AMD "Renoir" APU iGPU Configuration and Platform Spread Detailed

[INSTG8R]

The Surface Laptop 3 comes with either a 3580U or a 3780U. These are, for all intents and purposes identical to the 3500U and 3800U, based of April 2018's Zen+ launch. There's one more Vega CU in each of them, but neither model can use them because 15W and DDR4 2400 are the bottleneck here, not the CU count. It's Samsung's LP 12nm process and Zen+ architecture and DDR4 speeds with old, 100% Vega IGP architecture.

Given that the Surface Laptop 3 is only pre-ordering still (some lucky few might get theirs before Christmas), it's going to be something that's two years out of date by the time it's available for mainstream buyers. That's pretty awful for something that a lot of people will replace every 2-3 years.

Damn foiled my plans completely as a launch pad for their new SkUs I wonder if Intels mobile supply chain is also facing any kind of shortages where AMD could swoop in with something new and powerful to entice the usual suspects who‘ve barely looked at AMD since my Turion...

 

[Chrispy_]

Damn foiled my plans completely as a launch pad for their new SkUs I wonder if Intels mobile supply chain is also facing any kind of shortages where AMD could swoop in with something new and powerful to entice the usual suspects who‘ve barely looked at AMD since my Turion...

I don't know. Getting hold of a 10th Gen Intel seems to be hard work over here in the UK. A lot of listings that claim to be 10th-gen turn out to be 8th-gen, pre-orders, or available in 3-4 weeks.
Maybe it's because 10th Gen is too new to have propagated through the distribution channels still, maybe it's because there are genuine supply and yield issues that Intel isn't talking about.

I am cynical of Intel because they've been caught lying and with their pants down so many times now when it comes to 10nm, but I'm actually not sure about this. There are enough 10nm products out with reviewers and youtubers that it may be approaching a real launch to the public, rather than an ultra-low-volume marketing stunt to keep the shareholders and investors duped for another year.

 

[TheinsanegamerN]

How many LPDDR4X modules are out there in laptops ? How many laptops have more than 2400Mhz rated DDR4 modules ? Can you see the problem there ? It's a waste of effort and time for AMD to fit different memory controllers and change their architecture for a fringe memory type and speeds that aren't widespread. The only reason Intel did it is because it coincided with their newly developed architecture, 1065G7 is nothing short of a PR product showing supposedly that they are better than AMD. Go and tell me how many laptops with 1065G7 in them them that have 3733mhz rated LPDDR4X are there, I am waiting, then come back with a better theory.

LPDDR4X and higher speed DDR4 have no future with DDR5 around the corner which will make it worthwhile to change their APUs. Also, no, the memory performance isn't there, even with something like 3733 LPDDR4X these things are still memory bandwidth starved.

DDR5 is years away still. By the time it comes out, you'll be saying "well why would they support DDR5, DDR6 is coming soon!". Right now, LPDDR4X is the best mobile memory we have right now. If AMD doesnt want to develop a new memory controller for it, fine, but only supporting 2400 MHz DDR4? You yourself admit that even at 3733 mhz these things are memory starved, so the performance difference between 2400 and 3733 is going to be significant. Why leave all that performance on the table? Hell, why bother even makig the vega 11 iGPU if it is this memory starved? It is pointless to make such a big GPU if you cant feed it.

I cringe reading this, on the other hand it seems that you take this too seriously. I explained why, it's not worth it, as with any other company they care only about what's going to yield results. It is curious though that you bring this up, are you aware that there have been practically zero innovations in the last decade or so with anything that was GPU and mobile related from Intel ? When AMD trashed them in this area in terms of performance Intel was suddenly able to come back with a newly developed architecture with LPDDR4X and all that. If Intel was so concerned with the well being of their mobile users why have they sold them gloried display adapters all this time ? It seems like AMD raising the bar tremendously was just them not caring about mobile. These are some interesting theories indeed.

Lack of innovation from intel does not justify AMD kneecapping their own APUs with pathetically slow memory. Ryzen 3000 can support 3200 mhz memory, and the 2000 series (which this APU is built on) support 2933 on desktop. There is no excuse for the mobile chip to be hamstrung with 2400 mhz memory. Even ryzen 1000 supported 2666 FFS! The vast majority of intel laptops out there support 2666mhz memory at a bare minimum, and more recent models are pushing closer to 3000 as standard. I understand if you only ever look at AMD laptops you'd be confused, but the rest of the world has embraced high speed memory.

Too costly and complicated for now, if AMD ever gets some considerable market share I suspect they will pull this off eventually.

Always waiting. Always next year. Always the next generation. At some point AMD needs to wake up and release a product. They already have higher speed DDR4 controllers for the very arch used in their current APUs. They should be supporting 2933 just like their desktop bretheren.

By the time "eventually" hits Intel will have their 7nm Core I series out and AMD will have lost the process node advantage. If AMD just keeps waiting, intel will catch up again (just like with athlon 64) and AMD will become a permanent also-ran for another decade. Again. The mere existance of the G7 shows intel isnt just sitting idly, they are working on an answer for AMD, and AMD should be doing everything they can to make thir mobile parts attractive to buyers, since power usage is out the window AMD should be exploiting their performance advantage, that means supporting memroy speeds from 2019, not 2015.

As I wrote before, AMD should have released the 3000 series APUs only once 7nm was ready, with rDNA chips. They would have been more competitive then these re-heated 2000 series parts are, and slotted intel as second fiddle even with 10nm G7. They will never get considerable market share if this is how they treat said market.

In a laptop the power saved is negligible you can't really compare the two.

Ahem.

The Microsoft Surface Laptop 3 Showdown: AMD's Ryzen Picasso vs. Intel's Ice Lake

www.anandtech.com

Are you telling me those power numbers are solely due to intel's troubled 10nm arch? Reminder, MS worked with AMD to make the AMD surface, so if this is the best AMD can do, either AMD is completely incompetent or LPDDR4X has some impact, especially on idle power use where LPDDR4X pulls insignificant power compared to DDR4.

 

[INSTG8R]

I don't know. Getting hold of a 10th Gen Intel seems to be hard work over here in the UK. A lot of listings that claim to be 10th-gen turn out to be 8th-gen, pre-orders, or available in 3-4 weeks.
Maybe it's because 10th Gen is too new to have propagated through the distribution channels still, maybe it's because there are genuine supply and yield issues that Intel isn't talking about.

I am cynical of Intel because they've been caught lying and with their pants down so many times now when it comes to 10nm, but I'm actually not sure about this. There are enough 10nm products out with reviewers and youtubers that it may be approaching a real launch to the public, rather than an ultra-low-volume marketing stunt to keep the shareholders and investors duped for another year.

Well I absolutely agree and they’ve had a virtual strangle hold on the laptop space for longer than I can remembe. AMD need an “in” to reestablish themselves as a reliable option with theses new chips .

 

[Chrispy_]

Ahem.

The Microsoft Surface Laptop 3 Showdown: AMD's Ryzen Picasso vs. Intel's Ice Lake

www.anandtech.com

Are you telling me those power numbers are solely due to intel's troubled 10nm arch? Reminder, MS worked with AMD to make the AMD surface, so if this is the best AMD can do, either AMD is completely incompetent or LPDDR4X has some impact, especially on idle power use where LPDDR4X pulls insignificant power compared to DDR4.

I know you weren't reply to me, but I can't make any sense of that power use graph from your link, actually:

Clearly the Intel Surface Laptop beats the AMD in battery life, but the grey and red lines at the bottom of this graph are power usage in Watts. AMD seems to have both lower idle and peak power use compared to Intel, and the memory power usage is covered by the CPU power consumption, since the memory controller is on-die for both AMD and Intel.

Either:

• those measurements are wrong.
• those measurements for AMD are core-only and don't include the SoC power use.
• those measurements are right, and the Intel CPU is juicing hard but has a bigger battery.
• those measurements are right, and something else in the machine is chewing up the battery.

 

[Vya Domus]

By the time it comes out, you'll be saying "well why would they support DDR5, DDR6 is coming soon!".

This is by far the most unintelligent and basic thing you could have said, I am not even going to bother if this is the best you can do.

Lack of innovation from intel does not justify AMD kneecapping their own APUs with pathetically slow memory. Ryzen 3000 can support 3200 mhz memory, and the 2000 series (which this APU is built on) support 2933 on desktop. There is no excuse for the mobile chip to be hamstrung with 2400 mhz memory. Even ryzen 1000 supported 2666 FFS! The vast majority of intel laptops out there support 2666mhz memory at a bare minimum, and more recent models are pushing closer to 3000 as standard. I understand if you only ever look at AMD laptops you'd be confused, but the rest of the world has embraced high speed memory.

What Intel or AMD supports on paper is meaningless, what matters is what actually ends up being used. 2400Mhz DDR4 is vastly more common than any other memory type or speed, that's a fact. And it will probably remain like that until DDR5 gets here. Hell, most laptops are even sold with single-channel memory, the only thing pathetic here is your suggestion that everyone should support the most exotic and highest end memory type when the actual landscape paints a picture that's much more grim.

Yes, AMD doesn't support something that is found in less than 1% of laptops out there because they've concluded there's no point in doing so with their current situation and market share and no one can blame them. You know what will change if AMD starts shipping APUs that support faster memory ? Nothing, all manufactures will still happy fit a a cheap ass single channel 2400Mhz module in there because they care far less about this than you do.

Intel will have their 7nm Core I series out and AMD will have lost the process node advantage.

Yeah because TSMC will just disappear or something. Intel's leading process technology keeps falling behind for a couple of consecutive years now and you're sitting here telling me they'll somehow have the upper hand. Sure thing buddy, I have to say this blind belief in team blue is staggering.

Are you telling me those power numbers are solely due to intel's troubled 10nm arch? Reminder, MS worked with AMD to make the AMD surface, so if this is the best AMD can do, either AMD is completely incompetent or LPDDR4X has some impact, especially on idle power use where LPDDR4X pulls insignificant power compared to DDR4.

A laughable series of assumption you make there mate. Let's list them out :

MS worked with AMD but Intel ... didn't ? They just sort of like put an Intel chip in there without them knowing or what ? What makes you believe AMD received some sort of special treatment ?

LPDDR4X has some impact but can you prove it's a major one ? Do you just ignore the fact these are two completely different chips from different manufactures on two different nodes by the way.

Yeah, it must be that AMD is incompetent, it couldn't possibly be the fault of something else like Microsoft for instance, the ones that actually designed and put together the bloody thing.

You know, instead of wasting your time like this pulling out arguments from thin air you can just reply with "AMD bad Intel good", we'll get the message.

 

[r9]

I really hope AMD doesn't change the iGPU naming scheme again. Current Vega 3/8/11 is short and concise. If it's a mix of Navi and Vega like the leak claims just name it Vevi or better yet Naga.

I swear AMD is the fucking worst when it comes to naming their products. They ditched Radeon HD 8000, 9000 and went straight to Rn 200. Then after RX 500, they leaped back to RX 5000. Jesus fuck! And just forget about whatever they named their mobile and desktop APUs in the meantime. People make fun of Intel's 10k series naming scheme but at least they and Nvidia to some extent follow the numbering scheme.

Now tell us how you really feel, and this time don't hold anything back.

 

[Alpha_Lyrae]

I bet they simply can't find a solution for multi-screen power draw.

Look at RX5500 review on TPU:

AMD Radeon RX 5500 Review

The RX 5500 Navi is AMD's ambitious attempt to disrupt the sub-$200 graphics card market just as the RX 5700 series did to the sub-$400 market, bringing in much needed competition. It won't be easy as NVIDIA tried to preempt it with the GTX 1650 Super and GTX 1660 Super.

www.techpowerup.com

Vega idle power draw is very high, but multi-monitor is just +2-3W.
Radeon VII is exactly the same.
However, RX5500 gets +14W (4->18).

Even if we assume there's a 3W measurement error, Navi figures are absurd (even if it becomes something like 2W->9W in a smaller mobile SoC).

So AMD has to stay with Vega for most important segment: mobile <=25W SoCs (and likely the desktop 35W as well).

I guess dividing the lineup and making larger SoCs on Navi didn't make much sense.

Idle power figures for Vega rise substantially when using a single 120Hz 4K panel or even mixed 4K120 and 4K60 panels. It's about 9W, from 3W at 4K60 (chip only; actual card use is higher), and HBM idle clock bumps up to 500MHz. GPU stays at 26MHz, while SoC domain ventures up to 600MHz when data is transferred over SoC domain (ranges from 33-600MHz).

High refresh is still difficult for display controllers simply due to the amount of time the controller is on submitting frames to a panel refreshing at 120Hz. It doesn't really have a chance to idle. If aggressive power saving was employed with high refresh, you'd likely encounter sync issues and intermittent loss of signal. Bandwidth of DP1.4 cable is maxed at 4K120 using RGB 4:4:4: 32.4Gbps. DSC must be used to achieve higher than 8bpc.

AMD's display controllers are connected via Infinity Fabric in Vega and Navi via GPU SoC circuit. In Vega, display and media ASICs can access memory independent of GPU via IF/SoC domain. Navi might be similar, but to counter increased latency of GDDR6 (vs HBM2 in Vega), it may need to use maximum memory clocks, which increases power draw at idle.

 

[FinneousPJ]

Ryzen 7 APU? So that's at least 8 cores yeah? Fuck me that's amazing.

 

[notb]

Idle power figures for Vega rise substantially when using a single 120Hz 4K panel or even mixed 4K120 and 4K60 panels. It's about 9W, from 3W at 4K60 (chip only; actual card use is higher), and HBM idle clock bumps up to 500MHz. GPU stays at 26MHz, while SoC domain ventures up to 600MHz when data is transferred over SoC domain (ranges from 33-600MHz).

You're talking about a very niche use case - even more so for basic GPUs found in SoCs.

Dual 1080p monitors or notebook screen + monitor is how many laptops are used on daily basis (and TPU's test is fairly close to that).
It's just weird that a company would launch a SoC that has problems with such an important scenario.

Ryzen 7 APU? So that's at least 8 cores yeah? Fuck me that's amazing.

Dialed down and without SMT.
4700U benchmarks have leaked already - it's slower than existing 4C/8T Ice Lake U.

 

[R0H1T]

Dialed down and without SMT.
4700U benchmarks have leaked already - it's slower than existing 4C/8T Ice Lake U.

Shock horror - a pre-production likely QS or ES chip is slower than final ICL Si - meanwhile, in other news the same ICL ships are beaten in many benches by top of the line WHL or CML chips.

 

[Darmok N Jalad]

Ryzen 7 APU? So that's at least 8 cores yeah? Fuck me that's amazing.

Not necessarily. The current 3780U is a 4C/8T part.

 

[notb]

Shock horror - a pre-production likely QS or ES chip is slower than final ICL Si

Intel will launch Tiger Lake next year and that's what Zen2 APUs will have to compete with.
For the time being lets treat Ice Lake as pre-production TL.

And what do we know about TL? Expected to be 20%+ faster in CPU component and maybe 50% faster GPU (purely based on CU count).
Do you think there will be such a large gap between APU's ES and production?

Moreover, mobile Tiger Lake will have PCIe4.0 and all the goodies ICL already added.

- meanwhile, in other news the same ICL ships are beaten in many benches by top of the line WHL or CML chips.

Could be true. Why wouldn't they? ICL is not a successor of CML. They are 2 architectures on different nodes existing side by side.
And, clearly, both lineups are being improved. They're both quite polished and fast. I'm not sure why you're trying to turn this into Intel's failure.

Not necessarily. The current 3780U is a 4C/8T part.

3780U is still Zen+. 4000-series will be Zen2.

 

[Adam Krazispeed]

No Navi yet? VCN 1.0 would be perfect for a potential 4400G, especially in a HTPC for HEVC/H.265 decoding/transcoding.

BIG DISAPPOINTMENT... IM NOT BUYING An APU TILL APUS GET NAVI <WTF AMD WTF!!!!!

VEGA IS SHIT uARch, not worth my $$$$$$$$$$$$$$$$ F**K DAT

guess its not gonna be till 5 or 6000 series apus.. im just sticking to discrete GPUS and Desktop 8+ core CPus, screw this..

wish a could take a zen 2 apu and customly add my 5700xt die into it myself...

DDR5 is years away still. By the time it comes out, you'll be saying "well why would they support DDR5, DDR6 is coming soon!". Right now, LPDDR4X is the best mobile memory we have right now. If AMD doesnt want to develop a new memory controller for it, fine, but only supporting 2400 MHz DDR4? You yourself admit that even at 3733 mhz these things are memory starved, so the performance difference between 2400 and 3733 is going to be significant. Why leave all that performance on the table? Hell, why bother even makig the vega 11 iGPU if it is this memory starved? It is pointless to make such a big GPU if you cant feed it.

Lack of innovation from intel does not justify AMD kneecapping their own APUs with pathetically slow memory. Ryzen 3000 can support 3200 mhz memory, and the 2000 series (which this APU is built on) support 2933 on desktop. There is no excuse for the mobile chip to be hamstrung with 2400 mhz memory. Even ryzen 1000 supported 2666 FFS! The vast majority of intel laptops out there support 2666mhz memory at a bare minimum, and more recent models are pushing closer to 3000 as standard. I understand if you only ever look at AMD laptops you'd be confused, but the rest of the world has embraced high speed memory.

Always waiting. Always next year. Always the next generation. At some point AMD needs to wake up and release a product. They already have higher speed DDR4 controllers for the very arch used in their current APUs. They should be supporting 2933 just like their desktop bretheren.

By the time "eventually" hits Intel will have their 7nm Core I series out and AMD will have lost the process node advantage. If AMD just keeps waiting, intel will catch up again (just like with athlon 64) and AMD will become a permanent also-ran for another decade. Again. The mere existance of the G7 shows intel isnt just sitting idly, they are working on an answer for AMD, and AMD should be doing everything they can to make thir mobile parts attractive to buyers, since power usage is out the window AMD should be exploiting their performance advantage, that means supporting memroy speeds from 2019, not 2015.

As I wrote before, AMD should have released the 3000 series APUs only once 7nm was ready, with rDNA chips. They would have been more competitive then these re-heated 2000 series parts are, and slotted intel as second fiddle even with 10nm G7. They will never get considerable market share if this is how they treat said market.


Ahem.

The Microsoft Surface Laptop 3 Showdown: AMD's Ryzen Picasso vs. Intel's Ice Lake

www.anandtech.com

Are you telling me those power numbers are solely due to intel's troubled 10nm arch? Reminder, MS worked with AMD to make the AMD surface, so if this is the best AMD can do, either AMD is completely incompetent or LPDDR4X has some impact, especially on idle power use where LPDDR4X pulls insignificant power compared to DDR4.

for one intels icl is 10nm, and amds 3780u is 12nm zen+ this is not a fair comparison by anymeans, they needed zen2 4700u in the surface 3 to actually campair intel/amd cpus sncenc TSMCs 7nm and intels 10nm are kind of equal ir Similar in efficiency..