There's no way AMD is going to make up that much ground

What facts do you base this assertion on? Do you have insider access to the benchmark results? Maybe wait for the benchmark results before coming to Apple's rescue. Remember that it wasn't so long ago that AMD schooled Intel so hard that their CEO fell off.

Benchmarks of the 7840U are out. You can see for yourself it doesn't beat the M2.

I can't seem to find them, all I can't find are references to AMDs promise. Could you provide your source?

https://www.cpu-monkey.com/en/cpu-amd_ryzen_7_7840u

The AMD comes ahead in all benchmarks available on that site: https://www.cpu-monkey.com/en/compare_cpu-amd_ryzen_7_7840u-.... I'm not sure where it would fit in terms of perf-per-joule, but there seems to be a very limited amount of benchmarks available (e.g. there seems to be no benchmark for Blender for the AMD, which would be one of the more telling results). The jury is still out for this.

There's already Geekbench 5 scores of the 8 core 7840U up from devices like the Aokzoe A1 Pro with scores around 1900 in single core and 10K in multicore. The M2 scores mentioned above are for the 12 core M2 Max. The Aokzoe is a Steam Deck like handheld too so a larger chassis might allow for a little more out of the chip.

I confess to not really pay attention to the Apple ecosystem. Does anyone know enough about these benchmarks and chips to really know what is going on here? Is this straight up a difference in conventional superscalar/multicore throughput and efficiency?

Or are is there an aspect of Apple to non-Apple comparison here with the compiler, OS, and library ecosystem? I.e. do any of the benchmark calculations get executed off CPU in the GPU or other "matrix" coprocessing units in the Apple SoC? I am only interested in comparing CPU to CPU or GPU to GPU, but would want to exclude software-based differences like different compilers or support libraries.

The M1 has a lot of ALUs, an 8-wide instruction decoder, and a huge reorder buffer. It can decode and execute a lot of instructions on every clock and plenty of space to execute them in an advantageous order. Ignoring power draw, Intel and AMD beat the Mx then they hit higher clock speeds, the Mx top out at 3GHz, and/or pack in more P-cores.

I clearly need to invest in a new NUC. I have some anemic Celeron build that supposedly draws 10-20 W, but the performance is laughably bad. If these "real" chips have such low max draws, they should be able to fit within my informal <20W idle envelope, but still provide performance when needed.

Check out these systems based on AMD laptop chips - https://www.servethehome.com/beelink-ser6-pro-7735hs-edition...

Idle at 6-7W, but a full-powered multi-core CPU when you need it. Plus USB4 and 32 or 64GB RAM.

Reply on news.ycombinator.com