Question AMD Phoenix/Zen 4 APU Speculation and Discussion

Page 37 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Glo.

Diamond Member
Apr 25, 2015
5,658
4,416
136
I just saw video at BB stated the A1 steppings of PP has some issues that cause low score of Timespy of 2600 which is same as Rembrandt. There is new B steppings which supposedly to be retail version...no words on availability though o_O
BB? What's BB?
 

mikk

Diamond Member
May 15, 2012
4,112
2,106
136
If it's just timespy who cares, they wouldn't need a new stepping? It surely must affect other 3d workloads as well, maybe just certain workloads/games. I would guess it's related to some dx12 features which timespy relies on. Maybe that's why AMD didn't really showcase anything yet until they have a fully working silicon which they can use? Phoenix Point in not just around the corner if this story is true.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
If it's just timespy who cares, they wouldn't need a new stepping? It surely must affect other 3d workloads as well, maybe just certain workloads/games. I would guess it's related to some dx12 features which timespy relies on. Maybe that's why AMD didn't really showcase anything yet until they have a fully working silicon which they can use? Phoenix Point in not just around the corner if this story is true.
Phoenix laptop availability is planned for May, that's for IdeaPad Pro 5 14 and 16 according to notebookcheck. Link
Asus TUF A16 AE with Phoenix was also announced, but without mentioning availability.
Only 3 laptops?

edit: wrong info. Changed It
 
Last edited:

jams3223

Junior Member
Jan 8, 2023
6
4
41
Heard some rumors that the Zen 5 mobile on 3nm will have 10 cores, 8 performance Zen 5 cores and 2 efficiency low clocked Zen 4 cores and there will not be any efficiency gain on laptop but only more battery time with non sustained or idle load, also heard they will improve the AI and the RDNA4 will only have around ~20% performance gain from RDNA3 so around ~6300 TFLOPS FP32 (float) performance. should i wait next year or should i get this year version because i don't know what benefit does efficient core gives unless you're browsing and editing but AMD windows scheduler is very good at this already.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Heard some rumors that the Zen 5 mobile on 3nm will have 10 cores, 8 performance Zen 5 cores and 2 efficiency low clocked Zen 4 cores and there will not be any efficiency gain on laptop but only more battery time with non sustained or idle load, also heard they will improve the AI and the RDNA4 will only have around ~20% performance gain from RDNA3 so around ~6300 TFLOPS FP32 (float) performance. should i wait next year or should i get this year version because i don't know what benefit does efficient core gives unless you're browsing and editing but AMD windows scheduler is very good at this already.
Don't know where you heard those rumours from.
If TDP doesn't increase for Strix Point but performance increases, then efficiency increased, which is expected from a better process.
Phoenix has 12CU*128ALU*2FLOP*3000MHz= 9216 GFLOPs, but that's including dual issue.

You didn't even say for what usage you want a new laptop and we don't know Phoenix or Strix Point performance.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Here are the specs for Phoenix models.

AMD webpage: AMD Ryzen 7000 Series Processors
R9 7940HS -> 8C16T, Base: 4GHz, Boost: 5.2GHz, 16MB L3; TDP: 35-54W; IGP: 12CU 3000 MHz
R7 7840HS -> 8C16T, Base: 3.8GHz, Boost: 5.1GHz, 16MB L3; TDP: 35-54W; IGP: 12CU 2900 MHz
R5 7640HS -> 6C12T, Base: 4.3GHz, Boost: 5GHz, 16MB L3; TDP: 35-54W; IGP: 8CU 2800 MHz

Memory support:
DDR5-5600
LPDDR5x-7500
 

Tuna-Fish

Golden Member
Mar 4, 2011
1,324
1,462
136
Separate? IC is L4 cache for the CPU, not L3.

It's not. IC is an extension of the L3, not a separate level. That is, on a CPU that has IC, when a line is evicted from L2, it will be placed in a L3 slice based on it's hashed address, and some of the lines go to the IC, others go to normal L3.

Imo the far more interesting benefit of CAMM would be the ability to have socketed LPDDR. It could make laptops much more upgradable/repairable, and even make LPDDR viable for desktops. It is faster, after all, so if the soldered limitation is removed, why not switch almost everything to it?

LPDDR on a socket isn't faster or lower power than DDR. LPDDR achieves what it does precisely because of short traces and very-low capacitance paths. This is antithetical to sockets.
 

jams3223

Junior Member
Jan 8, 2023
6
4
41
Here are the specs for Phoenix models.

AMD webpage: AMD Ryzen 7000 Series Processors
R9 7940HS -> 8C16T, Base: 4GHz, Boost: 5.2GHz, 16MB L3; TDP: 35-54W; IGP: 12CU 3000 MHz
R7 7840HS -> 8C16T, Base: 3.8GHz, Boost: 5.1GHz, 16MB L3; TDP: 35-54W; IGP: 12CU 2900 MHz
R5 7640HS -> 6C12T, Base: 4.3GHz, Boost: 5GHz, 16MB L3; TDP: 35-54W; IGP: 8CU 2800 MHz

Memory support:
DDR5-5600
LPDDR5x-7500

I was planning on getting the 7940Hs audio production with the rtx 4050 for casual gaming and ai capability for programming as i am a vst and app programmer.

Yeah i understand that the limit will be in the 8-9 TFLOPS range but it's under high TDP load in the 45-55W, i meant that it perform 5.2 TFLOPS at low wattage.

I am still trying to figure out if i can get the rog flow x13 2023 along a 4050 or will the iGPU be enough ??? i don't know how much i can get from the 4050 though.

I use the Revi OS which is a faster version of windows 11, it's pretty snappy so everything will be smooth just like using apple silicon.
 
Last edited:
  • Like
Reactions: lightmanek

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
LPDDR on a socket isn't faster or lower power than DDR. LPDDR achieves what it does precisely because of short traces and very-low capacitance paths. This is antithetical to sockets.
The SODIMM form factor is limiting, but CAMM is close enough to soldered to support native, full speed LPDDR. There'll probably be articles sometime this year about it as the spec finishes making its way through JEDEC.
 
  • Like
Reactions: RnR_au

RnR_au

Golden Member
Jun 6, 2021
1,675
4,079
106
The SODIMM form factor is limiting, but CAMM is close enough to soldered to support native, full speed LPDDR. There'll probably be articles sometime this year about it as the spec finishes making its way through JEDEC.
Found a bit of an overview about CAMM. Sounds good. Thanks for the headsup :beercheers:

 
  • Like
Reactions: Kaluan

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
The SODIMM form factor is limiting, but CAMM is close enough to soldered to support native, full speed LPDDR. There'll probably be articles sometime this year about it as the spec finishes making its way through JEDEC.
Let's say DELL manages to make a standard of It this year. Then we still need to wait for It to be used in laptops of other manufacturers.
I don't think It will happen that fast.

Found a bit of an overview about CAMM. Sounds good. Thanks for the headsup :beercheers:

Thanks for the link. It's the best article about CAMM I saw so far.
Can't say I am a big fan of CAMM, It has Its disadvantages, but at least capacity is high and It's not soldered.
The soldered RAM is a no-go for me, especially when so many laptops are limited to 32 GB. and I am talking about the just announced ones.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
Let's say DELL manages to make a standard of It this year. Then we still need to wait for It to be used in laptops of other manufacturers.
I don't think It will happen that fast.
Sure, it probably won't see any serious adoption till '24 or even later, but it holds a lot of promise for the future.
Can't say I am a big fan of CAMM, It has Its disadvantages, but at least capacity is high and It's not soldered.
What disadvantages do you see? If it's that particular form factor, then I think Dell and Intel are pushing for a couple of different ones.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Doesn't the low performance of 780M's performance in Time Spy support the claim of only being 21% faster than Iris Xe?

Sounds like the stepping problem that was attributed to dGPU RDNA3, may have been mobile all along.

Alternatively, maybe the RDNA3 does lower performance/clock. The less-than-50% performance gain we see with 7900XTX is with the latter having 20% more "legacy" shaders and 50% more TMUs and ROPs, while bandwidth improved by 50% as well. 780M has the same shader count and doesn't increase on TMUs and ROPs, while bandwidth doesn't really improve.

There's a separate thing on top of that which is the actual perf/w gain.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
I was planning on getting the 7940Hs audio production with the rtx 4050 for casual gaming and ai capability for programming as i am a vst and app programmer.

I am still trying to figure out if i can get the rog flow x13 2023 along a 4050 or will the iGPU be enough ??? i don't know how much i can get from the 4050 though.
According to Ultrabookreview.com you can choose from 4 models. They mention the specs and comparison with the older model, check It out.
Asus ROG Flow X13 GV302XI – AMD Ryzen 9 7940HS + RTX 4070 60W + QHD display;
Asus ROG Flow X13 GV302XV – AMD Ryzen 9 7940HS + RTX 4060 60W + QHD display;
Asus ROG Flow X13 GV302XU – AMD Ryzen 9 7940HS + RTX 4050 60W + QHD display;
Asus ROG Flow X13 GV302XA – AMD Ryzen 9 7940HS + Radeon 780M + FHD display.
Amount of RAM and size of SSD is also different.
If the difference in price relative to price of the cheapest models is not that much, I would probably choose at least RTX 4050.
RTX 4070 and RTX 4060's performance gain over RTX 4050 is questionable thanks to 45+15W power limit.

Yeah i understand that the limit will be in the 8-9 TFLOPS range but it's under high TDP load in the 45-55W, i meant that it perform 5.2 TFLOPS at low wattage.
Where did you get this info about 5.2 TFLOPs at low wattage? I didn't see It yet.
5.2TFLOPs /2/128/12= ~1693 MHz for 12CU IGP

I use the Revi OS which is a faster version of windows 11, it's pretty snappy so everything will be smooth just like using apple silicon.
Didn't hear about this Revi OS, I will check It out. Thanks.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Sure, it probably won't see any serious adoption till '24 or even later, but it holds a lot of promise for the future.

What disadvantages do you see? If it's that particular form factor, then I think Dell and Intel are pushing for a couple of different ones.
Better than soldered RAM, that's for sure.
Yes, that form factor is really huge, but It's for 128 GB.
They mentioned It's possible to also put chips from the other side too.
I would prefer having modules from both sizes, this way It would be smaller.

They have to make different form factors, kind of like M2 for SSD. It would be hilarious If they put only 32GB on the same form factor as the one with 128GB.

edit: I thought this one was a 128GB version, It turns out to be only 32GB. Dimensions are 9.8 x 6.8 cm in size.
StorageReview-Dell-Precision-7670-1.jpg
csm_MG_7459_08a669abcd.jpg

It looks like 2 SO-DIMM slot with installed memory will take less space and could be next to each other or above one other or against each other.
 
Last edited:

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Doesn't the low performance of 780M's performance in Time Spy support the claim of only being 21% faster than Iris Xe?

Sounds like the stepping problem that was attributed to dGPU RDNA3, may have been mobile all along.

Alternatively, maybe the RDNA3 does lower performance/clock. The less-than-50% performance gain we see with 7900XTX is with the latter having 20% more "legacy" shaders and 50% more TMUs and ROPs, while bandwidth improved by 50% as well. 780M has the same shader count and doesn't increase on TMUs and ROPs, while bandwidth doesn't really improve.

There's a separate thing on top of that which is the actual perf/w gain.
That TimeSpy performance shows ~680M level of performance, but we don't know If 680M performs only 21% better in those 12 games on average than Iris Xe.
There could really be a problem with that stepping.

RDNA3 doesn't have lower performance/clock. Performance of N31 is not that bad considering dual issue is helping only 9% on average, so the 47% higher gaming performance is mainly coming from more WGPs(CUs) and higher clocks.

Frequency medianWGP (CU)TMUROPsInfinity cacheGDDR6 widthGDDR5 speedBandwidthTPU 4K
RX 6900XT2227 MHz (100%)40 (80)320 (100%)128 (100%)128MB (100%)256-bit (100%)16gbps (100%)512GB/s(100%)100%
RX 7900XTX2631 MHz (100%)48 (96)384 (120%)192 (150%)96MB (75%)384-bit (150%)20gbps (125%)960GB/s(188%)147%
I won't bother calculating Pixel or Texture Fill rate.

If BW or ROPs are not a bottleneck, then It doesn't matter how much increase they see, bottleneck will be elsewhere.
TMUs also shouldn't be a bottleneck, or AMD would have increased the amount per WGP.
The biggest problem with N31 is that AMD didn't increase WGPs more than 20% , yet they could have easily made a 60WGP model and then that dual issue SIMD.

Back to 780M. Yes, It probably has the same specs, but graphics frequency(boost?) is 3GHz or 25% higher than Rembrandt.
Thanks to a much better process, I expect significantly higher gaming frequency, but only 17% better BW is not much.
Extra performance should come with driver updates later.
780M will be faster than 680M, the question is by how much. It will depend on achievable clock speed. The highest gain we should see with U models.
 
  • Like
Reactions: Kaluan

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
@TESKATLIPOKA What you are saying is in theory though. Reality can be often different because it depends on the low level details.

Also, bottlenecks are generally not a hard limit, unless it's extremely unbalanced. So if you increase everything by 50%, you'll see 50% gains. Because inevitably with games that vary vastly between them, some will be fillrate bottlenecked, some will be compute bottlenecked, and some bandwidth. Especially at this low performance level.

I did find the lack of mention of graphics strange when they first announced it. Now I know why.
 
Last edited:
  • Like
Reactions: lopri and coercitiv

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Not sure which part you mean? What I said about bottlenecks? Yeah, It's theory.
I just wanted to point out that RDNA3 doesn't have lower performance/clock than RDNA2.

Yes, that's exactly what I mean. Clearly at the moment there's something seriously wrong, and that's what can happen in reality.

And since we cannot have a RDNA3 part with same CUs/TMU/ROP/Bandwidth what you are saying is still just an assumption. Not to mention iGPU is WAY different.

It happens even with CPUs where the performance can be much more predictable since it's all hardware. Like Rocketlake was a disappointment performance-wise that went under how a theoretical power uncapped Sunny Cove should have performed.
 
  • Like
Reactions: TESKATLIPOKA

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Yes, that's exactly what I mean. Clearly at the moment there's something seriously wrong, and that's what can happen in reality.

And since we cannot have a RDNA3 part with same CUs/TMU/ROP/Bandwidth what you are saying is still just an assumption. Not to mention iGPU is WAY different.
It can be done with N23 vs N33, specs are the same. But clocks need to be adjusted.
 

jams3223

Junior Member
Jan 8, 2023
6
4
41
According to Ultrabookreview.com you can choose from 4 models. They mention the specs and comparison with the older model, check It out.

Amount of RAM and size of SSD is also different.
If the difference in price relative to price of the cheapest models is not that much, I would probably choose at least RTX 4050.
RTX 4070 and RTX 4060's performance gain over RTX 4050 is questionable thanks to 45+15W power limit.


Where did you get this info about 5.2 TFLOPs at low wattage? I didn't see It yet.
5.2TFLOPs /2/128/12= ~1693 MHz for 12CU IGP


Didn't hear about this Revi OS, I will check It out. Thanks.

Yep it was a test someone i know from china did and he got these results but he didn't wanna give me further information.

The Revi OS is an incredible version of windows which is optimized to run fast on any machine, also it removes all the useless stuffs on you're machine and give you full utilization of your CPU, RAM and GPU.

Just make sure to undervolt the CPU, GPU and overclock the ram a little with the UXTU app paired with the Revi OS then you have a snappy machine.
 
Last edited:

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Yep it was a test someone i know from china did and he got these results but he didn't wanna give me further information.

The Revi OS is an incredible version of windows which is optimized to run fast on any machine also it removes all the useless stuffs on you're machine and give you full utilization of your CPU, RAM and GPU.

Just make sure to undervolt the CPU, GPU and overclock the ram a little with the UXTU app paired with the Revi OS then you snappy machine.
Didn't he at least say what was the power limit? 15 or 25W?

I already downloaded Revi OS. I will try It as a virtual machine. Not sure If I want to exchange this for my laptop windows.

Can't say I have a problem with my system. It's just an older 4C8T Intel + GTX 1650 but I have 24GB RAM installed and 2x SSD not in raid. I already use ThrottleStop.

In my next system, I would like to have 64GB RAM. Soldered memory is pretty much impossible for that, so only a laptop with 2 SO-DIM slots.
 

eek2121

Platinum Member
Aug 2, 2005
2,904
3,906
136
Are the new chips that use LPDDR5 using single channel or dual channel? I have seen mixed info. Single channel would be such a waste. Dual channel would serve the iGPU well.

So Adreno is a desktop class GPU architecture? This bodes well, since Qualcomm is looking to become a major player in the PC market with their upcoming Nuvia based SoCs

Make sure you hold your breath for that.

Unless you mean chromebooks and such. Even that is questionable due to the lawsuit. ARM will never take over the desktop unless it can offer superior perf/watt at similar/lower prices with the expansion capability of a modern x86 machine. Even then, the chip would have to be fast enough to emulate leading edge x86 chips at full speed to get enthusiasts, gamers, or power users to buy in. Basically impossible. Even for everyone else, Qualcomm’s love for margins means that they will have a very hard time competing with AMD/Intel. Also don’t forget that both AMD and Intel have the full capability to compete. Qualcomm would have a very hard fight on it’s hands against two multibillion dollar backers of x86.

I hope Microsoft ports Windows to RISC-V and builds an x86-64 emulator for it. As you can build a RISC-V chip royalty free (there are even open source implementations), RISC-V has much more promise.

Don’t get me wrong, a third competitor would be great, but Qualcomm is too scummy of a company to be that third competitor. There is a reason they have a near monopoly in the mobile/ARM space. Go look up their history.

Okay, I am done preaching now.