Discussion AMD's Soundwave ARM APU: The Beginning of Transformation !!!

[Panino Manino]

Hum... maybe Samsung wants to do more and better things offline in their Galaxies.
They are already the best phone "AI".

 

[Tup3x]

Hum... maybe Samsung wants to do more and better things offline in their Galaxies.
They are already the best phone "AI".

Imo server side AI is pretty pointless. It would be much better if on-device processing would increase.

 

[soresu]

Imo server side AI is pretty pointless. It would be much better if on-device processing would increase.

Thanks but I think on device usefulness peaked at low light photo noise reduction.

 

[Tup3x]

Thanks but I think on device usefulness peaked at low light photo noise reduction.

Not in Samsung's case.

 

[soresu]

Not in Samsung's case.

No that was my opinion.

That's me stating I really don't care for the rest of it, especially if it's in semi regular use when I'm not even using the phone.

 

[Tup3x]

No that was my opinion.

That's me stating I really don't care for the rest of it, especially if it's in semi regular use when I'm not even using the phone.

I meant that Samsung's images are extremely noisy. They aren't really doing anything meaningful on-device.

 

[soresu]

I meant that Samsung's images are extremely noisy. They aren't really doing anything meaningful on-device.

Ah k, I haven't really made much of an effort to check myself.

I know that they make a big noise about their ISOCell sensors, but I think usually Sony's sensors have the upper hand in quality, though you have to go out of your way to find out which any given phone is using.

 

[coercitiv]

I meant that Samsung's images are extremely noisy. They aren't really doing anything meaningful on-device.

It's mostly just gimmicks, like how they use machine learning to detect the moon in pics and introduce details in photos that do not come from the sensor. (more testing on reddit)

 

[Thibsie]

Ah k, I haven't really made much of an effort to check myself.

I know that they make a big noise about their ISOCell sensors, but I think usually Sony's sensors have the upper hand in quality, though you have to go out of your way to find out which any given phone is using.

Sony sensors are good. What manufacturers do with'em though is extremely variable.
Even Sony cameras sometimes produced crap.

 

[Tup3x]

Ah k, I haven't really made much of an effort to check myself.

I know that they make a big noise about their ISOCell sensors, but I think usually Sony's sensors have the upper hand in quality, though you have to go out of your way to find out which any given phone is using.

They'll be using the same sensors for base S26 as S23 (while the main camera isn't exactly the same as S22, in practice it is). Those cameras never were great hardware wise. UW doesn't have AF, tele is garbage pin sized 10MP camera that shoots upscaled images by default. Main camera has poor lense. They could probably achieve better results if they would completely redo their image processing but currently they are far behind everyone.

S23 especially had horrible sharpening yet it did nothing to improve the actual detail. Noise is horrible and there's visible unsharp ring in pretty much every image (especially noticeable if you shoot something like A4 paper).

I got so fed up with Samsung and their lack of progress that just bought vivo X300. Lets just say that the difference in battery life and cameras is not small. Not a fan of importing phones but anyway...

 

[Panino Manino]

I meant that Samsung's images are extremely noisy. They aren't really doing anything meaningful on-device.

Sony sensors are good. What manufacturers do with'em though is extremely variable.
Even Sony cameras sometimes produced crap.

From personal experience, Samsung's sensors are not good. The software has to work extra hard with Samsung sensors (and just like Sony doesn't seem to be able to use it's own sensors right on their Xperias, the same can be said about Samsung). When I switched from a phone with a Samsung sensor to a Sony sensor I got very surprised by the difference in quality.
From what I know OmniVision sensors are on par with Sony, and not both have 200mp sensors for those top dog phones's telephotos, Samsung market share will take a big hit.

 

[Tigerick]

Hoho, let's decode the above statement regarding Exynos 2600's iGPU: Xclipse 960 aka AMD's Juno RDNA IP. It is based on new architecture; definitely refer to new generation of RDNA. Before I proceed, let's assume the story between Samsung and AMD: Samsung wants to showoff more selling features of X960 but AMD does not want to reveal too many details about new RDNA IP because it is the same IP would be used for all future APUs (N3P). AMD definitely have a final word on the statement; thus, the claims should be correct.

Check my table regarding the relationship between STX (N4P) and E2500 (SF3): the performance delta is caused by the clock speed; the features should be similar...Exynos 2500 is using RDNA3+/3.5 IP; thus it is safe to assume Xclipse 960 is using RDNA4+/4.5 IP. However RDNA4s (N4P) are designed for desktop dGPU; some leakers keep saying RDNA4 are not for mobile SoC, then what IP Samsung/AMD is using? Let's check the features below:

• The computing performance of the Exynos Xclipse 960 GPU is twice as high as that of its predecessor. This is bold claim: Samsung claims computing performance is double than Xclipse 950. Hoho, even RDNA4's improvement is only 20% faster in rasterization's IPC. Then where is 100% performance claim from? Hoho, if you read my speculation about doubling SP per CU aka Compute Unit; then with same clock speed, I could claim double performance per CU as well. It is not full picture cause the number of CUs has been cut by half, that's why FP32 still remains the same.
• If above feature is considered overselling, then the second statement is underselling: This drives a ray tracing performance improvement of up to 50%. With 8CU, X960 still manage to offer 50% better RT performance than X950's 16CU/RT. Hoho, anyone wants to estimate how powerful is each Radiance core in X960?

As I speculated before, Soundwave (N3P) is like bigger brother of E2600. E2600 is having 8CU with 4.1TF, then how many CUs would Soundwave having? Don't believe what AMD marketing people told you; I rather believe in MLID's story regarding Senior Management about SWV. And if anyone still believes in Medusa Point is having RDNA3.5+ GPU, hoho...

DF has shown double frame rate (with RT) comparing 12Xe PTL and 16CU STX. Hoho, could you imagine RT performance of MDS1 with "only" 8CU RDNA3.5+ ? And the competitor of MDS1 is not even PTL; it is mobile NVL-H with faster iGPU and 80TOPS.

Above spreadsheet is created by me to compare Qualcomm and AMD mobile APU lineup. Qualcomm only need 2 dies to offer their WoA APU from top to bottom. FYI, Gorgon Point consists of 3 old dies: STX, KRK1 and KRK2 which will be replaced by MDS1, MDS2 and MDS3. Each die will be upgraded to support XDNA3 >80 TOPS: a feature also decoded by me before AMD's FAD. AMD seems to make changes in the CPU configuration to accommodate changes in iGPU and NPU. Still MDS1 (18 threads) is more than NVL-H (16 threads), it is the changes in iGPU I will be talking today.

The amount of CU will be cut by half in MDS1-3, do you really think AMD is going to use same old GPU architecture in newer N3P node: the same node as upcoming RDNA5? Here is my speculation, AMD will double SP per CU; thus the FP32 TF will remain if AMD is maintaining the clock speed which seems unlikely. For x86 APU, AMD will set higher clock speed for its iGPU and NPU: the removal of 4 E-core might be the indicator.

Before Samsung's claiming of 50% better RT performance, I already estimated 3-4 times faster RT in the thread: RDNA4.5 iGPU = Subset of RDNA5 GPU: turn out each Radiance core should be 3 times faster in order to meet the claim of 50% faster RT if the clock speed remains the same. With 3 times faster RT, MDS1's 8 RT should be comparable to NVL-H's 12 RT depending on the final clock speed. Geez, does it make sense to you now? Or do you really believe AMD is incapable of updating iGPU architecture?

 

[marees]

do you really believe AMD is incapable of updating iGPU architecture?

no idea. but I remember Vega being present in igpus for a very very long time

if you are talking specifically of handhelds, that should be the z3extreme. which I believe is the medusa premium (cpu = 4 + 8 + 2) (gpu = 24 CU / 12 wgp AT4)

 

[Tigerick]

Projected iGPU - Cyberpunk 2077 (1080p/Ultra/RT+Sun Shadows)

For those who do not understand how laughable if MDS1 is using RDNA3.5+ 8CU as iGPU, I created a stacked column chart listing the projected FPS based on data from DF. Each column comes with 20% faster FPS due to higher clock speed. PTL-H +20% is estimation of NVL-H with 12 Xe+. Even with +20%, MDS1 with RDNA3.5 8CU cannot even break 10 FPS. The reason is due to poor RT cores from RDNA3 generation. Do you really think AMD will bundle such old technology in APU which will be available from 2027 towards 2029. Hoho, use your brain for once, please.

I already estimate 3 times faster Radiance core in E2600's. By simple calculation, MDS1's RDNA4.5 iGPU's performance is comparable to PTL's B390. Clearly AMD is sandbagging the performance of iGPU. As I said, don't believe what the roadmap told you. The amount of CU is correct, just the RDNA generation is not. Juno IP is the next generation graphics IP from AMD which should be similar to RDNA4.5 which itself is subset of RDNA5. That's my assumption, let's see how truth it will be.

 

[marees]

Do you really think AMD will bundle such old technology in APU which will be available from 2027 towards 2029. Hoho, use your brain for once, please.

sir just one example

Vega GPU was released 2017
AMD was launching products with that igpu even after 5 years

AMD Ryzen 7 5700G Processor with Radeon Graphics Mini Review​

Posted by Sebastian Peak | Jun 28, 2022 |

https://pcper.com/2022/06/amd-ryzen-7-5700g-processor-with-radeon-graphics-review/

Now do this calculation:
RDNA 3 launched 2023
expect igpus for 5 more years until 2028

from 2029 onwards RDNA 5 should start taking over (except for embedded systems)

 

[ToTTenTranz]

if you are talking specifically of handhelds, that should be the z3extreme. which I believe is the medusa premium (cpu = 4 + 8 + 2) (gpu = 24 CU / 12 wgp AT4)

I honestly hope AMD just kills the Z line completely because it was a terrible idea to start with.

Selling a slightly less expensive APU that gets no driver support from AMD just resulted in a ton of frustration to everyone involved, especially the end customers.
Imagine selling a product with terrible driver support to the customers who are the most dependent on frequently updated drivers. What a genius idea.

 

[fastandfurious6]

AMD must really be missing good product people who work across domains

Looks like silos, "nobody thought of it" because there isn't anyone to think about it