1 hour ago, jabloomf1230 said:

You should have cut this off with his first post.

Freedom of speech (up to a point). Let's see what replies there are.

There are benchmarks already of pre production samples. The results might change a bit after release but still give a direction.

From those benchmarks I can tell that the "12900K" (the equivalent of the 12th gen top cpu model) may have an IPC value as large as 25-30% better than that of the 11900K and that is really good. If this holds it will be 60% faster than my 9900K. 

It might be possible to OC it to 5.3 GHz on all 8 "big" cores simulaneously. Its multicore performance will probably be better than that of the 5950X (honestly, it must be, anyway).

And, again it will probably be a very unefficient CPU.

Karl

Edited July 25, 20214 yr by kaha

I think I heard on a PC podcast that to really gain any advantage from the new desktop big/little architecture we'll need to wait for Windows 11?

11 hours ago, eaim1973 said:

I think I heard on a PC podcast that to really gain any advantage from the new desktop big/little architecture we'll need to wait for Windows 11?

Windows 11 will have changes to the scheduler to take proper advantage of the core types. No practical reason why it couldn't be in Windows 10 but obviously Microsoft wants to give people reasons to move to Win11. 

On 7/24/2021 at 9:16 PM, fluffyflops said:

Its a 20+ year old sim.

It amazes my why this urban legend is still circulating in some heads.

2 minutes ago, JoeFackel said:

It amazes my why this urban legend is still circulating in some heads.

Well both P3D and MSFS have roots in the original ESP code, even Asobo said that.  It will be interesting to see whether there is any effect on the performance of all the sims from the new architecture, but it won't be me evaluating it, my hardware is fixed for the next few years now!

8 minutes ago, neilhewitt said:

Windows 11 will have changes to the scheduler to take proper advantage of the core types. No practical reason why it couldn't be in Windows 10 but obviously Microsoft wants to give people reasons to move to Win11. 

They didn't say they wouldn't work on Windows 10, it's just that at this point of time Windows 10 isn't fully optimised for big/little yet, if ever?

2 minutes ago, kevinfirth said:

Well both P3D and MSFS have roots in the original ESP code, even Asobo said that.  It will be interesting to see whether there is any effect on the performance of all the sims from the new architecture, but it won't be me evaluating it, my hardware is fixed for the next few years now!

My system will be more or less the same except I may buy a 5800 or more likely a 5800xt if the latter CPU gets release towards the end of this year.

4 hours ago, eaim1973 said:

They didn't say they wouldn't work on Windows 10, it's just that at this point of time Windows 10 isn't fully optimised for big/little yet, if ever?

Yes, that's what I meant. Win11 will have changes to take advantage of the different core weights. Win10 will continue to work fine but it will treat all cores as if they were the same 'size'. 

The scheduler… it needs to determine what threads are doing a heavy workload and transfer them to the big cores of the load isn’t transient.

cheers

 

Process scheduling is something of a dark art, I think. I've never seen the code for the Windows scheduler, of course, but as I understand it, threads are scheduled primarily based on the priority assigned when they are started (runs from 0-31, only OS tasks can be priority 0) and how much of their time slices they use. A thread that gets called and then yields back to the scheduler before being pre-empted is considered lower priority than one which takes its whole time slice, because it's doing less work. This is also dependent on whether it's a foreground or background process, and whether it's waiting on I/O. Threads waiting on I/O are prioritised higher because they need to be ready when the I/O operation completes, and if they are waiting on an I/O device that the OS knows is 'interactive' (keyboard / mouse etc), those get prioritised higher again. To get an accurate measure of time taken per thread, it uses the CPU's cycle counter which increments for every instruction cycle in a given time-slice. 

For big/little, it would be a case of a) executing less busy threads on smaller cores, and b) adjusting the core loading to suit the power profile. In a mobile / lower-power mode, say on a laptop with a power-saving policy or just low battery, put more work on the smaller cores which consume less power. I assume there will also be ways for code to signal to the CPU that it is more suitable for big or little cores, but I'm just guessing there. 

12 hours ago, Luke said:

The scheduler… it needs to determine what threads are doing a heavy workload and transfer them to the big cores of the load isn’t transient.

cheers

 

We would not be amused if the sims main process runs on a little core.

4 hours ago, Rob_Ainscough said:

I'm going to have read up more on Intel's bit.Little, I'm just not seeing the advantage for desktop computing.  It seems to be about power efficiency yet the 12900 is a 228 Watt unit vs. AMD 141 Watt unit?  What am I missing?  All I'm getting out of it is less intense tasks go on the little cores while more intense tasks go on the bigger cores ... so why/how is that going to improve performance overall?

However, it seems Windows 11 will adjust the schedule to benefit Intel's big.Little according to this article.  I just don't understand how the lower performing cores is somehow a benefit in performance?

Cheers, Rob.

Rob, I have been wondering the same which is why I posted the question.

Having done some more reading up, it could be that with the new design and a move to smaller wafer sizes, higher clock speeds could potentially result, but again I have read that that advantage may also be put towards making the cores more efficient in operation aka reduced power consumption, so the gain is not a given at this stage.

 

The other possible benefit , although personally I think this is somewhat limited where P3D is concerned, the new design will offer additional threads to run those processes.  e.g.

If you have 8 core CPU today which is capable of running 16 threads with HT turned on, potentially the equivalent CPU when Alder Lake arrives is 8 BIG cores (with HT on)  which equates to 16 threads plus potentially another 8 cores that can be made available by Windows based on the Little cores (no HT available), so in essence you have 24 threads at your disposal (16 +8) .

So theoretically you have 8 more threads to play with, but of course can P3D really take advantage of all those available? In another forum post (I think it was Ray's) I had the impression that while P3D and its concurrently running add-ons could be spread across other, less heavily used cores, adding more and more cores (threads) has a diminishing rate of return. I guess it all depends upon how many other add-ons you run concurrently with P3D and this will differ from user to user.

 

And of course there are other things to consider with moving to this new Intel platform in that:

• you will need to invest in a new motherboard that can host the new LGA1700 CPU socket needed (+ CPU cooler kit/adapter)
• there's still open debate about the DDR5 memory but it also likely to be super expensive if it is supported by Alder Lake
• and from what I can tell PCIe 5.0 is still some way off even if a new motherboard can offer it, the GPU suppliers need to play catch up and even when they do, its likely to be ultra expensive graphics card.

 

Therefore, in my opinion, even if the performance advantage and power saving warrants the move up, I think its going to be a very expensive rig at the end of the day.

 

Best,

@787flyer I tend to agree with you Mark.  I've been intending to do a rebuild from my 9700k rig to the alder lake.  I saw the big/little news and was not too excited as it appeared it was heavily focused on power saving.  But I am going to keep an open mind. 

Quote

.....Ultra expensive graphics card....

So, no change there then

Edited July 27, 20214 yr by Christopher Low