- Mar 3, 2017
- 1,755
- 6,635
- 136
View attachment 117343
7ish hours until this, I hope slides get released online soonish.
Very curious about what claims they have about the CCX power vs Zen 4, V/F curve, and CaC numbers they have about Zen 5. I don't think AMD has yet to release a gen since zen where even at low power their next gen core hasn't had a perf/watt increase vs their past gen, but this doesn't seem to be the case with Zen 5.
slides usually get uploaded to IEEE xplore, just don't remember when (after the entire conference?, a bit later?).
Now, is Strix Halo's CCD different from Turin's CCD too?According to this Tweet, the CCD in Strix Halo is indeed 66 mm2.
And according to measurements, Granite Ridge CCD is 70 mm2.
[...]
View attachment 117339
So, perhaps, AMD really made a bespoke CCD for Strix Halo.
Looking at the die, and the short side shrunk implies GMI is gone
Are you going by artistically rendered "die shots" (into which I for one wouldn't read too much into), or by additional information?Not 'perhaps'. It is one.
TSV columns are gone too!
The only CPU Worth standing a Chance is ARL-HXCan you see the split wattages for the CPU and GPU? Reviews looked good because they were comparing Halo products to older Intel and Ryzen parts, but I think ARL-H might stand a chance. Considering it's using the full-fat Zen 5 cores, I found it a bit underwhelming. Single-core efficiency isn’t any better than Strix Point, and multi-core performance is only on par with ARL-H. On the other hand, the iGPU is a real home run - it’s consistently matching or even beating laptop 4060/70 GPUs while consuming similar wattages for the entire chip compared to a dGPU. That’s just f___ing insane. It’s a shame it’s only using RDNA 3.5 - imagine if it had RDNA 4 or UDNA with massively improved ray tracing and FSR 4. At that point, it would make mid-range NVIDIA GPUs completely obsolete.
The only CPU Worth standing a Chance is ARL-HX
different d2d interface.Now, is Strix Halo's CCD different from Turin's CCD too?
i know what it is lmao.Are you going by artistically rendered "die shots" (into which I for one wouldn't read too much into), or by additional information?
Any LLM inference benchmarks yet?
>... tried the 64G model with a real LLM test - DS Llama 3.1 70b yielding 4.4 tps beating dedicated a single 4090...
Procyon Inference: https://forums.anandtech.com/threads/amd-soc-halo-series-gpu-discussion.2624449/post-41397732Any LLM inference benchmarks yet?
what are the chances the burn is falsified? is it easy to do?
Damn non-representative multi-thread results. Son of a b. Makes sense though it should score right around the vanilla 9900 series.
Are you going by artistically rendered "die shots" (into which I for one wouldn't read too much into), or by additional information?