DCS fans - Here's good news: AMD Boosts Single Thread Performance

I know single-threaded CPU performance is critical to DCS framerates, so this should make some people happy. :slight_smile:

Via Tom’s Hardware…

…AMD’s new Precision Boost Overdrive 2 [for Ryzen 5000 processors only] … boosts single-threaded performance while retaining the benefits of the existing multi-core boosts…

AMD Introduces Precision Boost Overdrive 2, Boosts Single Thread Performance | Tom’s Hardware


They seemed to “boost” Cinebench score from 628 to 638 … :rofl:


der8auer had a piece on that I think, or something similar. with dynamic switching between boost and oc mode.

But he thought it was only the z570 Dark Hero motherboard from Asus that supported it. But I guess that’s only until other people implement or bios update to support it.


For those who buy overclocked chips from the likes of siliconlottery.com , this is likely to be meaningless. Such chips are usually overclocked equally on all cores. I have not had success overclocking individual cores beyond that, and even if I did, it is highly unlikely the MSW task scheduler could properly take advantage of that for DCS World.

Interesting nonetheless, thanks @neal_white_iii .

If you are thinking that is not significant, do not be fooled. Even 1.5% is very substantial to flight sim right now, and still significant for VR generally.

1 Like

What a silly thing to say… :rofl:

1.5% is 1.5%


On a bad day, the performance margins in DCS World get that thin all the time. :expressionless:

Well. That still doesn’t make 1.5% significant… :wink:

1 Like

Significant. As in, observable. I would say pushing the sim over the edge to complete failure to maintain 30Hz, causing all sorts of horrible double framing, is rather ‘significant’.

So You’re saying 1.5% is all it takes?

In that case I would say way too little margin.

It doesn’t take a lot to make that margin disappear.

Again, pretty silly thing to say… You’re running DCS on Windows, right? :wink:

… or did You build a custom realtime OS for Your simulator?

1 Like

Yes, yes, yes.

Yes, MSW. No, not realtime.

The fact that MSW is being used for VR at all is indeed awful. That is why absolutely nothing is installed on that MSW OS that doesn’t have to do with gaming, VR, or CAD modeling. I discovered the nicer TrackPoint drivers from Lenovo ate a ridiculous 20% into my DCS World performance margins, and uninstalled even that. I don’t dare even try to install the full application to reprogram my Logitech G502 mouse.

Anything I actually do, like anything practical, runs under Linux, or in a Virtual Machine. Thankfully, Virtual Box seems not to foul up DCS World in the slightest.

Look at it this way, It is fairly routine to set resolution to leave less than 10% margin in DCS World performance under the worst case loads. And surprisingly, that does work well, consistently. After that, add in a few extra planes in multiplayer, and whoops you’ve gone over the limit right at the merge.

ie. you lose

1 Like

What I’m trying to say is You’re still not enough in control to make that 1.5% margin “enough” in my opinion.

Anyway, I’ll stop hogging the discussion now… :slight_smile:

1 Like

That is indeed what I thought. Extensive, careful FPSVR experiments thoroughly proved otherwise, on both the CPU and GPU side. I adjusted resolution and clocks of both CPU/RAM/GPU/VRAM to test this stuff.

I am rarely surprised. It’s not what you would expect from such a bad software stack, but this really does happen.

Technically, this is not necessarily very hard realtime either. A latency impact of 0.3ms (about 1% at 30Hz) may be nearing the limit of typical realtime Linux, however, the entire frame rendering time is more like a batch job at 13ms, to be displayed while the next frame is already rendering. As long as your kernel doesn’t ‘spinlock’ for much more than the typical 5-7ms, scheduler ‘latency’ may not even be a factor.

At the end of the day, this is why I tell people, measure with FPSVR.


Then a single core boost might actually help. I don’t buy from silicon lottery, so I have no experience there, but I think this tweak allows 1 core to perform even better, by devoting more voltage and cooling to it alone. I would think this applies to specially binned chips too.

IIRC, you can already do stuff like that in BIOS, and until you hit thermal limits, usually all transistors on the chip will be exactly matched far beyond parts per trillion anyway. That is why we can ‘repurpose’ a CMOS logic inverter as an extremely high-gain amplifier, already biased to midpoint voltage. If there were the slightest imbalance between those MOSFETs - and MOSFETs are usually inherently extremely sloppy devices - the amplifier in this configuration would output Vcc (V+) or Vee (V-) - nothing in between.

No. Because they tend to be binned for all-core performance, and of such quality that all those cores are basically the same.

If this stuff did not work this way, then how would we overclock binned GPUs so that all their thousands of cores are >>50% faster?

1 Like

Have you actually read how Precision Boost Overdrive 2 works ?
It seems that you might not have, seeing it will lower voltages on cores when TJmax is hit, to lower the temp and there by being able to boost the clocks PER CORE.
The software does it for you so you dont have to fiddle with the BIOS settings :wink: and no spreadsheet needed to boot :crazy_face:


Yes, I understand that, and I don’t think it is useful in the context of chips that are already pre-binned, overclocked, and watercooled, all of which is what people who care about 1.5% will be dealing with.

Silicon Lottery gives me a printed document with the chip telling me all that stuff, even including motherboard Vdrop compensation settings the chip itself can’t possibly manage.

1 Like

Nope you PAY them do give you a printed document and test their chips they resell for a s…itload of extra money.
With the new AMD software you can just let it do the work for you, slap on a decent H2O cooling kit and you will be on your way.

If you really need those extra 1.5% just buy this device from EK and forget about prebinned chips

Or use it with a prebinned Chip you you really want to sprouse alot of money after 1.5%

And you can forget about helium or other gasses too :slight_smile: not needed at all :smiley:


TEC? Thermoelectric cooling? No way, there is no chance any sort of horribly inefficient Peltier junk has any place in my system. On top of that, Peltier coolers don’t even go down to low enough temperatures because… they are semiconductors themselves. >_>

I want another 50% higher clocks from cooling with LN2.

1 Like