Discussion AMD SoC Halo series GPU discussion

Page 21 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

marees

Senior member
Apr 28, 2024
892
1,182
96
Most of the IOD is occupied by a gigantic GPU with Infinity Cache, AMD themselves presented gaming benchmarks in the announcement slide deck and the very first product on shelves to bring Strix Halo was the Asus Republic Of Gamers Flow Z13 that even says the word GAMERS on the top.

View attachment 117587


But it wasn't designed for games. AMD just accidentally put there their biggest PC iGPU ever. The whole thing just slipped in there by accident during design.
They really didn't mean to compete with discrete GPUs like at all.

It's like Viagra.
Doesn’t compete directly with dgpus (in terms of perf/$)

But it is more the package as a whole to compete with 4060/4070 tier GPUs. The ROG Flow is a perfect showcase of this
 

ToTTenTranz

Senior member
Feb 4, 2021
380
726
136
Doesn’t compete directly with dgpus (in terms of perf/$)

But it is more the package as a whole to compete with 4060/4070 tier GPUs. The ROG Flow is a perfect showcase of this

Looking at the ROG Z13's pricing it looks like it competes with the 4070m pretty well in perf/$.

We'll see how the other Halo laptops are priced, though. HP says the G1a will cost "around $1500" though they didn't specify which Halo model that price would include. $1500 is about the cheapest you can get with a 4070m IIRC.


It will always outclass all the 8GB GPUs in recent and future AAA games so the price must reflect that, though. AMD will still have N44 to compete with the cheaper 4060m or even 5060m.
 

DavidC1

Golden Member
Dec 29, 2023
1,424
2,305
96
From what we've seen STX-H seems very similar to what STX+dGPU does for idle power.
If it is similar in battery life to a dGPU machine at low load, then it's not using 1W package power, or it can do 1W but it's useless such as with Icelake's great do nothing battery that plummets the instant you do more than the simplest 1 tab text browsing.

It still means there's a significant battery life cost of doing a big iGPU.
What do you mean "you need to start putting dedicated VRAM", the whole point of an APU is that you have a single shared pool of memory.

And on Strix Halo, just regular old LPDDR5X-8000 memory modules are used.
Ok, that's a positive I did not consider. Still you are redesigning for additional 128 lanes on the board which increases both layout complexity and actual cost.
The only real spot where you lose flexibility is if you need that APU to compete with multiple different dGPU dies, say for example you build a big IOD to compete with AD104, but you also need to cut it down to compete with AD106 based laptops. That... Doesn't seem to be the case here, from what it looks like. Strix Halo looks very focused on competing with AD106.
You can pair any CPU with the dGPU, Intel and AMD. I'm not talking about user flexibility, that has gone the way of the dodo since 15 years ago. dGPU is part of the board but it's still mostly separate.
That's speculation we don't really have a good basis for, frankly. What we do have are the prices of the old Z13 and the prices of the new ones. The top end of the older Z13 outstrips even the 128GB 395 model in terms of pricing. No matter how you look at it, the main issue with Strix Halo pricing is that evidently the 390 is priced much too close to the 395. That's it.
So to get their top config iGPU on desktop APU you need to buy a $350 chip, and for poor performance(respectable for an iGPU, but). And you look at the pricing on their graphics division, including the rumored ones for RDNA4.

And what they are suddenly going to give you significant discount over Nvidia on their halo APU?

How much do people believe Nvidia's portion of the total cost is? $1000 for RTX 4070 mobile?
 
Reactions: scineram

uzzi38

Platinum Member
Oct 16, 2019
2,745
6,623
146
If it is similar in battery life to a dGPU machine at low load, then it's not using 1W package power, or it can do 1W but it's useless such as with Icelake's great do nothing battery that plummets the instant you do more than the simplest 1 tab text browsing.

It still means there's a significant battery life cost of doing a big iGPU.

Well sure. I mean the memory bus is still twice as wide, even if you force it into the lowest power state for both products that won't change. It will always use more power.

Same situation as M4 vs M4 Pro/Max, mind you. The latter have slightly worse idle power for the same reason.

Ok, that's a positive I did not consider. Still you are redesigning for additional 128 lanes on the board which increases both layout complexity and actual cost.

As opposed to designing the board to accommodate for an additional chip with it's own 128b bus, yes.

You can pair any CPU with the dGPU, Intel and AMD. I'm not talking about user flexibility, that has gone the way of the dodo since 15 years ago. dGPU is part of the board but it's still mostly separate.

Designs using AMD and Intel chips feature two different motherboards. That's still two distinct boards you need to design.

Or at least that used to be the case. Nowadays, even for different Nvidia GPUs you need to design different boards again. In the case of Blackwell, 5060 and 5070 need a different board design to 5070Ti and 5080. I believe the same is also true for Ada, below 4080 is a different board design.

This is the reason why the G14 is shipping with both Phoenix and Strix Point this year, and why the low end models get both Phoenix and last year's chassis, while the higher tier models get Strix and the slightly larger chassis. Because they have totally different motherboards too.

So I'm afraid that flexibility you're talking about has already gone the way of the dodo.

So to get their top config iGPU on desktop APU you need to buy a $350 chip, and for poor performance(respectable for an iGPU, but). And you look at the pricing on their graphics division, including the rumored ones for RDNA4.

"Poor performance"?

It is within 5% of the 4070m at "equal" power - 60w package power vs 60w package power or 80W vs 80W. "Equal" power, as in Strix Halo includes both CPU and GPU, whereas 4070m is dGPU only.

There is no world where this is "poor performance".

And what they are suddenly going to give you significant discount over Nvidia on their halo APU?

How much do people believe Nvidia's portion of the total cost is? $1000 for RTX 4070 mobile?
When did I ever suggest that Nvidia is taking $1000 for a 4070?

Again, this pricing is firmly on ASUS. And it's the reason why comparing the Z13's pricing to bargain bin 4070 laptops being discounted ahead of 2025 models going on shelves is a fool's errand. The only thing this pricing shows is that ASUS take advantage of the fact that they know they have a product no-one else really competes with (a 13" laptop capable of console tier gaming performance), and factor it into their pricing.

It shows that you cannot compare the Z13's pricing to those aforementioned bargain bin 4070 devices and claim Strix Halo is far too expensive to make sense.
 

Joe NYC

Platinum Member
Jun 26, 2021
2,908
4,279
106
Most of the IOD is occupied by a gigantic GPU with Infinity Cache, AMD themselves presented gaming benchmarks in the announcement slide deck and the very first product on shelves to bring Strix Halo was the Asus Republic Of Gamers Flow Z13 that even says the word GAMERS on the top.

View attachment 117587


But it wasn't designed for games. AMD just accidentally put there their biggest PC iGPU ever. The whole thing just slipped in there by accident during design.
They really didn't mean to compete with discrete GPUs like at all.

It's like Viagra.

Strix Halo was designed to absorb dGPU.

As far as "gaming laptops", when my daughter started college (to study interior design), they were told what computer to get during their study, and were told to buy a "gaming laptop", which is a shortcut of telling them that the laptop needs to have a dGPU.

So, the "gaming laptops" have wider usage.
 

Glo.

Diamond Member
Apr 25, 2015
5,895
4,935
136
It's cool, but only long term will prove whether it'll be a lasting product or not.

The appeal for iGPUs are:
-Battery life
-Form factor
-Cost, of which can greatly justify the above two

You go high end, and you lose the battery life advantage, even against dGPU systems, and the since the manufacturer of the SoC isn't doing it for charity but an opportunity for more revenue, it's no longer a free one like iGPUs are but whether you are actually saving money over a dGPU.

You are losing the flexibility of a dGPU, and it isn't all about savings either as you need to start putting dedicated VRAM, most likely with exotic packaging techniques.

If it's say $500 cheaper on that Asus system, then you must consider how much of that $500 is artificial tax by both Nvidia and the system manufacturer versus actual cost of production. Remember Nvidia has absolutely monopoly on laptops, even more lopsided than on desktops. Will it even be $100 cheaper on a <$1500 system?
Costs of Intel CPUs are widely available on their Ark site. Now imagine that Nvidia charges more for laptop SKUs than for the desktop dies. Like 100-150$ more.

GA107 cost 300-350$ on desktop? Make it 400-500 on laptop 4070 SKU.

Now pay Intel 550$ for high end CPU, like highest end Core Ultra and you are already in 1000$. Add cost of higher complexity of the board, higher complexity of power delivery, higher complexity of features like Nvidia optimus and switching graphics, and you add costs.

Lets say that AMD charges 800-900$ for Strix Halo, but brings lower power draw, lower complexity of the board, lower complexity of cooling, lower complexity of power delivery, and those lower costs only add up, especially in high volumes.
 

Glo.

Diamond Member
Apr 25, 2015
5,895
4,935
136
Most of the IOD is occupied by a gigantic GPU with Infinity Cache, AMD themselves presented gaming benchmarks in the announcement slide deck and the very first product on shelves to bring Strix Halo was the Asus Republic Of Gamers Flow Z13 that even says the word GAMERS on the top.

View attachment 117587


But it wasn't designed for games. AMD just accidentally put there their biggest PC iGPU ever. The whole thing just slipped in there by accident during design.
They really didn't mean to compete with discrete GPUs like at all.

It's like Viagra.
They 100% intend to compete with dGPUs.

Days of dies like 106 and 107 from Nvidia are numbered, anyway.
 

poke01

Diamond Member
Mar 8, 2022
3,309
4,559
106
All that being said though, full CPU performance on battery is not necessarily the superior alternative you keep pushing for it to be.
I'm not pushing for anything you do you, configure all you like. The comment I made earlier is in regards to eek2121's experience with his ARM Mac on heavy/light loads and he wasn't happy with it and one reason I said is ARM Macs do not throttle CPU performance like x86 OEMs do. If ARM Macs throttled, would that have made a difference? Yes, I think so. It would have resulted in a longer run time.

I'll take the Handbrake benchmark as an example from Hardware Canucks, while unplugged the Z13 is slower than the M4 Pro Macbook because its running at a lower TDP to increase its battery life. The M4 Pro runs at 48 watts, isn't reducing performance and is faster at finishing the task. So one can assume the battery life isn't better than your x86 laptop or rather the efficiency isn't all its claimed to be. This was the point I was trying to convey.
 

jpiniero

Lifer
Oct 1, 2010
15,904
6,389
136
You're judging this by one single design? That targets one single 2kilobuck price point?

Given that Strix Point Laptops basically start at well over a thousand... yeah I'd expect the 256-bit SH models to start at 2k and up. The 128-bit SH model might be somewhere inbetween.
 

Glo.

Diamond Member
Apr 25, 2015
5,895
4,935
136
Given that Strix Point Laptops basically start at well over a thousand... yeah I'd expect the 256-bit SH models to start at 2k and up. The 128-bit SH model might be somewhere inbetween.
HP Z2 Mini G desktops start at 1100-1200$, depending on generation. How much will it cost with Strix Halo?

Hint, the same.
 

Joe NYC

Platinum Member
Jun 26, 2021
2,908
4,279
106
So they will be 256-bit? Just higher clocked? When is LPDDR6?

LPDDR6 will not be ready in time for Medusa Halo.

By "nuclear option" maybe @adroc_thurston means moving to Memory On Package, like Apple.

That may be the only way to compete with Apple on highest end, counter NVidia (which may also use Memory On Package).

Medusa Halo could then add more memory channels, go beyond 256 bit, to be able to scale with additional GPU capabilities. As a quick rule, LPDDR5 is about 1/2 bandwidth per bit, so 256 bit is equivalent to 128 bit in 4060 / 7600

AMD then could define a "package" or even a new socket that is independent of the type of memory used inside. When LPDDR6 arrives, it could be changed internally while still maintaining the same package for laptop makers.
 
Reactions: yottabit

ToTTenTranz

Senior member
Feb 4, 2021
380
726
136
Yeah but only do much they can do with memory on board.
They always have the nuclear option tho.

I wish the nuclear option was HBM3E or HBM4.

It's such a shame that AI hoarded all the HBM tech in the world for the past decade. A Halo APU with a single 12-hi 36GB HBM3E stack doing 1.2TB/s could probably forego its MALL and get an even wider GPU inside the IOD, on an ultimately tiny package.
 
Reactions: lightmanek

ToTTenTranz

Senior member
Feb 4, 2021
380
726
136
It really isn't, when LPDDR is simple and lets you pop up to 128GB.
LPDDR is good for client. HBM isn't.

It is if you want to maximize performance per watt above 20W and/or on a smaller form factor. Like I mentioned, a single HBM3E stack of 36GB would provide enough bandwidth for a 60CU iGPU.
 
Reactions: lightmanek

adroc_thurston

Diamond Member
Jul 2, 2023
5,232
7,310
96
It is if you want to maximize performance per watt above 20W and/or on a smaller form factor
Also nope, max perf/W will be a MoP'd wider LPDDR.
Like I mentioned, a single HBM3E stack of 36GB would provide enough bandwidth for a 60CU iGPU.
Capacity sucks, pop options suck.
The sooner you forget about HBM like a bad dream, the merrier.
It's never gonna be alive outside of HPC.
 
Reactions: scineram

TESKATLIPOKA

Platinum Member
May 1, 2020
2,600
3,160
136
Strix Halo turned out as I expected -> top tier CPU + mediocre iGPU.
Not sure how good RDNA4 will be, but I would have preferred It over RDNA3.5.
I don't have other issues with the chip Itself.

Pricewise It's expensive, but It's Asus ROG, and that was always expensive.
I expect cheaper machines with higher TDP from other vendors, the question is when.

I don't understand why AMD allowed Strix Halo to be placed first in a gaming tablet instead of a proper laptop with 100-125W TDP.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,600
3,160
136
The iGPU is the top-tier thing there.
As a pure iGPU yes, but compared to dGPU It is mediocre, at least no Vram limitation is present.
It does not scale to 100W.
Says who? AMD stated cTDP is 120W.
At best you can say It's not worth the worse power efficiency, but It should still scale up to 120W, If not the iGPU because of BW then at least the CPU.
 

Glo.

Diamond Member
Apr 25, 2015
5,895
4,935
136
In a mini-PC:


I'm not impressed for gaming. He's running most games at 1080p medium and still needs FSR to get over 60 FPS.

It looks like it cost more than my PC I built with an RTX 4070 last year.
Uhhh... Thats NOT Strix Halo. Thats Strix Point with 16 CU GPU.

 
Reactions: gdansk and marees
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |