AMD "Renoir" APU to Support LPDDR4X Memory and New Display Engine

[XiGMAKiD]

Integrated praphics should not be meant for gaming. The whole idea, that it should be good for low end gaming is really what is stopping AMD from getting additional money.

Intel CPUs are basically the No. 1 choice when it comes to office PC, home servers, work servers, worstations.
iGPU is only needed to show the view, and no workload is needed. We are buying things like 8700k, 9900k not because they are intel, but because they are powerful and have iGPU.
I wish AMD added iGPU into their high end models - NOT FOR GAMING.

While AMD is doing a good job on providing 4 core APU, it is however, based previous generation, which is not acceptable in my opinion.

I don't think we'll see 6-8 cores AMD with iGPU in near future

 

[CityCultivator]

My experience with STAPM and a couple of 2000-series APUs is that AMD are so nearly there with their current-gen APUs. Battery life suffers compared to Intel because AMD's 1st-gen Zen/Zen+ CPU cores seem to eat up power at lower clockspeeds and don't clock down as hard as Intel's U-series do when idle.

I actually think that getting the CPU portion of the power draw down is more important that the Vega IGP, so if Renoir APUs are 7nm Zen2 cores and 12nm IGP and IO, they'll be fine.

Ideally, the whole thing will be at 7nm but with the 2700U I bought myself in February, and with some STAPM power-budget tweaking thanks to modified BIOSes and RyzenAdj, I can say for certain that the 4C/8T of Raven Ridge load can comfortably use 15-18W and will never consume less than 5W in a normal system. The Vega10 IGP will add maybe 8-9W at full load and peak clocks but rarely get to run at those speeds because most laptops OEMs set a STAPM limit of somewhere between 15-25W (20W in my case). That means that before long, the CPU is using the lion's share of the power budget and the Vega cores are throttled back to 25% of their ideal clocks, despite being the more important part of the equation when the IGP is active in a 3D or GPGPU compute application.

At 15-18W potential peak CPU core usage the IGP really suffers in a Ryzen APU, and most people will be buying Mobile Ryzen for the Vega cores, otherwise they'd just get an Intel with worse graphics but better battery life. In the case of the 15W ultrabooks, the IGP is throttled down to pointless speeds. In the case of 20W and 25W models, the IGP is running sub-optimally if the CPU is busy. TSMC's 7nm seems to have huge efficiency gains so even if ONLY the CPU cores in Renoir were moved to 7nm, that 15-18W CPU peak draw could drop down to Maybe 10-12W, leaving far more headroom for the IGP to do their thing in the ideal 15W power budget.

My preference would be for AMD to better balance their APU power budget in hardware so that the Vega cores get first dibs on whatever power is available. It's actually self-balancing, because if the CPU clocks drop too far from IGP greed, the IGP will stall and free up power budget for the CPU. The current CPU-first implementation doesn't really work. The CPU clocks up and overfeeds the IGP which then rejects frames because it can't process them fast enough. It's a stupid waste

I also have a 2700u. I used ryzenadj to keep the system always in fast mode (by setting the slow and stapm time to forever). My stapm limit is a measly 15W. Without ryzenadj the gpu would be useless. On sustained loads, the cooling can dissipate about 25W (with ryzenadj).

 

[Chrispy_]

I also have a 2700u. I used ryzenadj to keep the system always in fast mode (by setting the slow and stapm time to forever). My stapm limit is a measly 15W. Without ryzenadj the gpu would be useless. On sustained loads, the cooling can dissipate about 25W (with ryzenadj).

It's annoying isn't it. We bought our laptops for the Vega10 IGP yet default AMD behaviour cripples them to keep the CPU cores running as fast as possible.

Raven Ridge's CPU portion has no advantages over the equivalent i7 (slower, hungrier, lower IPC) and it sucks that AMD doesn't provide software to control the APUs properly. We rely on third parties for that.

One thing I found helpful for gaming was to run the game with processor affinity limited to two consecutive threads (one physical core with SMT). Assuming the game runs on two threads, it's usually an improvement in IGP performance. Just create a shortcut to your game with the following:

cmd.exe /c start "Program Name" /affinity 3 "Full path of application file"

The 3 is hexdecimal for 11, implying core0 and core1 are active.

affinity 1 = 1C/1T
affinity 3 = 1C/2T
affinity 5 = 2C/2T
affinity 1E = 2C/4T

I don't think there's any point going higher than that because the IGP will always be the gaming bottleneck even with only 2C/4T active.

 

[CityCultivator]

I stopped playing on that machine, I do still use it for madvr playing.
There is less CPU usage then (especially when hardware acceleration is present).

 

[bug]

@Chrispy_ 3 in binary is 11.
Hexadecimal 3 is... 3.

 

[Chrispy_]

You need to use the hex code of the binary value. Which reminds me, I derped up; 2c4t is 1E, not 15!

 

[bug]

You need to use the hex code of the binary value. Which reminds me, I derped up; 2c4t is 1E, not 15!

So you meant 3 is hex for binary 11. Gotcha.

 

[Chrispy_]

So you meant 3 is hex for binary 11. Gotcha.

Yep. Brainfarts are my specialty.

 

[bug]

Yep. Brainfarts are my specialty.

No sweat I routinely go back and edit my posts for things like these. Or typos. And I'm pretty sure I'm not catching them all

 

[danbert2000]

Here I was all excited for Ryzen 3 APUs because we could finally get decent IPC on the CPU side and a Navi iGPU, but apparently that makes too much sense so we must wait for two whole cycles of APUs to get the best AMD can make. Big miss for AMD, and you have to wonder if:

• AMD can make Navi work with an APU, but they don't want to provide a GPU solution in the APU that makes casual gamers just skip a discrete AMD GPU when building their computer.
• AMD developed Navi with Sony and Microsoft money and there's some provision that AMD can't use Navi in APUs until after console release so as not to overshadow the custom APUs.
• AMD is out of time and money to make Navi work with the APUs that aren't really their profit center right now.
• The GPU part is going to be integrated into the IO chip and Navi was never validated on the larger lithography.

Whatever the case, AMD will be looking pretty weak on APU performance if Intel is able to push out Gen 11 integrated graphics on their chips anytime soon. Hopefully they figured out some way to up the number of compute units to at least beat the existing chips in graphics performance.

 

[bug]

Here I was all excited for Ryzen 3 APUs because we could finally get decent IPC on the CPU side and a Navi iGPU, but apparently that makes too much sense so we must wait for two whole cycles of APUs to get the best AMD can make.

Eh, by the time they make a Navi APU, their best will probably be Arcturus already. The only problem I have with that is newer GPU arch tends to be more power efficient and that's a big win for APUs.

 

[Brusfantomet]

Here I was all excited for Ryzen 3 APUs because we could finally get decent IPC on the CPU side and a Navi iGPU, but apparently that makes too much sense so we must wait for two whole cycles of APUs to get the best AMD can make. Big miss for AMD, and you have to wonder if:

• AMD can make Navi work with an APU, but they don't want to provide a GPU solution in the APU that makes casual gamers just skip a discrete AMD GPU when building their computer.
• AMD developed Navi with Sony and Microsoft money and there's some provision that AMD can't use Navi in APUs until after console release so as not to overshadow the custom APUs.
• AMD is out of time and money to make Navi work with the APUs that aren't really their profit center right now.
• The GPU part is going to be integrated into the IO chip and Navi was never validated on the larger lithography.

Whatever the case, AMD will be looking pretty weak on APU performance if Intel is able to push out Gen 11 integrated graphics on their chips anytime soon. Hopefully they figured out some way to up the number of compute units to at least beat the existing chips in graphics performance.

Well, AMD is capable of making a bigger GPU for an APU solution, just look at the Intel i7 8809g.
Note that the Vega chip in those implementations has its own dedicated HBM2 framebuffer, removing the need to share bandwith with the CPU.

If you compare the Vega 8 in the 2200g with the Vega 11 in the 2400g, there is a 50% increase in raw power, but the gaming performance is really close. Shure, "Renoir" has potentially higher ram speeds (4266 MHz compared to 3200 MHz) but that is peanuts (68 GB/s and 51 GB/s) compared to the dedicated bandwidth of the 8809g at over 200 GB/s.

In short; memory interface speed is holding AMD APUs back, and nothing short of a dedicated HMB2 chip for GPU memory on package will solve that.

 

[Chrispy_]

Here I was all excited for Ryzen 3 APUs because we could finally get decent IPC on the CPU side and a Navi iGPU, but apparently that makes too much sense so we must wait for two whole cycles of APUs to get the best AMD can make. Big miss for AMD, and you have to wonder if:

• AMD can make Navi work with an APU, but they don't want to provide a GPU solution in the APU that makes casual gamers just skip a discrete AMD GPU when building their computer.
• AMD developed Navi with Sony and Microsoft money and there's some provision that AMD can't use Navi in APUs until after console release so as not to overshadow the custom APUs.
• AMD is out of time and money to make Navi work with the APUs that aren't really their profit center right now.
• The GPU part is going to be integrated into the IO chip and Navi was never validated on the larger lithography.

Whatever the case, AMD will be looking pretty weak on APU performance if Intel is able to push out Gen 11 integrated graphics on their chips anytime soon. Hopefully they figured out some way to up the number of compute units to at least beat the existing chips in graphics performance.

I think it's simpler than that. Demand for Zen2 processors FAR outstrips supply. I've been waiting on orders for 6+ weeks for any kind of bulk purchases and I'm looking outside of the UK to EU vendors too.

TSMC's 7nm node is busy. Everyone wants a piece, not just AMD. So AMD have limited 7nm capacity available to them and all of that is going on Zen chiplets for Epyc in the case of the most efficient yields (where the profit margin is 10x what it is in the consumer space) and the leakier, higher-voltage chips become consumer Ryzen.

This is why the only Navi parts coming out right now are going into the high-end market. There's not enough spare 7nm production capacity to justify making low-end consumer parts with tiny profit margins, so the only APUs we're likely to see in the immediate future are probably going to be 8C/16T Zen2 chiplets that are super-defective but can be salvaged as 4C/8T. All AMD have to do to make Renoir from those is throw them onto a package with existing Vega and IO controller made on the cheap and unconstrained GloFo 14nm process.

 

[IceShroom]

Well, AMD is capable of making a bigger GPU for an APU solution, just look at the Intel i7 8809g.
Note that the Vega chip in those implementations has its own dedicated HBM2 framebuffer, removing the need to share bandwith with the CPU.

If you compare the Vega 8 in the 2200g with the Vega 11 in the 2400g, there is a 50% increase in raw power, but the gaming performance is really close. Shure, "Renoir" has potentially higher ram speeds (4266 MHz compared to 3200 MHz) but that is peanuts (68 GB/s and 51 GB/s) compared to the dedicated bandwidth of the 8809g at over 200 GB/s.

In short; memory interface speed is holding AMD APUs back, and nothing short of a dedicated HMB2 chip for GPU memory on package will solve that.

Perfect example of APU is APU's used in console like Xbox 1/PS4. Another is Subor Z+.

 

[Brusfantomet]

Perfect example of APU is APU's used in console like Xbox 1/PS4. Another is Subor Z+.

Yes, but both the Xbox and PS4 are running a scaled back OS compared to a standard PC.
Additionally, both of them have memory in the same ballpark as the HBM2 chip in the 8809g for the GPUs (Xbox one x 326 GB/s and PS4 is 176 GB/s)
The Subor Z+ sits at 256 GB/s according to this.

I still think it’s a question of memory bandwidth for the consumer APUs from AMD.