Jump to content

AMD announce new processors


Ray Proudfoot

Recommended Posts

Wilhelm Zwirchmayr
2 hours ago, Sang Hyun Yoon said:

runs much smoother than the previous 12700kf I had 

 

 

IMG_20221026_212510.jpg

Interesting,

how much more fps do you have or is it just holding the fps better with your new cpu

I have a 5900x & RTS 3080, I think only changing the graphics card will bring an improvement

maybe if one of you has an RTX 4090, I would be interested in the performance on the WQH resolution

Link to comment
Rob Ainscough

I finished my baseline testing (with video and data) for P3D/MSFS on my 5950X nv3090 and will post those results later this week -- I got some interesting results and definite pattern on how MSFS performs (inverted U chart).  Next test will be 5950X nv4090, then I move onto 7950X and nv4090.  Transitioning all my stuff from 5950X to 7950X will take some re-install work but hopefully done by the following weekend (Nov 5th).

I'll include raw data links so folks can validate or go thru my data however they like.

If you plan to turn up graphic settings with heavy add-ons (aircraft/scenery) installed, you'll definitely want min 16GB VRAM and 32GB RAM regardless of sim platform choice.

Cheers, Rob.

  • Like 1
  • Thanks 2
Link to comment
Sang Hyun Yoon
23 hours ago, Wilhelm Zwirchmayr said:

Interesting,

how much more fps do you have or is it just holding the fps better with your new cpu

I have a 5900x & RTS 3080, I think only changing the graphics card will bring an improvement

maybe if one of you has an RTX 4090, I would be interested in the performance on the WQH resolution

 

 

60fps locked and it's looking quite good.    

 

 

 

 

Link to comment
Ray Proudfoot

I’ve been checking Chillblast here in the UK for the i9-13900K CPU but it isn’t available yet for custom builds. I guess it takes a couple of months for them to get over here. It looks to deliver a nice boost over the i9-11900K.

Somewhat disconcertingly the fastest card is the 3070Ti with just 8Gb. I’ll wait until the 13900K is available and then ask if they’ll be getting the 3080Ti or perhaps a 3090. Unlike @Andrew Wilson I’m unlikely to get a good deal on the 3090Ti. :(

Look forward to seeing your performance report @Rob Ainscough.

Link to comment
Rob Ainscough

 

I'm going to order a new PCIe power cable (cable mods) for the 4090 after all the reports of them melting and catching fire. 

The socket design is bad (shame on you nVidia/UL/ANSI) ... that's 600-650 watts (600/12 = 50 Amps), so adapting to existing PSU (1600watt EVGA T2 rated 133A on 12v) to run eight 12v 14 gauge wires.  Standard PSU 14 gauge wire is rated to 6 AMPS (more if mil-spec wire is used at 17 AMPS, but no one uses mil-spec wire in a PC) so 8 wires at 14 gauge still comes up short at 48 AMPs.  In addition, the insulation for that much current needs to increase (at the socket and wire) or be of better quality material (i.e. mil-spec).  Considering this is a $1800 GPU it's shameful disregard to safety, I can guarantee you lawsuits will follow if nVidia doesn't do something about this quickly.

AMD have taken note of this issue and have proactively decided to change their socket design on their next gen RNA3 7900XTX GPU so as to be "safe" and avoid potential fires and subsequent legal battles.

Hopefully end of this weekend I can start releasing the performance data (it's a long slow process).  

Tossing out FPS numbers is fairly meaningless unless they're accompanied with flight location, add-ons used, all (no exceptions) the graphics settings, and logged data to file over time so averages, max, min values are recorded/calculated ... this takes time and effort which is my objective.  In my logging I used 100ms sample rate (lowest I could go with my tools) ... many tools default to 500ms or even 1000ms which is just not enough to report accurately.  In fact, if I could go lower than 100ms I would but my logging tools seem don't permit.  When you consider that FPS is operating anywhere from 6ms to 60ms you can see how lower sample rates would be beneficial.

Here is a quick teaser for baseline 5950X with nV 3090 (graph of FPS overlay at end):

Full flight with data points shown at end of video:

This is just one "Baseline" set for my "Complex" add-on scenario ... I have two more baseline sets for moderate (CRJ-700) and light add-on scenarios (PA34 Seneca V) that I'll post later.  This are just baseline (starting point).  

I'll run all 6 of these tests again with AMD 5950X and nv 4090.  Then 6  more times for AMD 7950X and nv 4090.  Then 6 more times for AMD 7950X and AMD 7900XTX (when available). 

But again, graphics settings and add-ons are everything, just one change in one graphics setting can make a huge difference in performance ... for example the render scaling in MSFS is left at 100 (TAA), move that slider to the right drastically impacts MSFS performance at 4K resolution, if I maxed that out I would looking at 2 FPS and that's just not flyable, so my goal was to try to keep both sims above 20 FPS on average and those setting I list in video permit that goal.

FYI, this is with P3D V5.3 and MSFS latest build (not Beta build).  I'm not permitted to show future version ;) if such versions exist nor admit their existences even if they did exist.

EDIT: I was going to use FSL A321-SL for my baseline, but since there isn't FSL MSFS version I used PMDG ... I would have preferred to use FSL since IMHO it's the higher quality and most complex aircraft I've ever used in any simulator platform.

  • Like 3
  • Thanks 1
Link to comment
Ray Proudfoot

Thanks for posting those Rob. That’s definitely the hardest test in any sim. Even the fastest kit can’t run with all sliders right.

I’ll be avoiding the 4000 cards until those problems are resolved.

Link to comment
Rob Ainscough

Not the fastest kit, that's my baseline (now old computer) ... the fastest kit tests are yet to come but now I have my baseline to compare with AMD 7950X and nv4090 ... stay tuned.

Waiting for good power cables to arrive, not going to plug my 4090 in with so many folks experiencing fire/smoke/melted sockets.

Cheers, Rob.

Link to comment
Ray Proudfoot

Ah, in that case I look forward to a consistent 30fps throughout the flight. :D

Link to comment
Rob Ainscough

Maybe ... but I can get them both to go lower, increase P3D to 8X SSAA or increase MSFS render scaling above 100 (which moves from TAA to SSAA).  Heck, not even using weather or AI traffic which will increase the load and lower FPS even more ... I didn't include weather/AI because those two features will provide very different results and not be consistent for each sim platform (variance in AI traffic count, variance in cloud depiction/layers, etc.).

Cheers, Rob.

Link to comment
Wilhelm Zwirchmayr

Hello, I did a test with my existing system (5900x 4.5 GHZ 3080 MSI water cooling 32 GB ram Monitor WQHD).

My Prepard3d is blocked at 30 FPS
Aircraft.: FS LABS A319
my graphics card is actually only used to a maximum of 50% at the airport in daylight and only about 30% when flying
When my FPS drop briefly, my CPU runs into the limit for a moment

the same at night (with fog) and that is more demanding for my graphics card, i have max. 80% utilization.
in flight, the graphics card is also loaded a little more, around 40-55%
The CPU is exactly the same as daily.


So a better CPU would bring more stability for me and if I set the FPS to unlimited, my system heats up and I get a lot more stutters.

With my current system, I have a maximum of 71 °C for the cpu and a maximum of 57 °C for the graphics card
the clock rate of the graphics card remains stable at 1970-2000 mhz

Link to comment
Rob Ainscough

Just a quick update, installed the 4090 in my existing AMD 5950X based system and performance didn't really increase much ... my settings and add-ons are very CPU intensive.   With that said, I was able to increase P3D to 8XSSAA, set Dynamic Reflections to Ultra and with Dynamic Lights present there was 1 FPS drop and GPU load increased from 43% to 67% ... so clearly I'm still very CPU bound.

Rob.

Link to comment
Wilhelm Zwirchmayr
18 hours ago, Rob Ainscough said:

Just a quick update, installed the 4090 in my existing AMD 5950X based system and performance didn't really increase much ... my settings and add-ons are very CPU intensive.   With that said, I was able to increase P3D to 8XSSAA, set Dynamic Reflections to Ultra and with Dynamic Lights present there was 1 FPS drop and GPU load increased from 43% to 67% ... so clearly I'm still very CPU bound.

Rob.

Exactly, that's what I noticed with my system too.

Since I'm using an amd 5900x, only a cpu upgrade would help to keep the 33 fps.

Link to comment
18 hours ago, Rob Ainscough said:

Just a quick update, installed the 4090 in my existing AMD 5950X based system and performance didn't really increase much ... my settings and add-ons are very CPU intensive.   With that said, I was able to increase P3D to 8XSSAA, set Dynamic Reflections to Ultra and with Dynamic Lights present there was 1 FPS drop and GPU load increased from 43% to 67% ... so clearly I'm still very CPU bound.

Rob.

Curious to see the uplift in MSFS, which seems to take better advantage of hardware.

Link to comment
Rob Ainscough

I ran the same test last night for MSFS, I'll publish video and data later tonight ... MSFS gained 0.5 FPS to 25.1 average from 24.6 average.

BUT, I need to check a few things because my watt usage being reported is not accurate for the 4090 in both MSFS and P3D.  When I run 3DMark TimeSpy, I get about what most report 28051 score for a normal 4090 so I know the card is using more current or else it couldn't reach that TimeSpy score.

16 minutes ago, Alex Pugh said:

which seems to take better advantage of hardware

Not sure what you mean there ... can you give me some specifics and supporting data so I can evaluate?  Based on the videos I posted above you can see P3D CPU usage at 34% AVG and GPU 76% AVG and MSFS CPU usage at 33% AVG and GPU 46% AVG ... you can probably attribute the higher GPU usage in P3D to SSAA vs. MSFS TAA.

Both P3D and MSFS operate DX12 API, however, I didn't use DX12 in MSFS (per video) because the performance was actually worse in my testing so I stayed with DX11 for MSFS ... not exactly sure what is going on there (with DX12 in MSFS) but my guess is the WASM layer MSFS operate under isn't fully leveraging the very hardware specific functions available to DX12 graphics API ... I know Asobo did have to go two code paths, one for XBOX and one for Desktop Windows.  

Whenever you add a translation layer (WASM no matter how efficient) there will be some loss of performance and direct access restrictions, that's unavoidable ... on average WASM based application/game will lose about 10% from it's "native" performance potential.  With that said, WASM is pretty efficient (I'm current building a web app based on WASM and Blazor for UI) but it's still a translation layer that must be supported by the browser in my case.

Cheers, Rob.

Link to comment
Rob Ainscough

Test results so far (need to investigate why GPU power consumption for the 4090 is so low for both P3D and MSFS):

uc?export=view&id=1cvC2Dp3QeXiXKQdYJrzwt

The highest consumer of VRAM was MSFS at 15874 max.

The highest consumer of RAM was P3D at 23131 max.

The best FPS goes to MSFS at 25.4 avg.

Raw data files are here for anyone to validate/review.

PMDG 737-8 4090 Video P3D

 

PMDG 737-8 4090 Video MSFS

 

Shifting to a moderate complexity location EGPP to EGNM in lighter weight aircraft AS CRJ-700.

AS CRJ-700 Video P3D

AS CRJ-700 Video MSFS

Move away from London FPS improves with P3D trading places and taking the top spot at 36.0 FPS with MSFS at 34.2 FPS.

Testing continues ... 

Cheers, Rob.

 

 

  • Thanks 3
Link to comment
Ray Proudfoot

Thanks for that Rob. The 4090 hardly improves fps at all. I’m eagerly awaiting the 7950X processor. That should give things a nice boost.

  • Like 1
Link to comment
Rob Ainscough

I must admit, I was expecting higher FPS even with a heavily CPU dependent scenario (aka London and PMDG).  I need to research why the GPU usage and power consumption was so low (much lower than my 3090).  In TimeSpy, the 4090 pulls a full load and the score is similar to other 4090s ... not sure if it's just the logging software that's not working with new GPU or something else, driver issue?

Or maybe the 4090 is just not good with flight simulators and "requires" DLSS support to perform?

Looks like it will be a month or two before I get my AMD 7900XTX GPU, announce on the 3rd Nov (tomorrow) but that's NOT the availability date.  AMD 7900XTX raster performance is even higher than 4090 and they use a more conventional (read doesn't catch on fire) power socket.

Delivery of the EK full front/back waterblock for the 4090 has been pushed back until later November (was supposed to ship next week).

But, I'll push along with AMD 7950X CPU now that I have baselines ... but that's will take some time as I have to install new OS, P3D, MSFS all over again.

Link to comment
Rob Ainscough

Quick update on the Power consumption being logged by RTSS, it's wrong for nv 4090, correct for the 3090.  So it seems RTSS needs an update. 

I tested with HWINFO64 (after updating it) and I see correct power consumption for the 4090 ... in my case reporting 589 Watts under peak load in P3D (8X SSAA at 4K res added 40 watts).

Asus have a PSU power recommendation chart for 4090 + CPU and in my case they recommend 1200 Watt PSU (I'm using 1600 Watt EVGA T2).  

Waiting for 1pm PT today (Nov 3) to get RNA 3 (AMD 7900XTX GPU) price and availability.

Cheers, Rob.

Link to comment
Edgar Del Valle

Thanks Rob for all the good info that you share. I am due for an upgrade. Current hardware is an I7 4790K and an Nvidia 1080Ti. Planning on waiting on AMD's new Ryzen 7000 3D

V-Cached CPU and either the new AMD Graphics card or the Nvidia 4090.

I am also considering the AMD 5800X3D but being an older platform I am not sure that is the right call.

Came to this thread thinking that you may have been experimenting with 5800X3D but it is not the case.

Anyways, I am following this thread. Thanks.

Ed

Link to comment
Rob Ainscough

Hi Ed,

Bang for the buck is probably the AMD 7600 or Intel 13600K.

AMD just announced the new RNA 3 GPU cards today 7900XT and 7900XTX.  the 7900XTX is $999 available Dec 13th

uc?export=view&id=1svML4QNMr-tgHznuW9t2A

7900XTX is 24GB VRAM and 355W and has better raster performance than the nv4090.  Don't know ray tracing numbers.

I wouldn't go with the 5800X3D, it's a great processor but doesn't support DDR5 and PCI 5.0 which is what you get with AMD 7600X.  I believe the 7000X3D is Q1 2023?  I think, not sure.

On to setting up the AMD 7950X + nv4090 testing now ... 

Cheers, Rob.

 

 

Link to comment
Wilhelm Zwirchmayr
11 hours ago, Rob Ainscough said:

Hi Ed,

Bang for the buck is probably the AMD 7600 or Intel 13600K.

AMD just announced the new RNA 3 GPU cards today 7900XT and 7900XTX.  the 7900XTX is $999 available Dec 13th

uc?export=view&id=1svML4QNMr-tgHznuW9t2A

7900XTX is 24GB VRAM and 355W and has better raster performance than the nv4090.  Don't know ray tracing numbers.

I wouldn't go with the 5800X3D, it's a great processor but doesn't support DDR5 and PCI 5.0 which is what you get with AMD 7600X.  I believe the 7000X3D is Q1 2023?  I think, not sure.

On to setting up the AMD 7950X + nv4090 testing now ... 

Cheers, Rob.

 

 

I also saw the presentation
and I find it very exciting.

I believe that the top model will be positioned between 4080 and 4090.
which is perfectly fine for the price and consumes a maximum of 100 watts less electricity

now nviidia has a big problem with the 3080 & 4080 or TI.

Link to comment
Rob Ainscough

Agree, $1700 for my Asus TUF OC edition 4090 to get less than 1 FPS improvement in high load scenario is definitely not good value in my book. Being able to run 8X SSAA at 4K res without a performance hit is nice, but not $1700 nice.

I think AMD are poised to take considerable market share away from nVidia.  Come December 13th I suspect 4090 retail prices will drastically drop.

Cheers, Rob

  • Like 2
Link to comment
  • 2 weeks later...

I believe that frame generation (without DLSS) will be an invaluable asset for simmers. On MSFS it has allowed me to achieve 60 fps+ @ 4K/Ultra using the Fenix A320 on major hubs with loads of AI traffic. Unfortunately no such option on P3D or XP12.

Link to comment
Rob Ainscough

Quick update: my main FS PC is back together ... work and life has kept me busy so haven't spent much time with it and have Win11 22H2 loaded with Cinebench R23 and TimeSpy.

I've done some very basic overclocking to the AMD 7950X, Asus UEFI provides a lot more options to adjust this time around and different strategies.  No problem hitting 5.9Ghz on 12 cores and 5.5Ghz on remaining 4 cores with peak temp at 82C when running Prime95 and Cinebench R23 (AMD say 95C all day long is normal operating temps under max loads and can run all day at 95C).  RAM is 6000Mhz CL30 - EXPO II profile.  I should be able to hit 6Ghz without a problem once I dig more into the OC options.  I haven't experimented with the Async mode where the BClk adjustments impact only the CPU and not entire board (RAM etc.).  Line load calibration is at Level 5 (recommend for OC) but I haven't experimented.

I don't have a Gen 5 M.2 only the older Gen 4's which come in at 7000MB/s reads (Gen 5 is around 10000MB/s).  On the video/GPU side of things to note is that if I use more than one M.2 the PCIEX16_1 will run at the 4090 Gen 4 @ 8X.  So if you want absolutely maximum performance, then stick to just ONE M.2 ... I haven't found any Gen 5 M.2s available yet.  Maybe in the future I'll replace my two M.2's 2TB Gen 4's for one M.2 4TB Gen 5 ... not sure how much faster initial loads times will be in MSFS or P3D (both are painfully slow even with 7000MB/s Gen 4 M.2s).

Cinebench R23 MultiCore score for the AMD 7950X was just a tad over 39,102 which is a significant improvement over my AMD 5950X at 27383.  I'm certain I can get over 40,000 as I have 13C of headroom and the CPU seems to respond well to lower temps (my Silent Chiller operates on both CPU and RAM). 

Adapting the DDR5 RAM modules to my existing EK RAM water block took some work with getting the correct thermal pad thickness on each side for optimal heat distribution and physical fitment, but happy with how that turned out.  Stock DDR5 RAM 6000Mhz modules can get over 72C ... my chilled ones peak at 39C under load.  

Hopefully this weekend I can gather up all the data and post results.

Cheers, Rob.

  • Like 4
Link to comment
Rob Ainscough

Still installing P3D and MSFS, this may take some time.

Consistent with prior Intel HT and AMD SMT testing, SMT OFF is the optimal way to go for performance in games/sims - 10% increase overall with 13.6% CPU performance increase in TimeSpy.

AMD 7950X 4090 -SMT ON (29073 - 34655/15201):

image.thumb.jpeg.0b0798ef9038d0db41e662f4a64825b8.jpeg

AMD 7950X - 4090 - SMT OFF (32005 - 37674/17276):

image.thumb.jpeg.e02f9880c55114926a217e08d406cc6a.jpeg

AMD 5950X - 3090 - SMT OFF (20265, 21214/16170):

image.thumb.jpeg.ee62fcdf4c154fa5f0d5de7b1dbd9e5f.jpeg

Intel 9900K - 2X 2080Ti (SLI) - HT OFF (23547 28792/11588) :

image.thumb.jpeg.ed16c48549ad5f11a588c0961b359a46.jpeg

Looking at TimeSpy CPU scores:

AMD 7950X vs Intel 9900K = 49% improvement (note, the 9900K was tested under an older version of TimeSpy).

AMD 7950X vs AMD 5950X = 6.8% improvement

Cinebench R23 a different story ...

Intel 9900K Single Core (1305):

image.thumb.jpeg.3e87270049d1868a77f379ddb564beb7.jpeg

AMD 5950X Single Core (1653):

image.thumb.jpeg.a09ee612135dcec7bcde84cf42d73dd3.jpeg

AMD 7950X Single Core (2086):

 

Use Cinebench Single Core the performance variance:

AMD 7950X vs. Intel 9900K (5.3Ghz) = 59.8% improvement

AMD 7950X vs. AMD 5950X = 26.1% improvement

Moving onto flight sim platform testing will continue with P3D and MSFS scenarios per original baselines to get sim world variances (might not be this weekend, maybe over Thanksgiving).

I'll also post my UEFI settings for the AMD 7950X once I finalize, I've only done very basic OC testing on the 7950X.

EDIT: oops forgot to include that last CinebenchR23 screenshot ... updated.

image.jpeg

  • Like 1
  • Thanks 1
Link to comment
Rob Ainscough

I converted the ASUS 4090 TUF Gaming OC Edition GPU to a water cooled (front/back) block from EK.  The EK manual was simply wrong in a few locations and missing 3 screws (fortunately I have a good collection).

With that said, glad to get my 4090 down to a more manageable size ... now I can get back to testing P3D/MSFS performance.  This is dual water loop setup, 4090 goes thru 480 / 60mm radiator, the CPU and RAM go thru my external silent chiller (vapor chiller for sub-ambient).

Cheers, Rob.

IMG_5529.jpg

IMG_5530.jpg

IMG_5536.jpg

IMG_5537.jpg

IMG_5539.jpg

IMG_5540.jpg

IMG_5541.jpg

  • Like 5
Link to comment
Rob Ainscough

FYI for anyone else doing this type of build ... moving my GPU to PCIEX16_2 slot (because it no longer fits in PCIEX16_1 due to a poorly place motherboard display unit) means my GPU will only be able to run X8 and not X16.  

I "assumed" that a $1000 motherboard would:

  1. Allow for full coverage water cooled GPUs to fit in PCIEX16_1 slot
  2. Allow me to select which one of the two PCIEX16 slots to run at X16

Unfortunately that is NOT the case, the ONLY slot this Asus X670E Extreme motherboard can operate a X16 is PCIEX16_1 slot (regardless of M.2 configurations).  That does beg the question why Asus falsely label the slot as PCIEX16_2 when it has no ability to run at X16 regardless of hardware configuration?

This is clearly a motherboard limitation as the CPU PCIE lanes are sufficient to support:

image.jpeg.bb519d51a87e5955671a9d309d5f825a.jpeg

  • nVidia 4090 GPU (16X) PCIEX16_1
  • M.2 4X via Chipset (4X)
  • M.2 4X via General Purpose PCIe (4X)
  • USB 3.0 PCIe card (4X) PCIEX4

All within lane spec usage ... however because I moved the GPU to PCIEX16_2 this is what I'm forced with:

  • nVidia 4090 GPU (8X) PCIEX16_2
  • M.2 4X via Chipset (4X)
  • M.2 4X via General Purpose PCIe (4X)
  • USB 3.0 PCIe card (4X) PCIEX16_1

Really not happy with ASUS right now ... maybe other manufacturers do the same?  ASUS does NOT permit me to set PCIEX16_2 to 16X and PCIEX16_1 to 4X - why??

I don't know exactly how much this has impacted performance?  I was able to improve TimeSpy score both CPU/GPU to 32379 and very good temps with peaks at 58C CPU and 47C GPU under full stressed loads.

I did run a quick test of P3D using same EGLC to EGLL route with same 3rd party addons installed and eyeballed FPS range 33 to 57 FPS but will do a more formal test with supporting data later this week or weekend.

Rob

 

Link to comment
Rob Ainscough

Here's the TimeSpy data:

CPU Frequency peak at 5.94 Ghz

CPU Temp Peak 54.7 C

GPU Temp Peak 57.1 C

GPU +100Mhz GPU VRAM +1100Mhz Volt +50mv

SMT-OFF-TimeSpy-4090OC-MSI-50-100-1100.jpg

SMT-OFF-TimeSpy-4090OC-MSI-50-100-1100-details.jpg

SMT-OFF-TimeSpy-4090OC-MSI-50-100-1100-details2.jpg

SMT-OFF-TimeSpy-4090OC-MSI-50-100-1100-details3.jpg

  • Like 1
Link to comment
Rob Ainscough

What I discovered on the CPU side is that setting all cores at a fixed frequency is a bad idea.  Overall performance is much better when using PBO curve optimizer and let it manage CPU frequency/voltages based on temps so long as there is sufficient cooling.  I do "force" a cooler point value and removed the "cooler learning" which has provided a more consistent optimizer.  You can also add CPU performance conditional logic based on temps/volts in the UEFI.  I did update the UEFI rev 0805.

To obtain optimal performance I recommend NOT using more than two DDR5 RAM slots and stick to 32GB RAM ... hopefully that changes in the future ... there are 32GB modules (X2 for 64GB) but they run slower (currently).

  • Like 1
Link to comment
Rob Ainscough

And here is some data for 7950X using same scenario (Apples to Apples) in P3D:

P3D-EGLC-EGLL-737-8-7950X-nv4090_Totals.thumb.jpg.47d2402b32b2f4cfc31edc4ba143f388.jpg

P3D-EGLC-EGLL-737-8-7950X-nv4090_Chart.thumb.jpg.ef7d4b51c6008b73d5524384c6775fde.jpg

 

P3D-EGLC-EGLL-737-8-7950X-vs-5950X-nv4090_Chart.thumb.jpg.95be3233ec6a658a6356839a7ec62846.jpg

In this relatively extreme scenario of add-ons and graphics settings the:

AMD 5950X (nVidia 4090) was 23.4 FPS AVG 

AMD 7950X (nVidia 4090) was 34.0 FPS AVG

Performance improvement going from an AMD 5950X to AMD 7950X is 45.2%

As you can see from the data, overall throughput is improved with much less variance and fewer long spikes in FPS.  Overall, this is probably the biggest CPU improvement I've seen in a while going from prior gen CPU to current gen CPU.

For those interested in the data files they can be found here:

Raw RTSS logged data.

Converted and Charted RTSS data to Excel.

Here is the video of the test flight (give YouTube some time to resolve it to 4K resolution):

This weekend I hope to be able to run the same test in MSFS for another Apples to Apples compare between the 5950X and 7950X AMD CPUs.

Cheers, Rob.

 

  • Like 2
  • Thanks 2
Link to comment

Unreal engine is probably the version after the next for p3d. Even so I think the home user sim from Lockheed is a thing of the past. Feels like end of the road for p3d

Link to comment
Rob Ainscough

You gents need to be a little more negative … geez … anyway happy holidays!  
 

I do like UE5 but no one has done a flight simulator with it … it’s all about render distance (LOD). 

So I guess it’s end of road for MSFS and XP because they perform at very similar FPS per my prior results?  Not sure I follow or subscribe to that logic?

Anyway, the objective is to demonstrate the relative performance difference with hardware … it’s nothing to do with future platforms, value, etc.  I’ll happily take a 45% improvement and more timeframe consistency.

 

Cheers Rob.

  • Like 1
Link to comment
12 minutes ago, Rob Ainscough said:

You gents need to be a little more negative … geez … anyway happy holidays!  
 

I do like UE5 but no one has done a flight simulator with it … it’s all about render distance (LOD). 

So I guess it’s end of road for MSFS and XP because they perform at very similar FPS per my prior results?  Not sure I follow or subscribe to that logic?

Anyway, the objective is to demonstrate the relative performance difference with hardware … it’s nothing to do with future platforms, value, etc.  I’ll happily take a 45% improvement and more timeframe consistency.

 

Cheers Rob.

me end of road comment wasn't meant to be negative in any way.  I just personally don't think we will see lockheed put any effort into selling home user sims anymore.  I'm not saying that's good or bad.  Either way my intent wasn't to derail and appreciate the scientific look you are giving us to the upgrade.  Can't wait to see what the new "x3d" AMD chip looks like when it's out.  

Link to comment
Rob Ainscough

Marketing effort, No.   Development effort, Yes.  But you know I can’t say more sooo…

From my perspective I see MSFS as restricted to whatever XBOX and WASM can or can’t do.  P3D has no such restrictions.  No restriction future is what I want from a flight simulator.

Rob

 

  • Like 3
Link to comment
Wilhelm Zwirchmayr
5 hours ago, Alex Pugh said:

$580 processor and a $1,600 video card gets you..... 34 fps. lol. Ready for the switch to UE5, if that's actually happening for V6. 

I also read this message a few days ago.
once they start, it will take years, possibly version 7

Link to comment
Wilhelm Zwirchmayr
11 hours ago, Rob Ainscough said:

And here is some data for 7950X using same scenario (Apples to Apples) in P3D:

P3D-EGLC-EGLL-737-8-7950X-nv4090_Totals.thumb.jpg.47d2402b32b2f4cfc31edc4ba143f388.jpg

P3D-EGLC-EGLL-737-8-7950X-nv4090_Chart.thumb.jpg.ef7d4b51c6008b73d5524384c6775fde.jpg

 

P3D-EGLC-EGLL-737-8-7950X-vs-5950X-nv4090_Chart.thumb.jpg.95be3233ec6a658a6356839a7ec62846.jpg

In this relatively extreme scenario of add-ons and graphics settings the:

AMD 5950X (nVidia 4090) was 23.4 FPS AVG 

AMD 7950X (nVidia 4090) was 34.0 FPS AVG

Performance improvement going from an AMD 5950X to AMD 7950X is 45.2%

As you can see from the data, overall throughput is improved with much less variance and fewer long spikes in FPS.  Overall, this is probably the biggest CPU improvement I've seen in a while going from prior gen CPU to current gen CPU.

For those interested in the data files they can be found here:

Raw RTSS logged data.

Converted and Charted RTSS data to Excel.

Here is the video of the test flight (give YouTube some time to resolve it to 4K resolution):

This weekend I hope to be able to run the same test in MSFS for another Apples to Apples compare between the 5950X and 7950X AMD CPUs.

Cheers, Rob.

 

thank you very much for doing these tests and has my full respect

it would be impossible to fly with this setting with my system ( 3080 and AMD 5900 )

could you repeat this test with different weather settings
e.g. fog, rain, night
I think that's where the 4090 starts to sweat and runs into the limit

I have the same problem with the 3080 at higher settings

Link to comment
Rob Ainscough
34 minutes ago, Wilhelm Zwirchmayr said:

could you repeat this test with different weather settings
e.g. fog, rain, night

Oh you want a typical English day ... hehe ... yes, but first I'll do MSFS test (probably this weekend time permitting as these test take a while to setup correctly and perform) so I can compare 5950X vs. 7950X in MSFS.

This specific test is focused on what I believe to be one of the most demanding locations and very demanding add-on aircraft with demanding add-on airports.  As you can see from my other tests in less demanding locations the FPS increases significantly ... ex: Islands like Tenerife in Carenado aircraft using the same graphics settings is over 100 FPS.   

I plan to do another test with AI traffic loads to see how that impacts the CPU ... I know there are several users who enjoy significant amounts of AI traffic which tends to be CPU bound as the AI models aren't very poly intensive.

  • Like 1
Link to comment
  • 2 weeks later...
Rob Ainscough

Finally got around to doing MSFS test for the AMD 7950X:

MSFS-EGLC-EGLL-737-9-7950X-nv4090_Totals.thumb.jpg.2873efec2201658f590f87b687a25fe3.jpg

MSFS-EGLC-EGLL-737-9-7950X-nv4090_Chart.thumb.jpg.d5e631e5837c50dde71744f5b21001df.jpg

MSFS 5950X -4090 = 25.4 FPS
MSFS 7950X -4090 = 38.5 FPS
51.5% increase

Now here comes the strange part, I tested MSFS with DLSS set to "quality" and "UltraPerformance" (both were about the same) and FPS was actually lower and image quality was not great either (blur any display with rapid number changes and "The O2" looked very blurred as did runways)??

MSFS-EGLC-EGLL-737-9-7950X-nv4090-DLSS_Totals.thumb.jpg.ac99ab9f383104e8a7120c3f36b85128.jpg

MSFS-EGLC-EGLL-737-9-7950X-nv4090-DLSS_Charts.thumb.jpg.5f3bf175990396808a84d1eab3e93c40.jpg

MSFS 7950X -4090 DLSS OFF = 38.5
MSFS 7950X - 4090 DLSS ON = 35.6
7.5% decrease

Not sure what to say here, I'm running the latest drivers from nVidia 527.37??

Here is MSFS 7950X - 4090 with DLSS OFF (both 4K res native):

Here is MSFS 7950X - 4090 with DLSS ON (both 4K res native):

I also ran some tests with DX12 vs. DX11 in MSFS and will post those later.

Cheers, Rob.

 

  • Like 1
  • Thanks 3
Link to comment
Wilhelm Zwirchmayr
7 hours ago, klaus legrand said:

Great piece of work @Rob Ainscough.

I am considering myself upgrading my rig in the next few months. Do you consider the new AMD processors better for sim than the likes of i7-13700k ?

hello because of Rob Ainscough I'm testing and have come to the following conclusions.
I have a 5900 x 4.5 GHZ, 32 GB Ram, 3080 RTX 10 GB wqhd 33 FPS fixed
at higher settings and bad weather - fog, high cloud density, rain etc.
the graphics card is very demanding and the FPS suffer and my 10 gb vram are running out
my CPU gets bored at larger airports with FSLABS

Every now and then my CPU has 100% utilization but only briefly and maybe 1-3 fps less but the graphics card is the bigger issue

I have the same issue with MSFS 2020 and my 10gb vram is really a problem

You should have at least 32 RAM installed

Link to comment
Rob Ainscough
8 hours ago, klaus legrand said:

Do you consider the new AMD processors better for sim than the likes of i7-13700k

I have no experience with the i7-13700K, sorry.

59 minutes ago, Wilhelm Zwirchmayr said:

I have the same issue with MSFS 2020 and my 10gb vram is really a problem

You should have at least 32 RAM installed

Agree, as per my testing with very high graphics settings and dense location with complex (ish) aircraft:

  • P3D I'd recommend min 16GB VRAM and 32GB RAM
  • MSFS I'd recommend min 20GB VRAM and 32GB RAM

On a side note, I tried using MSFS DX12(beta), "Frame Generation", with and without DLSS, Latency to set to Ultra, and Hardware Accelerated Graphics enabled and was greeted with HORRIBLE long pauses every 2-3 seconds along with significant input latency.  I'm not sure if this is an AMD CPU/Chipset issue or something else, but it makes the use of DX12 with HAGs and frame generation impossible in MSFS on my setup.  HAGs doesn't seem to impact P3D, get a slight (very slight) increase in performance. 

Is MSFS ever going to move to native DX12 and get out of it's DX12 "Beta" designator?  Having to run a 13 year old DX11 API from 2009 just doesn't seem fitting for "next generation" anything.

Cheers, Rob.

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...