Massive AMD Ryzen 8000 Zen 5 leak points to Strix Halo chiplet APUs targeting Apple silicon efficiency with RTX 4070-class graphics along with new Strix Point and Fire Range APUs

[Redaktion]

New information about AMD's upcoming mobile APUs have leaked online indicating significant upgrades to CPU and GPU performance. Next gen APUs include Hawk Point and Strix Point for the Premium Mobile segment, Strix Halo that is slated to take on Apple silicon with RTX 4070-class RDNA 3.5 graphics, and Fire Range, which is the purported successor to Dragon Range.

https://www.notebookcheck.net/Massive-AMD-Ryzen-8000-Zen-5-leak-points-to-Strix-Halo-chiplet-APUs-targeting-Apple-silicon-efficiency-with-RTX-4070-class-graphics-along-with-new-Strix-Point-and-Fire-Range-APUs.709446.0.html

[NikoB]

Actually, the whole value of these rumors about Zen 5 (when the false AMD still cannot release Zen4 Phoenix laptops, although it is already the end of April 2023) is that AMD admits to its significant technological loss to Apple by more than 2.5 years !!!

256-bit memory controller (moreover, soldered, like Apple, and therefore significantly limited in volume, and we need ALREADY 64GB minimum, if the built-in video chip requires 16-24GB for share memory - otherwise what is their point in games?), in 2 times slower than Apple's 512-bit memory controller in the M2 Max, which is already in production and on sale.

At the same time, GDDR6x video memory in NVidia solutions is still 2-5 times faster, which means that there will be no performance level of GTX4070, a priori. At best, some "GTX4050Ti". Sometime in AMD's distant hazy future....

The shameful 128-bit memory controller in the Zen4 7945HX cannot even physically handle the 28 pci-e 5.0 lanes that AMD integrated there for some reason, instead of the integrated graphics of the Zen4 Phoenix level (which we have not even seen in any review), even without Taking into account the integrated video card, chips of the 7945HX level already need memory with a bandwidth of at least 400Gbyte/s, at least 200. But in reality, the slow 7945HX memory controller has a shameful bandwidth of only 60-65Gbyte/s! It's like putting tractor wheels on a Formula 1 car...

The entire x86 industry is in a dead end - the memory bandwidth ALREADY should be at least 3 times more than its best solutions in the consumer mobile segment (and this is no more than 85Gbyte/s on average and only for Intel, not for AMD, whose memory controllers are traditionally lose to Intel controllers by about x1.3-x1.5) to cater for all ports and technologies...

Apple needs to finish off the x86 camp by making its Arm platform freely available to all laptop manufacturers. Only in this way will it be able to turn the tide with x86's overwhelming market share and overwhelming software market share. Someone at Apple, as brave as Jobs once was, must finally make this decision.

[rk]

The entire x86 industry is in a dead end - the memory bandwidth ALREADY should be at least 3 times more than its best solutions in the consumer mobile segment

Memory controller is not specific to x86. You can have quad channel, HBM or other memory (like in consoles).

The shameful 128-bit memory controller in the Zen4 7945HX cannot even physically handle the 28 pci-e 5.0 lanes that AMD integrated there for some reason

It's a desktop CPU binned for laptops.

Apple needs to finish off the x86 camp by making its Arm platform freely available to all laptop manufacturers.

You already can buy Qualcomm Windows on ARM laptops with "big" uplift coming end of 2024 with Nuvia based designs (depending how the lawsuit will go). Aside of that there are Mediatek offerings, mostly Chromebooks.

WoA has close to no adoption and software compatibility leaves a lot to be desired. Not sure why I would want an ARM SoC right now. If it runs Linux, has 32GB RAM, one or two M.2 slots and has performance no weaker than like Ryzen 4800U then I can happily work on it. But if it as expensive as current Qualcomm devices then I want also Windows and x86 app, drivers and games compatibility and ability to run some lighter games.

ARM SoCs are way more proprietary and limited than x86. It's HARD to get an ARM based device with upgradable RAM (or one with 32+GB of soldered RAM for work, no to mention 16GB is rare) while M.2 storage is also only available in selected models.

2 times slower than Apple's 512-bit memory controller in the M2 Max

And 99% of the time you don't need that bandwidth or unified memory controller. M2 Max is a top shelf high end part and most sales volume is done in every lower bracket. None of the laptop vendors needs this to print a lot of money from selling laptops.

At the same time, GDDR6x video memory in NVidia solutions is still 2-5 times faster, which means that there will be no performance level of GTX4070, a priori. At best, some "GTX4050Ti". Sometime in AMD's distant hazy future....

680M and 780M are memory starved already but the performance is good for what they are. For anything bigger that we have now it's rather obvious the design would be different. Either caches like Intel is trying or some sort of VRAM like HBM E tile on a chiplet based SoC (similar to Kaby Lake G). UMC for those future Ryzens could also be an option. CPU wants low latency while GPU wants high bandwidth and that's why we have two different memory types.

The thing with such CPU+GPU packages as shown by Kaby Lake G is that you concentrate heat generation to one spot instead of two separate. CPU and iGPU will cannibalize each other thermals...

[kuro68k]

Okay, but what about Thunderbolt/USB 4 support?

To date both Ryzen 6000 and 7000 series have been hobbled by not having native USB4/TB support, and the few laptops that do have a couple of ports suffer from poor performance on them.

It seems like Ryzen mobile has too few PCIe lanes in general, because you never see 2xNVMe and 2xUSB4 in the same machine.

[NikoB]

To date both Ryzen 6000 and 7000 series have been hobbled by not having native USB4/TB support

Zen4 Phoenix - first Zen with USB40 integrated 2 ports. But USB40 not equal by capabilities for TB4 controllers from Intel...

At the moment, both USB40 and TB4 are morally obsolete - even TB5 at 80 Gbps, not even appearing in laptops, is already morally obsolete, because. you already need at least 200-300Gbps to service modern video cards in eGPU mode.

Copper cables can no longer provide the required speeds. time to switch to optics. But as I already wrote in other topics, the problem is in the extremely slow memory controllers of all x86. In order for everything to work without friezes and lags with such ports and the pci-e 5.0 bus, memory controllers must be 3-5 times faster than they are now.

x86 is at a dead end when it comes to RAM performance.

[lol]

To date both Ryzen 6000 and 7000 series have been hobbled by not having native USB4/TB support

Zen4 Phoenix - first Zen with USB40 integrated 2 ports. But USB40 not equal by capabilities for TB4 controllers from Intel...

At the moment, both USB40 and TB4 are morally obsolete - even TB5 at 80 Gbps, not even appearing in laptops, is already morally obsolete, because. you already need at least 200-300Gbps to service modern video cards in eGPU mode.

Copper cables can no longer provide the required speeds. time to switch to optics. But as I already wrote in other topics, the problem is in the extremely slow memory controllers of all x86. In order for everything to work without friezes and lags with such ports and the pci-e 5.0 bus, memory controllers must be 3-5 times faster than they are now.

x86 is at a dead end when it comes to RAM performance.

I hope you realise that x86 requires less bandwidth and ram due to it's cisc architecture you got the Dunning Kruger effect

Actually, the whole value of these rumors about Zen 5 (when the false AMD still cannot release Zen4 Phoenix laptops, although it is already the end of April 2023) is that AMD admits to its significant technological loss to Apple by more than 2.5 years !!!

256-bit memory controller (moreover, soldered, like Apple, and therefore significantly limited in volume, and we need ALREADY 64GB minimum, if the built-in video chip requires 16-24GB for share memory - otherwise what is their point in games?), in 2 times slower than Apple's 512-bit memory controller in the M2 Max, which is already in production and on sale.

At the same time, GDDR6x video memory in NVidia solutions is still 2-5 times faster, which means that there will be no performance level of GTX4070, a priori. At best, some "GTX4050Ti". Sometime in AMD's distant hazy future....

The shameful 128-bit memory controller in the Zen4 7945HX cannot even physically handle the 28 pci-e 5.0 lanes that AMD integrated there for some reason, instead of the integrated graphics of the Zen4 Phoenix level (which we have not even seen in any review), even without Taking into account the integrated video card, chips of the 7945HX level already need memory with a bandwidth of at least 400Gbyte/s, at least 200. But in reality, the slow 7945HX memory controller has a shameful bandwidth of only 60-65Gbyte/s! It's like putting tractor wheels on a Formula 1 car...

The entire x86 industry is in a dead end - the memory bandwidth ALREADY should be at least 3 times more than its best solutions in the consumer mobile segment (and this is no more than 85Gbyte/s on average and only for Intel, not for AMD, whose memory controllers are traditionally lose to Intel controllers by about x1.3-x1.5) to cater for all ports and technologies...

Apple needs to finish off the x86 camp by making its Arm platform freely available to all laptop manufacturers. Only in this way will it be able to turn the tide with x86's overwhelming market share and overwhelming software market share. Someone at Apple, as brave as Jobs once was, must finally make this decision.

[dr strange]

At the moment, both USB40 and TB4 are morally obsolete - even TB5 at 80 Gbps, not even appearing in laptops, is already morally obsolete, because. you already need at least 200-300Gbps to service modern video cards in eGPU mode.

Have you taken your medicine today?

[davidm]

Lots of good info in this thread, though it would be nice if people wouldn't resort to insults. Everyone is learning, right?
I am surprised local AI isn't discussed more often for x86 performance. Surely you'd want a local AI. It benefits immensely from fast memory.
Also not mentioned, battery life. We're talking laptops, right? Most of the time, it's more important than raw performance.

[Sinocelt]

Hard to get excited about Phoenix+ when Phoenix was "launched" in January yet still is nowhere to be seen. When will we see Phoenix+ notebooks? In Q3 or Q4 of 2024? So it'll compete with Lunar lake and Apple's M3 or even M4, not with Apple's M2.

[kek]

"Strix Halo that is slated to take on Apple silicon with RTX 4070-class RDNA 3.5 graphics"

lmao, sounds like AMD fanboy mental delusions. Not even their current card lineup can match Nvidia (and that's without power limits like on a laptop).

Maybe they meant "RTX 2070"

[bk]

"Strix Halo that is slated to take on Apple silicon with RTX 4070-class RDNA 3.5 graphics"

lmao, sounds like AMD fanboy mental delusions. Not even their current card lineup can match Nvidia (and that's without power limits like on a laptop).

Maybe they meant "RTX 2070"

Totally agree that a 4070 in an APU sounds delusional, but both of AMD's 7900 cards are quite a bit faster than the RTX 4070 and implying that they aren't is just not true.

[Poster]

In a joker voice I have this to say about AMD announcements/leaks, BLA,BLA-BLA,BLA,BLA...

Paper news/launches, broken promises and products nowhere to be found and then released 1.5 years later at the same price of new products. One minute of temporary peaked interest then, *YAWN* - moving on..

[RobertJasiek]

both of AMD's 7900 cards are quite a bit faster than the RTX 4070

Depends on applications: 3D game raster: yes. Compute: mostly no.

[A]

The entire x86 industry is in a dead end - the memory bandwidth ALREADY should be at least 3 times more than its best solutions in the consumer mobile segment (and this is no more than 85Gbyte/s on average and only for Intel, not for AMD, whose memory controllers are traditionally lose to Intel controllers by about x1.3-x1.5) to cater for all ports and technologies...

Apple needs to finish off the x86 camp by making its Arm platform freely available to all laptop manufacturers. Only in this way will it be able to turn the tide with x86's overwhelming market share and overwhelming software market share. Someone at Apple, as brave as Jobs once was, must finally make this decision.

The day that x86 becomes obsolete and we get everything ARM soldered is the day I buy the WEAKEST ARM laptop with best battery life and use it as a zero client. Memory usage of stuff is only going up, the paltry 8gb memory they make the minimum is not enough, and I am not fond of keeping my data on a soldered computer. I would have to sync everything to cloud and smash the computer if they fail. Most don't offer options for 64+ gb ram, and if they do expect to pay crazy overpriced prices.

So at that point, why do I need a powerful laptop anymore? I can get far more power from remoting into a desktop and it will cost me less too

[usacomputer]

I find it incredible how everyone is betting on AMD ZEN 4 7040 Phoenix. We can see it in all the small consoles that come onto the market, for example: ASUS ROG Ally, AOKZOE A2, AYANEO NEXT 2. What the public wants now is to see it in light laptops of less than 2.2lb since with RNDA 3 it has achieved graphic power and being able to play AAA games and render Videos and 3D smoothly impossible on intel.
What many people are interested in about this ZEN 4 7040 is the artificial intelligence with XDNA architecture developed by Xilinx and everything at 4nm vs 10nm from Intel, how this AMD artificial intelligence will be integrated into current laptops and in the future Windows 12, Office and the new Direct XII APIs.
We want to see laptops with ZEN 4 7040 Phoenix ORANGE label NOW, because it is the only profitable quality/price