AMD Releases Ryzen Balanced Power Plan - Test Results Inside

Subject: Processors | April 6, 2017 - 06:03 PM |
Tagged: ryzen, Power Plan, Power Management, Balanced, amd

AMD Releases Ryzen Balanced Power Plan - Test Results Inside

AMD has published Community Update #3 to their blog. This update details a new Power Plan that should yield improved gaming performance for those who were previously using the Windows default Balanced Power Profile. There has been lots of speculation on reasons for performance differences when gaming in various power modes and even on different Operating Systems. With this new Ryzen Balanced profile also came some info that should help us clear up some of the other misconceptions out there.

After we determined that the Windows 10 Scheduler was not at fault for the Ryzen performance issues we were seeing in some applications, we received some testing feedback from those who had noted performance differences between Windows 7 and Windows 10. While many believed that to be confirmation of scheduler differences between both Operating Systems, the actual cause was down to how Windows 7 and Windows 10 park their cores, as demonstrated by the points AMD sent us earlier today:

  • Windows 7 only parks SMT cores, keeping all physical cores awake.
  • Windows 10 keeps the first core awake (logical core 0 + 1 on a HT system) and parks the remainder when possible.
  • Windows 10 disables core parking by default on Intel CPUs (Speed Shift support).

Since Windows power management (not the scheduler) is not yet Ryzen aware, its default settings result in overly aggressive core parking when driving a Ryzen CPU. Until a lower level change can take place, AMD has released a custom Ryzen Balanced Power Plan that tweaks some of the P-state transition values and a few other settings to help realize the performance gains previously seen by folks shifting to the High Performance mode while keeping idle power consumption much closer to that of the Balanced plan. Here are AMD’s claimed performance gains (vs. Balanced) with their new Ryzen Balanced Power Plan:

View Full Size

AMD provided claimed gains for Ryzen Balanced profile vs. default Windows Balanced profile.

Realize these gains are all going to be nearly identical to any prior comparison showing Balanced vs. High Performance profile deltas, but this profile retains most of the idle power savings accomplished by the Balanced plan. We’ve been doing some testing with the tool and can partially confirm the above results, while adding in some more of our own that were not included in AMD’s data:

View Full Size

The blue highlighted bars denote the overlapping titles tested. A few other titles we tested showed lesser (or no) gains, but that’s not necessarily the fault of this new profile as those same titles saw similar results with a switch to High Performance mode when tested previously.

I did a bit of digging into exactly which power profile parameters are being tweaked and how. Laymen poking around in Windows Power Management will only find this single difference:

View Full Size

However, deconstructing the actual profile data reveals more changes that do not appear in the Windows GUI. Here are the low-level changes we discovered, including the ‘Minimum processor state’ previously noted above:

View Full Size

Note: Units differ varying by parameter in this chart - compare within each set of 3 bars.

As you can see, changes were made to help minimize the parking of Ryzen cores, and to also speed up their waking when required. It may not be a perfect solution as it is another step that the user must perform to get good ‘out of the box’ Ryzen performance, but it does help alleviate the dilemma of running your desktop machine at full tilt 24/7 or having to switch power modes on either end of your gaming sessions. This is a solid stop-gap until native Ryzen support makes its way into Windows, so all of you Ryzen users out there, run over to the AMD Blog and grab/install the Ryzen Balanced Power Plan!

Source: AMD

April 6, 2017 | 06:41 PM - Posted by Mr.Gold (not verified)

I have the default balanced profile with my 1800x, all stock.

Scrolling in chrome (task manager in front)
I see that only CPU 0 to 3 are used, the rest is parked.
When I type CPU 0 1 are used, the rest parked.

So typing this my CPU is running at 2.09GHZ on 2 core, all the rest parked.

I opened HWInfo and it shows a CPU + SoC power of 9.7w
with a core voltage of 0.769v and SoC voltage of .944v
(Ram is at 2666mhz.. considering its DDR4 with an XMP of 2400mhz, I'm not complaining)

I will try the AMD tweaked profile, but I'm fine with the performance/power of the balanced profile.

side note: This chip overclock, but not past 4ghz.
Since I reach 3.7ghz in all 16 thread heavy apps, and 4.1ghz on light threaded... a 10% boost at best and a 3% slowdown otherwise, I elected not to overclock.

But I do on my R7 1700 system (but stayed at 3.7ghz with the stock cooler). And its amazingly near silent IBT
Paired this one with a Strix Rx 480 and an EVGA G3. Both turn off their fan during desktop use.
This R7 1700 machine is the quietest build I ever made.
(in a crystal 460x, with fan profile tweaked down)

I know, "cool story bro"... but I'm pretty happy with those 2 builds.

April 7, 2017 | 01:14 PM - Posted by John H (not verified)

cool story bro ;-)

Actually i'm glad there's a newer profile here. I have a Core i5 on a win 2012 r2 server that really suffers on lighter threaded workloads (1-2) because it won't turbo without all 4 cores being hit heavily.. (when in balanced profile).

April 6, 2017 | 08:07 PM - Posted by Anonymous (not verified)

In other news: Scorpio will not use Ryzen so Ryzen optimizations will only happen on PC Ports for the PC Minuscule Race.

"AMD fans just need to keep the faith for a few months, and that soon Ryzen's full power will be revealed"

Maybe AMD can just throw money at Stardock and have them optimize all the PC games for them...

in other other news: Nvidia doesn't even bother with the hype train and releases a new titan. No competition not even worth pre-hype

April 6, 2017 | 10:23 PM - Posted by Anonymous (not verified)

I don't care in the slightest about new nvidia cards. There hasn't been an interesting video card release since the RX 480. Everything since then has been ridiculously expensive, low volume, marketing parts that I, and probably 99.99% of other consumers, have no interest in buying.

Anyway, you sound like a troll, but the Xbox one CPU is a two module, 8 core part, which probably requires similar, if not the same thread/CPU affinities to be set for the best performance. Scorpio will be a similar two module device.

April 6, 2017 | 10:42 PM - Posted by Anonymous (not verified)

Oh you have the interest, but no money.

April 7, 2017 | 01:26 PM - Posted by Anonymous (not verified)

I love my MSI RX480 Gaming X. I only got the 4GB, but I have yet to run into an issue with that. It was an insane improvement over my old Radeon HD 6950 2GB that was constantly running at 88 C. My 480 rarely breaks 65 C.

April 7, 2017 | 01:39 AM - Posted by Anonymous (not verified)

Ryzen is a consumer brand name for CPUs based on the Zen microarchitecture. Scorpio will be using some unknown Jaguar variant core design or maybe even some Zen/Light variant with the SMT hardware removed and make use of some custom on APU on die interconnect variant, no one really knows for sure at this time. AMD could make up a Zen core microarchitecture APU/Scorpio variant that has a custom CCX/8 little core configuration who knows. All the Ryzen desktop parts are based on the modular/scalable Zeppelin Die(Single Zeppelin die for Ryzen 7, 5, and 3 series SKUs) that is intended for server usage, but the Raven Ridge APUs are not going to be using any Zeppelin dies, so who can be sure.

It's pure speculation at this point in time as to what Scorpio's GPU CUs will be based on. Could it be Polaris or even Vega or some variant that is in-between. Maybe Scorpio will make use of primitive shaders like Vega. There is no way of telling what features that are in Vega that may have been intended for Polaris or even tested during the development of the Polaris GPU microarchitecture that may not have made in into the final Polaris designs and where instead pushed back to be included in the Vega GPU microarchitecture. So AMD can probably have some semi-custom CUs loosely based on Polaris that may have some features intended for Vega, it depends on how far back in time the design process was frozen and the tape-out was finished for the semi-custom APU in Scorpio.

AnandTech speculates(Jaguar, Excavator, or Zen) on what CPU options that AMD may have available for Scorpio, but I think that AMD was designing for the server market with its Zeppelin Die and Ryzen was more dependent on that Server development time-line owing to the fact that Ryzen(7, 5, 3) is directly based on that Modular/scalable Zeppelin die. So the Zen CPU core microarchitecture was completed shortly before Jim Keller left AMD and thus that Zen microarchitecture has been available for a longer time than maybe the Zeppelin dual core complex Die design intended for the server market may have had its design work completed. And the Zen microarchitecture is not dependent on the Infinity Fabric UN-core IP, as APUs may or may not need to make use of the Infinity Fabric IP as much in any semi-custom designs.

Ryzen arrived when it did because Ryzen was based on the Zepplelin Die and not because the Zen microarchitecture was not finished. And AMD was looking to get back into the server market first and formost, so the Zen microarchitecture may have been available much eariler than that Zeppelin Die had its design work completed and certified for the server market. The Zen microarchitecture has been available before September 18, 2015 and Jim Keller's last day at AMD.

April 7, 2017 | 04:32 AM - Posted by Kusanagi (not verified)

It was confirmed that Project Scorpio uses an upclocked Jaguar core.

April 7, 2017 | 12:09 PM - Posted by Anonymous (not verified)

Source please to confirm your statment.

April 7, 2017 | 02:45 PM - Posted by Vince (not verified)

See the Digital Foundry Youtube video detailing Project Scorpio with information officially provided by Microsoft--no rumors: Project Scorpio Exclusive: Final Specs Revealed!
Link: https://www.youtube.com/watch?v=RE2hNrq1Zxs

April 8, 2017 | 03:03 PM - Posted by Anonymous (not verified)

Yes is see that Anandtech has also updated their info, so here is a question. The ROP count on the RX 480 is 32, and the new XBOX Scorpio(?). What is very interesting is the DX12 implementation in the command processor(That has to be firmware up-datable maybe(?) and this is some specialized ASIC/controller/CPU(?). The Jaguar cores maybe they are more customized are they maybe getting more Floating Point resources per module(?). Scorpio's GPU is using 40 CUs and what about the Polaris Graphics in Scorpio maybe getting a different ROP ratios compared to the Polaris RX 480 GPU's 2304 SPs, 144 TMUs, 32 ROPs, and maybe that primitive shader technology for Scorpio's custom Polaris GPU build.

I hope that a good deal of this will be some more Info forthcoming. That DX12 built into the GPU's command processor. Maybe AMD can do something similar for all its GPU SKUs with the GPUs Command processor having a brain so to speak, and AMD has been throwing that brain word around for Vega, so I'd like to see maybe the ability to take any graphics API into the firmware of a command processor and accelerate things that way.

April 8, 2017 | 03:22 PM - Posted by Anonymous (not verified)

Oh yes that TSMC 16nm FF process so Scorpio can be clocked higher and is a little more mature 16nm FF process. So M$ chose TSMC over GF, and I'll bet that this has nothing to do with AMD's GF contract where AMD has to pay GF for any chips fabricated at any another foundry. This is M$'s project and M$ is the customer while AMD is just a semi-custom subcontractor for Scorpio's APU design.

That DX 12 in the ommand processore needs to be matched in Vega with any Graphics API able to be loaded in the firmware of the command processor because that's a revolutionary improvment for gaming. Maybe Vega will have some ability to put the Vulkan API in the command processor and really improve latencty for gaming, because that a big thing for Scorpio and 4k.

April 8, 2017 | 03:25 PM - Posted by Anonymous (not verified)

Edit: That DX 12 in the ommand processore
To: That DX 12 in the command processor

April 7, 2017 | 08:03 AM - Posted by Anonymous (not verified)

Wow while you tried to make a point you only ended up with embarrassing yourself with your lack of knowledge... It's a good idea if you just stop posting.....

April 8, 2017 | 10:56 PM - Posted by Anonymous (not verified)

Really with the very same Zeppelin Die(2 CCX units die used in Ryzen) used to make the Naples(4 Zeppelin dies)/lower server variants, and you don't think that the optimizations for the Zen microarchitecture are going to continue and probably accelerate. And Ryzen(Consumer Branding) is based on the very same Zen microarchitecture that Naples will be using.

Don't you Know that the server/HPC/Workstation market will have AMD/server makers and the academic community tweaking Zen until the Zen 2 microarchitecture arrives, and even after that, to get every lettle bit of performance/Watt out of Zen for the professonal server market that measures and works constantly to reduce power usage for the big multi-megawatt eating server farms. And that pro markets R&D and tweaking reaches into the consumer market because Intel's server R&D finds its way down into the the consumer market the same way. Any workloads for Zen and the server market may just match up with the gaming workloads in the consumer market as the pro markets will be using AMD's and Nvidia's GPUs also.

How clueless and totally DAFT can most gamers be, even when presented with the facts for months on end.

Ryzen is not a microarchitecture, its a consumer brand name, Zen is the F-ing name for the microarchitecture.

P.S. don't make fun of Scorpio that APU is custom and It's GPU command processor has the DX12 API baked in, so that part was relatively unknown, and even the Xbox one's APU apparently also had that ability with the graphics API in the GPU's command processor for the APU in the XBox one.

That's some interesting technology for consoles that I hope find its way into the discrete GPU market and for consumer APUs. I'd love to have an APU with the Vulkan API, DX12, or Metal/other loaded in to the GPU's command processor as that will reduce the bandwidth on the bus and cut down on latency, and latency is a big factor in that FPS metric that gamers all fapp about all of the time.

April 7, 2017 | 08:41 AM - Posted by gamerk2 (not verified)

Stupid question:

Intel has a CPUID bit that informs the OS/Applications which processors contain HTT, so Microsoft and other OS vendors don't have to manually modify the scheduler every time Intel releases a new CPU architecture.

What's preventing AMD from doing the same, or even repurposing the same CPUID field? Wouldn't that solve all these issues that both Bulldozer and Ryzen are running into?

April 7, 2017 | 05:47 PM - Posted by serpico (not verified)

I think the problem is that even if the OS can read which CPU it is, it doesn't know what to do with the proessor for max performance. AMD didn't sit down with Microsoft beforehand and let them know exactly what to do when ZEN architecture is detected.

April 11, 2017 | 01:52 AM - Posted by Anonymous (not verified)

Because that process of AMD working with M$/Others is called the optimization process and AMD needs to get cracking on getting their optimization manuals into the hands of edvelopers. Intel has had years of optimization work for it's specific x86 micro-arch.

How can AMD sit down with M$ beforehand when Zen/Ryzen is so new to the market and AMD's is still issueing Micro-Code tweaks to get Zen/Ryzen and Zen/Naples SKUs up to speed and all the kinks worked out for Zen Version 1, and Zen Version 2 is on the way. Now that the first consumer Zen/Ryzen products are to market there will be the usual time lag before the optimizations are done in the OS/software and Games/Gaming engines ecosystems that target the Zen Micro-Arch. Rome wasn't built in a day.

April 7, 2017 | 11:33 AM - Posted by Subsailor

Good work Allyn!

April 7, 2017 | 11:45 AM - Posted by Anonymous (not verified)

Still buying a 1800x to replace my 4930K. This gives us even more performance. 1800x destroys my 4930K in rendering. Even when I OC to 4.3ghz. Very excited about this update to power profiles.

April 7, 2017 | 01:05 PM - Posted by Anonymous (not verified)

how can you call it "balanced" when the processor minimal frequency is 90% of max freq? the settings delta implies this is "high performance" re-branded

April 7, 2017 | 03:16 PM - Posted by Anonymous (not verified)

Cores shut down still and there is a power savings. Ryzen is still doing its own internal power savings. Ryzen is much more advanced then you are giving it credit.