Question AMD Phoenix/Zen 4 APU Speculation and Discussion

Page 40 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
I

insertcarehere

Senior member
Jan 17, 2013
712
701
136
TESKATLIPOKA said:
If It had the same FPS/$ as CPU+dGPU including Vram, no one would want to buy It.
Do you know how much profit does AMD need to make on an APU to be worth It, because I don't?
Click to expand...

We're talking in circles here then. You're arguing AMD needs to provide value with an APU vs CPU/DGPU setup while I am arguing that AMD would probably be better off selling more CPUs/dGPUs at the pricing where such APU would provide value, classic catch-22.

Remember that a significant portion of APU demand comes from laptop OEMs, and Dell/Lenovo/Asus et al aren't gonna pay higher prices for processors in XPS 13s/Thinkpads/Zenbooks just because there's a bunch more GPU in the die, given said GPU can't be utilized due to inherent form factor limitations, and that even they did, the end consumer may not pay a premium for such extra capability anyways.

TESKATLIPOKA said:
This is not true for this case, because this big Zen4 APU is using N4 and Raphael is using N5, so different production lines.
Click to expand...

Doesn't matter here, AFAIK, wafer costs for both processes are similar and AMD doesn't order N4/N5 separately. Rembrandt being N6 did not prevent supply being utterly terrible initially, presumably because AMD wanted the wafers in N7 form for their other products.
 
Last edited:
  • Like
Reactions: lightmanek
T

TESKATLIPOKA

Platinum Member
May 1, 2020
2,696
3,260
136
insertcarehere said:
We're talking in circles here then. You're arguing AMD needs to provide value with an APU vs CPU/DGPU setup while I am arguing that AMD would probably be better off selling more CPUs/dGPUs at the pricing where such APU would provide value, classic catch-22.

Remember that a significant portion of APU demand comes from laptop OEMs, and Dell/Lenovo/Asus et al aren't gonna pay higher prices for processors in XPS 13s/Thinkpads/Zenbooks just because there's a bunch more GPU in the die, given said GPU can't be utilized due to inherent form factor limitations, and that even they did, the end consumer may not pay a premium for such extra capability anyways.

Doesn't matter here, AFAIK, wafer costs for both processes are similar and AMD doesn't order N4/N5 separately. Rembrandt being N6 did not prevent supply being utterly terrible initially, presumably because AMD wanted the wafers in N7 form for their other products.
Click to expand...
I didn't say this big APU would be for everyone, or that It should have been made instead of the current Phoenix. This was meant as an extra design.

I originally started this talk about big IGP with Strix Point in mind, which I expected as a chiplet design and where IGP could be on a separate chiplet.
Zen4 24CU monolith was a by-product when I calculated cost for Phoenix with different IGP sizes.
I thought It could be profitable to make and there would be demand for this.
You had a different opinion and started looking at "possible" profit loss compared to 16C Raphael.

It looks like your issue is not with this big APU per se, but with AMD not reserving enough wafers.
I automatically calculated with AMD buying(reserving) extra wafers for this APU, so It won't affect other products.
So the question is If there are enough wafers or not.

P.S. Why do you think AMD is reserving capacity for N4/N5 or N6/N7 together?
It should be a different production line, so this doesn't make any sense to me.
TSMC doesn't have an unlimited production capacity for each process, and there are other customers who also buy wafers depending on what they need.
Nvidia for example uses only N4 and no N5 for their consumer GPUs.
If AMD doesn't reserve a specific amount of wafers for each node, then TSMC wouldn't know how much of the unused capacity for either N5 or N4 can be sold to someone else.
 
Last edited:
  • Like
Reactions: Lodix
Heartbreaker

Heartbreaker

Diamond Member
Apr 3, 2006
5,062
6,636
136
Attempt to redirect the APU discussion from the general RDNA 3 architecuture thread, to the actual APU thread.

Mopetar said:
Maybe some company does need to work with AMD for them to want to make product like this, but as the market changes and evolves I think we'll see stronger APUs that creep upwards in capabilities to capture the eroding low-end of the GPU market.
Click to expand...

Isn't this what we have seen for years? Continual iGPU performance evolution/creep upward.

The only disagreement going on is between just that evolution continuing, or the big leap forward (~double more gains in one generation).

I don't think that big of a leap is likely due to memory BW and cost constraints.
 
  • Like
Reactions: Mopetar
Glo.

Glo.

Diamond Member
Apr 25, 2015
5,930
4,991
136
guidryp said:
It's very big for a mainstream APU, and double Phoenix. 100% increase is a VERY BIG increase.

But it's just a rumor, Phoenix was also rumored to be 24 CUs.

Because doubling something is great click bait...
Click to expand...
PHX was rumored to be 6 WGPs, 12 CUs, with each one WGP double the ALUs - 1536. We ended up with 6 WGPs, 12 CUs, 768 ALUs with dual issue.
 
  • Like
Reactions: Tlh97, Mopetar, Lodix and 3 others
Heartbreaker

Heartbreaker

Diamond Member
Apr 3, 2006
5,062
6,636
136
Heartbreaker

Heartbreaker

Diamond Member
Apr 3, 2006
5,062
6,636
136
Glo. said:
View attachment 75434

Snippet from this:

CPU and WGP count the same as Rembrandt(8 CPU cores, 6 WGPs).
Click to expand...

16-24 CU rumor, From the Same Guy 2 months earlier, "coincidentally" same guy is also the source of Strix 24 CU rumor this time.

We are further from Strix Point, than the earlier 24 CU rumors of Phoenix were from Phoenix Launch.

Generally the further from release the more inaccurate (and made up) the rumors.

So his early 24 CU rumor for Strix, is no more reliable than his early 24 CU rumor for Phoenix was.
 
  • Like
Reactions: Tlh97, Kaluan, Mopetar and 3 others
Heartbreaker

Heartbreaker

Diamond Member
Apr 3, 2006
5,062
6,636
136
Another redirect from the Architecture thread.

Joe NYC said:
AMD competition (for the APU, iGPU) ranges from less capable Intel processors to more capable Apple M1/M2, to mobile CPUs paired with dGPU.
Click to expand...

Newsflash. You need to pick one. There is no "one size fits all" in this competition.

Aim low and you can't perform like higher end parts. Aim high and you can't compete on price with the lower offerings.
 
  • Like
Reactions: scineram
T

TESKATLIPOKA

Platinum Member
May 1, 2020
2,696
3,260
136
20% faster with 7500MHz LPPDR5 is just not worth It.
You can't even expand memory.
At least we know why they didn't add more CU, because It's useless.
Did I mention It will be expensive?

It looks like dGPU is the only option for me unless I want worse performance than I currently have with my GTX 1650M. :D
 
L

leoneazzurro

Golden Member
Jul 26, 2016
1,114
1,867
136
TESKATLIPOKA said:
20% faster with 7500MHz LPPDR5 is just not worth It.
You can't even expand memory.
At least we know why they didn't add more CU, because It's useless.
Did I mention It will be expensive?

It looks like dGPU is the only option for me unless I want worse performance than I currently have with my GTX 1650M. :D
Click to expand...

If you have 20% increase in score for 33% increase in bandwidth, this means bandwidth is not the only limiting factor BTW, especially when clock speeds are not known atm.
 
  • Like
Reactions: Tlh97, Kaluan, lightmanek and 2 others
Glo.

Glo.

Diamond Member
Apr 25, 2015
5,930
4,991
136
leoneazzurro said:
If you have 20% increase in score for 33% increase in bandwidth, this means bandwidth is not the only limiting factor BTW, especially when clock speeds are not known atm.
Click to expand...
33% more MEMORY bandwidth. Phoenix has 100% larger caches, with the same amount of resources(768 ALUs) so it should be much less starved for resources than Rembrandt.

And yet, something is not working correctly.

Because Rembrandt already was scoring 3100 pts in TS Graphics with 6400 MHz DDR5 memory. So you have a downgrade in performance with higher bandwidth available to execution units.

I'd say, lets wait for final tweaks. It might fix it, it might not change anything. We shall see.
 
T

TESKATLIPOKA

Platinum Member
May 1, 2020
2,696
3,260
136
leoneazzurro said:
If you have 20% increase in score for 33% increase in bandwidth, this means bandwidth is not the only limiting factor BTW, especially when clock speeds are not known atm.
Click to expand...
This is also true.
BW is 34% better(7500MHz LPDDR5 vs 5600MHz DDR5), but score improved by just 9%.
 
L

leoneazzurro

Golden Member
Jul 26, 2016
1,114
1,867
136
Glo. said:
33% more MEMORY bandwidth. Phoenix has 100% larger caches, with the same amount of resources(768 ALUs) so it should be much less starved for resources than Rembrandt.

And yet, something is not working correctly.

Because Rembrandt already was scoring 3100 pts in TS Graphics with 6400 MHz DDR5 memory. So you have a downgrade in performance with higher bandwidth available to execution units.

I'd say, lets wait for final tweaks. It might fix it, it might not change anything. We shall see.
Click to expand...

IIRC larger caches are only partly true for Phoenix, but only with N31/N32 parts you get the full enhanced config. N33 and Phoenix, except IIRC L0, should have the same cache sizes as RDNA2.
 
A

Abwx

Lifer
Apr 2, 2011
11,868
4,846
136
Computerbase put the 680M at 2400 in their comparison, FTR they tested this IGP in some article.

www.computerbase.de

AMD Radeon 780M: RDNA-3-iGPU aus der Phoenix-APU erstmals im Benchmark

Die AMD Radeon 780M bildet den GPU-Part der Phoenix-APUs der Serie AMD Ryzen 7040 und ist erstmals im Benchmark aufgetaucht.
www.computerbase.de www.computerbase.de

680M review :

www.computerbase.de

AMD Radeon 680M: Die Ryzen-6000-iGPU im Test

AMD Ryzen 6000 setzt erstmals auf eine iGPU mit RDNA 2. Der Leistungssprung, den die Radeon 680M auf dieser Basis im Test macht, ist groß.
www.computerbase.de www.computerbase.de
 
  • Like
Reactions: scineram
HoveringStyle

HoveringStyle

Junior Member
Dec 11, 2022
22
26
51
On the 3DMark website the average TS graphics score for the 680M is indeed ~2400 but the best ones are almost 3000. You can check for yourselves here.
 
Y

yuri69

Senior member
Jul 16, 2013
676
1,208
136
RMB peaks at 12 CUs at 2.4GHz. PHX has also 12 CUs but at 2.8GHz. Arch-wise it's RDNA2 vs RDNA3 so ~20% is about right.

The rumors about doubled CU count likely originate in the "RDNA3 being dual".
 
  • Like
Reactions: Kaluan, TESKATLIPOKA and scineram
E

eek2121

Diamond Member
Aug 2, 2005
3,406
5,043
136
TESKATLIPOKA said:
So he was testing 7940HS, you originally wrote 7840HS.
Even with 54W It can't clock to 3GHz.
Click to expand...
Erm, that would be shared between the GPU and CPU cores unless something has changed. Huge improvement over Cezanne.

Though I will likely wait another generation for my laptop (I rarely use it), I can see the appeal in upgrading.
 
  • Like
Reactions: Kaluan
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom