AMD announce new processors

[Rob Ainscough]

Do NOT attempt to control CCD via mapping, this resulted in having to reset my BIOS via button on the Asus 670E Extreme MB (do NOT change):

 

[Rob Ainscough]

Running 3D Mark Time Spy 4K results show about 8.4% performance increase (SMT off):

Also make sure Windows 11 game bar is installed and enabled and configured:

Apparently the job scheduler will perform differently making the 3D vCache cores have a higher priority.

Next I'll break out Process Lasso and see if it helps with enforcing core assignment.

Cheers, Rob.

[Rob Ainscough]

Prior to using Process Lasso I tested using P3D's built-in core affinity settings with the objective to keep all processing on cores 0-7 (3D vCache) ... with SMT OFF:

AffinityMask=255 (make cores 0-7 available)
P3DCoreAffinityMask=170 (set P3D to use core 1,3,5,7)
MainThreadScheduler=0 
RenderThreadScheduler=2 
FrameWorkerThreadScheduler=4

With that set, ran a quick flight from same location EGLC to EGLL all maxed for Graphics.  I performed 1/2 the flight with no AI, then started AIG at 100% (80% of available airline routes selected in AIG manager) ... no significant change FPS.  Promising ... next I added 50% road/ship/boat traffic (the streets and water way of London were packed with cars/boats) ... averaged about a 2 FPS drop.  

Overall FPS compared to older 7950X in the same scenario was about 2 FPS higher which closely corresponds to the 8% TimeSpy score difference between the CPUs.

Very promising, the world was alive with road, ship, aircraft traffic.  I'll get actual data posted this weekend.  But from what I've seen so far it's about what I expected.  But getting the CPU setup for this type of activity was tricky but I did confirm that cores 8-15 were doing nothing while the cores 0-7 (3D vCache) were working per affinity settings above. 

I'll get to MSFS testing of AI/road traffic impact on the 7950X3D after P3D testing.  I suspect results will be similar to that of P3D.

[Ray Proudfoot]

@Rob Ainscough, this sounds very tricky to setup. Other applications - AS, ChasePlane etc. can be assigned to the remaining cores so the CPU is fully utilised.

You don’t mention resolution. 3840*2160?

[Rob Ainscough]

Same settings as before on the graphics side 3840 x 2160 and same flight from EGLC to EGLL same add-ons same PMDG 737-8 …  isolating just AI aircraft and road traffic settings to see what impact they may or may not have.

Yes those will get assigned to other cores, either via tsk mgr or process lasso.  However, Process Lasso does come with a slight performance penalty.

Cheers Rob.

 

[Ray Proudfoot]

Guru3D have published a full review of the 7950X3D. Scrolling towards the last few pages showing various game performance a few are slightly faster with the AMD, a few are faster with the i9-13900K.

https://www.guru3d.com/articles-pages/amd-ryzen-9-7950x3d-processor-review,1.html

The primary benefit of the AMD seems to be far cooler running.

[Rob Ainscough]

On 3/11/2023 at 9:09 AM, Ray Proudfoot said:

Scrolling towards the last few pages showing various game performance a few are slightly faster with the AMD, a few are faster with the i9-13900K.

I usually respect Guru3D, but in the article you linked they failed on a few fronts:

Above is from Guru3D website, there is no BIOS/EFI "9922" for the Asus Crosshair X670E Hero?

But here is where Guru3D failed to setup the CPU correct for game testing:

Again, this is from that Guru3D article on their website ... notice at the bottom left you see "SMT Unit #1"... absolutely do NOT enable SMT for games.  So from this point forward their gaming results are now all incorrect.  Did they set Dynamic Core Preference to "Cache"?

Here is my AIDA64 CPUID:

As you can see I have SMT disabled.  I'm assuming when Guru3D screen captures CPUID they happened to hit a low frequency state for the CPU clock and multiplier?  Even so, if you notice Guru3D have the CPU voltage at 1.376v even when running at low frequency??  Compare that to my setup where I have the CPU at 1.119v at much higher frequency.

But finally, the kicker ... where did all 3D vCache go?  Guru3D shows 64MB and I have 192MB (which is accurate)?  We're both running the same version of AIDA64.

With that said, my Time Spy scores (above are considerably higher than what Guru3D reports).  Since they didn't setup the 7950X3D for gaming, all their gaming scores are irrelevant/skewed.  

If the belief at Guru3D or anyone for that matter is they can drop in a CPU and not setup anything in the EFI/BIOS, that's just wasting performance (and one's money).

Cheers, Rob.

[Ray Proudfoot]

26 minutes ago, Rob Ainscough said:

absolutely do NOT enable SMT for games. 

I have no idea what that means. The equivalent of Intel’s Hyperthreading?

With so many cores I wouldn’t have thought it necessary given P3D can’t use more than 6 or so.

I see you use a lower voltage. How did you determine that?

[Rob Ainscough]

16 minutes ago, Ray Proudfoot said:

The equivalent of Intel’s Hyperthreading?

Correct ... no game runs faster with more than 8 real cores so it's pointless to enable.

Once you exceed 8 cores (which P3D will do if allowed) the effectiveness on performance actually deteriorates.

16 minutes ago, Ray Proudfoot said:

I see you use a lower voltage. How did you determine that?

That's a secret ... gotta have some secrets or else these "builder" companies will run with it as their own secret  

But like I said, the Guru3D folks need to figure out what happened to their L3 cache, they're missing a huge chunk of it.

Cheers, Rob.

EDIT: and Guru3D need to turn off their "Integrated Video" in the EFI ... ugh, Guru3D very sloppy for game testing ... shame on them

[Andrej Ropret]

Amazing review Rob, simply outstanding job you do comparing the CPUs. Thank you.

Reading this review:

https://www.tomshardware.com/reviews/amd-ryzen-9-7950x3d-cpu-review/6

shows us massive increase in MSFS performance, if I quote them:

Quote

The 7950X3D’s performance in Microsoft Flight Simulator 2021 is almost unbelievable — the X3D chip is 53% faster than its vanilla counterpart, the 7950X...

... Microsoft Flight Simulator 2021 obviously benefits tremendously from L3 cache

Do you agree with them recommending -X3D version for us using P3D5 and more often even MSFS2020?

Thank you for your thoughts about the Tomshardware review.

Cheers,

[Rob Ainscough]

I have not done full data logged 7950X to 7950X3D compare in MSFS or P3D yet ... was hoping to do that last weekend, but got diverted with other projects and probably not be able to get to it this weekend either.

It's a little tricky to setup the AMD CPU to optimize the 192MB of vCache ... but one completed, performance was definitely higher ... but I'm not seeing 53% higher when I do a quick visual run MSFS, but I promise I'll get to testing within the next couple of weeks.

I'm currently setting up a 18U rack mount with 4 computers (in addition to my 2 computer SimPit) of varying performance that I eventually plan to test with on lower order hardware.

I've been trying to get an AMD 7900XTX GPU that does NOT have a defective heat pipe cooler ... none of the manufacturers (XFX, Sapphire, etc.) are prepared to confirm a new unit purchase will have a none-defective heat pipe cooler.  They seem to be playing a game off buy, try it, and if it overheats then ask for an RMA ... probably why the GPU has dropped in price. 

Anyway, more testing to come in a couple a weeks.

Cheers, Rob.

[Kael Oosthuizen]

Hi @Rob Ainscough,

 

I just wanted to say thanks for all the work you've put into these benchmarks and testing - very much appreciated. 

I am just building a new rig (waiting on the 7950X3D to arrive hopefully next week) and was tossing up between going back to P3D or continuing to use MSFS. I'd moved away from P3D largely because running P3D with heavy AI, 3 party aircraft and 3rd party airports was killing my FPS, more so than it did on MSFS (noting you can run FG with MSFS).

 

That said, It looks from your results with the 4090 and the 7950X3D that you're able to run heavy AI in complex airports and aircraft and stil maintain quite a bit above 30 FPS which is the sweet spot for me. So I'll give some thought to reinstalling P3D instead (I don't like running both sims).

Keep up the good work!

 

[Rob Ainscough]

I'm using AIG with just about every route available at 100%.  I give it about 10-15 mins to ramp up aircraft flow ... then start testing.

The AI density in MSFS seems to be about 1/3rd that of P3D using the same 100% settings ... and of course the spinning AI aircraft in MSFS.  I find that about 35% P3D/AIG AI traffic = about 100% MSFS/AIG AI traffic.  I normally don't run AIG AI above 80% or I'll be waiting in line a long time at Heathrow (not just taxi, but arrivals are lined up in the sky).

I use AILRP (AI lights reborn) for P3D with AIG to get perfect night lighting (highly customizable).

Here is some real world footage from PDX, notice how the landing lights on some aircraft (not all) pulsate (dim/bright).

Cheers, Rob.

 

[Kael Oosthuizen]

On 3/18/2023 at 5:04 PM, Rob Ainscough said:

I'm using AIG with just about every route available at 100%.  I give it about 10-15 mins to ramp up aircraft flow ... then start testing.

The AI density in MSFS seems to be about 1/3rd that of P3D using the same 100% settings ... and of course the spinning AI aircraft in MSFS.  I find that about 35% P3D/AIG AI traffic = about 100% MSFS/AIG AI traffic.  I normally don't run AIG AI above 80% or I'll be waiting in line a long time at Heathrow (not just taxi, but arrivals are lined up in the sky).

I use AILRP (AI lights reborn) for P3D with AIG to get perfect night lighting (highly customizable).

Odd, the difference is AI numbers wasn't something I observed between MSFS and P3D. I used AIG and AI Companion to monitor numbers and seem to recall seeing similar numbers at similar settings. Not in a position to test anymore having binned P3D on the current system.

 

Agree AILRB was a great addon for P3D. Using FSLTL models (instead of AIG models) in MSFS results in far better lighting and sounds. I understand AIG are working on improvements for both lighting and sound but looking at what FSLTL have done its clear to see how much improvement is possible.