Discussion AMD SoC Halo series GPU discussion

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

gdansk

Diamond Member
Feb 8, 2011
4,074
6,749
136
CPU R23 nT scores are also solid. Comparable at the top end to 16C dragon range from last gen. 60 W or so seems to be the sweet spot. Efficiency should be comparable to M4 Pro/Max/Strix point for R23 nT specifically.
The Cinebench scores at 55W are worse than a 7940HX I have. IOd improvements aren't showing up in any CPU benches? Not even in efficiency. Do I start weeping for Zen 6 again?
 

techjunkie123

Member
May 1, 2024
145
313
96
The Cinebench scores at 55W are worse than a 7940HX I have. IOd improvements aren't showing up in any CPU benches? Not even in efficiency. Do I start weeping for Zen 6 again?
STXH looks slightly better than notebook checks data: https://www.notebookcheck.net/AMD-R...cient-than-Intel-Raptor-Lake-HX.705034.0.html.

But dragon range's idle power consumption (IOD overhead + others) is only about 8 W.

I mean Zen5 isn't much more efficient on desktop either. Basically same node so 😐. Not sure what you were expecting.
 
  • Like
Reactions: Tlh97

Meteor Late

Senior member
Dec 15, 2023
289
314
96
Yep so around 4060 performance at higher power levels, the laptop they are comparing against has a RX 7600S, and from what I've seen, RTX 4060 Mobile is around 10-12% higher than 7600S, difficult to find reliable data however.
 

Meteor Late

Senior member
Dec 15, 2023
289
314
96
STXH looks slightly better than notebook checks data: https://www.notebookcheck.net/AMD-R...cient-than-Intel-Raptor-Lake-HX.705034.0.html.

But dragon range's idle power consumption (IOD overhead + others) is only about 8 W.

I mean Zen5 isn't much more efficient on desktop either. Basically same node so 😐. Not sure what you were expecting.

Zen 5 loses it's efficiency advantage at lower power levels compared to Zen 4, just go check Geekerwan's data for single core efficiency of both at lower power levels. Worse in int, only slightly better in FP at around 7W. When there are 16 cores, it's not going to look pretty (against Zen 4).
 

Meteor Late

Senior member
Dec 15, 2023
289
314
96
Zen 5 vs Zen 4 efficiency.

That's what you get for AMD skipping 3nm for Zen 5. Of course, you will still pay premium (Strix Point) or Halo (Strix Halo), though.
 

Attachments

  • Z5-4int.jpg
    Z5-4int.jpg
    383.7 KB · Views: 23
  • Z5-4fp.jpg
    Z5-4fp.jpg
    381 KB · Views: 23

gdansk

Diamond Member
Feb 8, 2011
4,074
6,749
136
I mean Zen5 isn't much more efficient on desktop either. Basically same node so
I expected a preview of what Zen 6 can bring beside IPC and I think I got it. IOd may not be holding back Zen 5 as much as popular belief suggests.

There's nothing to show in efficiency for a newer node IOd connected with lower pj/b interfaces. That's mildly unexpected.
 

techjunkie123

Member
May 1, 2024
145
313
96
I expected a preview of what Zen 6 can bring beside IPC and I think I got it. IOd may not be holding back Zen 5 as much as popular belief suggests.

There's nothing to show in efficiency for a newer node IOd connected with lower pj/b interfaces. That's mildly unexpected.
Fair enough. Let's wait for reviews to see if there's a benefit. CB nT is not bandwidth limited so wouldn't benefit from the new IOd necessarily.
 

CouncilorIrissa

Senior member
Jul 28, 2023
623
2,418
96
Well, it's the only decent data I've found so...
1736373211989.png

Zen 5 performs kind of weird in SPECint compiled with Clang. I guess part of the reason is that Geekerwan uses Clang 14 from 2022. Whereas the `znver5` enablement was only merged a few months ago to Clang 20 branch and backported to 19.

SIR compiled with GCC paints a different picture.

And it's not the first time Geekerwan's meme SPEC curves raised questions.

The Cinebench scores at 55W are worse than a 7940HX I have. IOd improvements aren't showing up in any CPU benches? Not even in efficiency. Do I start weeping for Zen 6 again?
Additional bandwidth is probably not going to show up in 1t benches (unless it's some ridiculous SIMD-heavy code) when the core's backend is latency-bound.
 
Last edited:

DavidC1

Golden Member
Dec 29, 2023
1,449
2,357
96
I expected a preview of what Zen 6 can bring beside IPC and I think I got it. IOd may not be holding back Zen 5 as much as popular belief suggests.
That is precisely the reason why IO die can be put on an older node.
Zen 5 vs Zen 4 efficiency.

That's what you get for AMD skipping 3nm for Zen 5. Of course, you will still pay premium (Strix Point) or Halo (Strix Halo), though.
AMD just changed the V/F curve so it benefits the higher power more on the Strix Point generation over predecessors. Newer node wouldn't have really changed this(except shift the whole curve).

This is likely to compete with Intel's Raptorlake doing better on the higher power(but flunking on the lower power range).
 

marees

Golden Member
Apr 28, 2024
1,012
1,356
96
Almost good enough for handhelds

 

ToTTenTranz

Senior member
Feb 4, 2021
387
757
136
Most high end laptops have a GPU and cooling solution to justify it. Is there really a market for 100W+ handhelds? It is surprising to me as I haven't heard of such a thing.
If implemented in a handheld, I don't think Strix Halo would be set to 100W.
We'd be looking at single CCD 8-cores or perhaps 2*CCD 6 cores each very modestly clocked, plus 40 or 32CU also clocked lower than the full laptops.
With a TDP set at 35-40W, it would still be a gaming powerhouse.


99watts is maximum. Due to some kind of regulation

Asus & MSI already have 80watts

That's battery capacity, not power consumption, and only on flights. You must bring batteries with less than 100W.h in a flight. Which is why you'll probably never see any battery over 99.9W.h in any portable device.
 

eek2121

Diamond Member
Aug 2, 2005
3,321
4,880
136
At least 1 mini PC has been announced so far. Hoping to pick one up in the spring provided AMD has gaming and media optimized drivers available.

It makes for a perfect Steam Machine and media player.

Shoot, if the GPU were faster it could replace my desktop.

I hope it sells well so AMD is encouraged to do more of this type of thing in the future.
 

Meteor Late

Senior member
Dec 15, 2023
289
314
96
I've seen some prices.
-Asus ROG Flow Laptop, 32GB and a fairly standard IPS display without Mini Led. With Ai Max 390 it's 2000$, with AI Max 395 it's 2200$
-HP Z2 Mini G1a Mini PC, 1200$ with the cheapest AI Max 380, with 6 CPU cores and 16 CU, unknown RAM capacity for said price.

Yeah, Strix Halo is DOA. If at least it was RDNA4 with the much improved upscaler...

I think Medusa Halo will be the point where this product concept is more interesting.
 
Last edited:

ToTTenTranz

Senior member
Feb 4, 2021
387
757
136
-Asus ROG Flow Laptop, 32GB and a fairly standard IPS display without Mini Led. With Ai Max 390 it's 2000$, with AI Max 395 it's 2200$
-HP Z2 Mini G1a, 1200$ with the cheapest AI Max 380, with 6 CPU cores and 16 CU, unknown RAM capacity for said price.

Yeah, Strix Halo is DOA. If at least it was RDNA4 with the much improved upscaler...

Why is Halo DOA?

The ROG Flow 13 is actually a Surface-like tablet. The IQ + performance in it when I briefly played Ratchet & Clank at AMD's booth is nothing short of impressive for a device of that size. It really is PS5 performance in a tablet. I don't know if that demo of Ratchet & Clank was already running FSR4, but if it wasn't then all the more confident I am in the product.

Where did you see that price for the HP mini pc? It can go all the way up to the full 16x Zen5 + 40 CU model.
 

Bryo4321

Member
Dec 5, 2024
53
109
66
It makes for a perfect Steam Machine and media player.
The low end 6 core 16 cu halo is weird. I wonder if valve could cut some deal with AMD to get a “custom” halo with a 6 core cpu chiplet and 30 cu gpu chiplets and juice the heck out of the tdp for a relatively cheap steam machine or something.

Is it possible to cut the gpu portion down to 15 wgp? I’m not really sure if such a configuration is possible.

But basically buying up all the chips that didn’t quite make the cut for the more premium devices to keep cost down for valve.