7800X3D, 7900X3D, 7950X3D..

[trevoC]

For those interested, I had zero problems and didn't follow any special instructions to get this working on a 7900x3D with game bar.

Just do the standard stuff.

Install AMD driver set after windows install then install DCS. (I made sure game bar was updated but that was the extent of it). Worked first time a charm. Not sure what all the fuss is about on youtube with 7x3D testers going mad about all the proper ways you have to follow to ensure it works.

I honestly didn't do a single thing outside what you would normally have to do.

[some1]

Changing DCS FPS limit to 400 fps also makes a significant change in test results for older CPUs. Below is my run on 5800X3D, 3600MHz RAM, RTX4090, mirrors off, 1080p and 400fps limit. Comparing to the results posted here, it looks like 7950X3D is faster than 5800X3D by 20% to 25%. At least in this particular test. 

 

Edited March 10, 2023 by some1

[okletsgo]

Who's going to repeat the benchmark with and without MT?

[Tom Kazansky]

Here is an interesting one by @phant

 

[tepl]

Plazma track  at 1280*768 (because of GPU bottleneck) and second track - F-18 launch on SuperCarrier (90 second)/ 5800x3d , RTX 3080

 

SuperCarrier.trk

[some1]

Same conditions as my previous test above https://forum.dcs.world/topic/318603-7800x3d-7900x3d-7950x3d/?do=findComment&comment=5168474

A solid FPS increase with MT and even a little bump without MT, but one part of the mission shows more stutter now than before the patch. Also without MT. 

Without MT:

With MT:

Previous Open Beta:

 

Edited March 10, 2023 by some1

[Th3ChosenOn3]

Haven't had time to do 5 benchmark aggregates, but the MT patch gives about a 30% uplift to averages, and a little bit to 1% lows. Almost cancelling the effect of having mirror and tacview on. I'm also running into a GPU Bottleneck at the end of the benchmark:

 

Also saw the same stutter bit as @some1 with Tacview and Mirrors enabled, not entirely sure what it is:

[sammich318]

I have a 7900x3d with a 3090 and 64gb of ram at 5200.  When i do the fps counter in game it says im cpu bound - main thread.  Ive done all the checks to make sure the unused cores are parked.  idk what else to do.  I dont think i should be CPU bound.  I tested on Syria f-16 free flight.  Motion reprojection enabled i get 30-40 fps.  MP off, I get 45-50..... IDK man, something isnt working right

[trevoC]

Also CPU bound according to DCS frame times and see the same fps across all 4 scenarios:

single threaded - parked cores (running on high cache cores)
multi threaded - parked cores
single threaded - no parked cores (all cores enabled)
multi threaded - no parked cores (all cores enabled)

All 4 scenarios are within 1 or 2 fps from each other. This is very strange behavior.... vcache seems to have zero effect along with multi threading.

this is with 7900x3D/4090/G2

no gains for me yet with MT or vCache.

[Hoirtel]

I would be grateful if anyone can help with RAM choice for a new X3D chip - as yet unpurchased but hoping to get a 7800 if I can be quick enough.

I have been looking for RAM kits and found some kingston ones that I could get but I am unsure what the best move is between these two:

Kingston Beast 64Gb 6000Mhz C36 - Says EXPO and XMP

Kingston Vengeance 64Gb 6000Mhz C32 - Only says XMP

The vengeance is a little bit more but not so much that its a concern. I am concerned about XMP/EXPO compatibility especially as I now own a Asus X670E Strix F and worried about the memory problems that lots of people seem to be having. I would of course like to avoid that. Neither are quite on the QVL list but the 32gb counterparts are. QVL list seems a bit old.

Even cheaper still are 5600Mhz kits at C36 or C40. As I understand these are all likely Hynix and perhaps could go the same speed. I have never overclocked RAM and would may the extra to have a decent setting without having to learn, but its something I will probably look into at some point especially as it seems there is perf to be gained. with Ryzen. 

Thanks to anyone that can spare some advice.

 

 

 

4 minutes ago, trevoC said:

Also CPU bound according to DCS frame times and see the same fps across all 4 scenarios:

single threaded - parked cores (running on high cache cores)
multi threaded - parked cores
single threaded - no parked cores (all cores enabled)
multi threaded - no parked cores (all cores enabled)

All 4 scenarios are within 1 or 2 fps from each other. This is very strange behavior.... vcache seems to have zero effect along with multi threading.

this is with 7900x3D/4090/G2

no gains for me yet with MT or vCache.

No gains, pre patch with vcache? Have you tried disbaling non vcache CCD entirely?

Edited March 13, 2023 by Hoirtel

[trevoC]

10 minutes ago, Hoirtel said:

I would be grateful if anyone can help with RAM choice for a new X3D chip - as yet unpurchased but hoping to get a 7800 if I can be quick enough.

I have been looking for RAM kits and found some kingston ones that I could get but I am unsure what the best move is between these two:

Kingston Beast 64Gb 6000Mhz C36 - Says EXPO and XMP

Kingston Vengeance 64Gb 6000Mhz C32 - Only says XMP

The vengeance is a little bit more but not so much that its a concern. I am concerned about XMP/EXPO compatibility especially as I now own a Asus X670E Strix F and worried about the memory problems that lots of people seem to be having. I would of course like to avoid that. Neither are quite on the QVL list but the 32gb counterparts are. QVL list seems a bit old.

Even cheaper still are 5600Mhz kits at C36 or C40. As I understand these are all likely Hynix and perhaps could go the same speed. I have never overclocked RAM and would may the extra to have a decent setting without having to learn, but its something I will probably look into at some point especially as it seems there is perf to be gained. with Ryzen. 

Thanks to anyone that can spare some advice.

 

 

 

No gains, pre patch with vcache? Have you tried disbaling non vcache CCD entirely?

 

No need to disable it... I had no problem with Game Bar assigning single threaded DCS to proper CCD with vCache.

I'm not alone... looks like vCache is not giving anyone any gains in DCS on 7x3D's

[Hoirtel]

11 minutes ago, trevoC said:

No need to disable it... I had no problem with Game Bar assigning single threaded DCS to proper CCD with vCache.

I'm not alone... looks like vCache is not giving anyone any gains in DCS on 7x3D's

Really, I had thought this thread said otherwise? Compared to what though? "standard" ryzen CPUs?  Is this with non MT version too? 

From your post - Have I misunderstood??

I think this may be the largest system upgrade jump I've ever taken, and I rarely miss an upgrade in CPU or Graphics generation. (I'll skip the TI variants of each unless I skipped the original)

I will still make the video when I get all my peripherals installed and the system overclocked and tuned and stress tested etc.... but here was what I got essentially.

My most common fps measure was around 64 fps but it would basically float around 63-73 fps so I'm guessing that would average somewhere in between. I may have witnessed more 63's than 73's so we'll say maybe 66 fps ish. My average 99% fps was higher than the last system's average fps and it never went below 40 fps. I saw as high as 90 fps when things were not as busy and the lowest fps I witnessed was 53 for a second.

Edited March 13, 2023 by Hoirtel

[Th3ChosenOn3]

2 hours ago, trevoC said:

No need to disable it... I had no problem with Game Bar assigning single threaded DCS to proper CCD with vCache.

I'm not alone... looks like vCache is not giving anyone any gains in DCS on 7x3D's

I've done a run on the Normal CCD of the 7950X3D, and got an average of 120, after I fixed my game bar issue and DCS was running on the 3D V-Cache CCD, my average jumped to 170 pre framecap update, and over 200 after the frame cap update.

You should see substantial gains if you're on the 3D V-cache CCD, unless you're being bottlenecked somewhere else.

Edited March 13, 2023 by Th3ChosenOn3
You're

[Voyager]

Does anyone know how sensitive the X3D chips are to ram speed here? I recall the 5800X3D more or less did not care about ram speeds. 

[trevoC]

6 hours ago, Hoirtel said:

Really, I had thought this thread said otherwise? Compared to what though? "standard" ryzen CPUs?  Is this with non MT version too? 

From your post - Have I misunderstood??

I think this may be the largest system upgrade jump I've ever taken, and I rarely miss an upgrade in CPU or Graphics generation. (I'll skip the TI variants of each unless I skipped the original)

I will still make the video when I get all my peripherals installed and the system overclocked and tuned and stress tested etc.... but here was what I got essentially.

My most common fps measure was around 64 fps but it would basically float around 63-73 fps so I'm guessing that would average somewhere in between. I may have witnessed more 63's than 73's so we'll say maybe 66 fps ish. My average 99% fps was higher than the last system's average fps and it never went below 40 fps. I saw as high as 90 fps when things were not as busy and the lowest fps I witnessed was 53 for a second.

 

You haven't misunderstood.... the 7x3D/4090 combo is the largest jump in performance I've witnessed in DCS.

What I have learned since with more benchmarking than I care to remember over the last 5 days is that the vCache doesn't make a difference.

The package is still a good package. I'm wondering if the standard Ryzen 9 non vCache would be identical results.

If I run on vCahc ccd or not, same fps, Single threaded or MT, same fps....

I'm not saying this is my final conclusion ATM about vCache, but I'm not alone on this one.

If you notice, the benchmarks here, you'll notice that they have also come to the same conclusion (not in this thread) and many didn't notice the cores not parking correctly on MT executable.

I've discovered a lot in the last 5 days but unfortunately, I have more questions than answers after all of it.

I have hours of video of all those benchmarks and don't even know where to start as the data is strange and the conclusions are few.

On that note, my system is extremely well configured (I've been doing this for 30+ years) and I have some great Memory speeds dialed in and the Ryzen cores are OC'd to about 5350mHz on vCache cores and 5750mHz on others now. Figure I can get that to 5500/5900 with some more work stable.

I have some more testing to do but will probably just play some DCS for a bit.... it is still much nicer than my last setup, just bugging me when something doesn't makes sense.

4 hours ago, Th3ChosenOn3 said:

I've done a run on the Normal CCD of the 7950X3D, and got an average of 120, after I fixed my game bar issue and DCS was running on the 3D V-Cache CCD, my average jumped to 170 pre framecap update, and over 200 after the frame cap update.

You should see substantial gains if you're on the 3D V-cache CCD, unless you're being bottlenecked somewhere else.

 

I find it hard to believe that all things were equal between those two tests.

Nowhere will vCache give you 75% gains.

Edited March 13, 2023 by trevoC

[trevoC]

7 hours ago, Hoirtel said:

Really, I had thought this thread said otherwise? Compared to what though? "standard" ryzen CPUs?  Is this with non MT version too? 

From your post - Have I misunderstood??

I think this may be the largest system upgrade jump I've ever taken, and I rarely miss an upgrade in CPU or Graphics generation. (I'll skip the TI variants of each unless I skipped the original)

I will still make the video when I get all my peripherals installed and the system overclocked and tuned and stress tested etc.... but here was what I got essentially.

My most common fps measure was around 64 fps but it would basically float around 63-73 fps so I'm guessing that would average somewhere in between. I may have witnessed more 63's than 73's so we'll say maybe 66 fps ish. My average 99% fps was higher than the last system's average fps and it never went below 40 fps. I saw as high as 90 fps when things were not as busy and the lowest fps I witnessed was 53 for a second.

 

I think some of the frustration you may feel in some of my posts was the let down of being super excited of the new setup (my first account of it that I posted) and then no sooner than I experienced that high and MT was released and I'm thinking this is the best week ever expecting even more gains from MT (I'd have been happy with any gains here) and being let down when no amount of testing will show me gains in this area. I had just finished testing that night and noticed the update but didn't assume it was anything more than bug fixes so I didn't update DCS so that my benchmarks could be as close to apples to apples as possible.

Went to bed and woke up and realized it was the MT update as the forums had exploded.

Obviously EFF apples to apples I want me some MT. A little disappointed that day and have been confused ever since.

It bugs me because I can spend hours squeezing 1 or 2% more out of my overclocking efforts from my prior setup but if something is seriously wrong or out of place that is where you want to focus first as there might be some serious gains to be had if I can fix what might be a major issue.

To be crystal clear, I'm super stoked about the R9/4090 setup.

I stress-tested my last OC on memory/CPU and I never saw temperatures above 45-50c on CPU/GPU under full load and the fans only kick on once and a while. I definitely have tons of thermal headroom here but I will be capped by the power limits on the GPU and stability limits on the CPU.

[BitMaster]

Are you talking about the "same" DCS at all ?

According to your both signatures, one runs VR and the other runs 2D... that may explain the different experiences.

Also other DCS settings may effect your possible gains as well as add-ons, 3rd party modules etc...   and maybe not having deleted fxo and metashader, there are many variables to be considered when you "just compare".

 

I didn't get more fps, heck..the 1080ti is my bottleneck, but in 10+ years it has never been that smooth.

just my 2 cents

[Th3ChosenOn3]

4 hours ago, trevoC said:

I find it hard to believe that all things were equal between those two tests.

Nowhere will vCache give you 75% gains.

 

The 120 average fps run on the non 3D v-cache CCD was first run after getting the system up and running, and noticing that DCS was running on the incorrect CCD and so there were many things unoptimized. I'm just mentioning it as it was the only experience I've had on the non 3D V-Cache CCD, but I never used it in any of my charts or data. After fixing game bar, and seeing DCS run on the 3D v-cache CCD I started gathering data.

It's also worth mentioning that my 60-75% uplift claim is from upgrading my 5800X to the 7950X3D.

[trevoC]

5 hours ago, BitMaster said:

Are you talking about the "same" DCS at all ?

According to your both signatures, one runs VR and the other runs 2D... that may explain the different experiences.

Also other DCS settings may effect your possible gains as well as add-ons, 3rd party modules etc...   and maybe not having deleted fxo and metashader, there are many variables to be considered when you "just compare".

 

I didn't get more fps, heck..the 1080ti is my bottleneck, but in 10+ years it has never been that smooth.

just my 2 cents

In zero situations will vCache give you 75% gains. There are no things to consider. Lets move on.

1 hour ago, Th3ChosenOn3 said:

The 120 average fps run on the non 3D v-cache CCD was first run after getting the system up and running, and noticing that DCS was running on the incorrect CCD and so there were many things unoptimized. I'm just mentioning it as it was the only experience I've had on the non 3D V-Cache CCD, but I never used it in any of my charts or data. After fixing game bar, and seeing DCS run on the 3D v-cache CCD I started gathering data.

It's also worth mentioning that my 60-75% uplift claim is from upgrading my 5800X to the 7950X3D.

Is that worth mentioning now after responding to me in regards to considerable testing I'm trying to give to the community that I'm crazy and should see massive gains in the form of vCache from CCD1 to 0. That is what you said.

We'll move on, but this helps noone.

null

[hhhttt]

May be DCS is hard coded with 8 core in mind because in MT FAQ it keep refering P core and E core. As we know 13900k have 8 P core, but 7900x3d only have 6 core on vcacheccd.

If DCS is coded incorrectly to always use up to 8 core, rather than aware of 7900x3d's 6+6 situation, it may mess things up the speed as cross ccd is somewhat slow.

 

 

[blkspade]

Something to be noted is that game bar doesn't technically do any scheduling. It basically only engages the core parking so that scheduling shouldn't be needed, because the other CCD is then ineligible. Games can/will jump CCDs if other things wake the cores and make them eligible. Probably won't be an issue in most cases, but if you were to willfully use the other CCD for anything intensive enough you may want to set the affinity. You can alter program shortcuts to launch with a desired affinity though, and the masks for 16 core is FFFF for CCD0, and FFFF0000 for CCD1. So you can place cmd /C Start /Affinity FFFF " " in your shortcut target line before the path to the exe. EG:

cmd /C Start /Affinity FFFF " "  "C:\Program Files\Eagle Dynamics\DCS World OpenBeta\bin-mt\DCS.exe"

cmd /C Start /Affinity FFFF0000 " " "C:\Program Files\obs-studio\bin\64bit\obs64.exe"

All quotes needed.

[trevoC]

I'll address a few things after another few hours of benchmarking.

Part of the problem with anyone doing a 7x3D review is that they basically built and started benchmarking the same day or a few before MT dropped. This definitely started to add to the confusion.

This confusion was also had by me who right before MT dropped I got the system running long enough to just launch DCS and take a flight. I immediately noticed around an 80% improvement (sometimes closer to 100%) in frames based on a prior benchmark on the old system and I wasn't CPU bound but instead (for the first time in a very long time as my 3090 slept for most of its life) I was GPU bound. Remember, this is just the 7900x3D/4090 upgrade.

I only had time for that one flight and then went to sleep. I had seen the update for MT but didn't know it was MT at the time and was trying to keep everything apples to apples.

This is a screenshot from the Single threaded DCS (what I saw before going to sleep):

Notice I'm GPU limited, all cores are parked nicely and I've really not attempted any optimization for DCS in software yet (windows settings) but have just got my system RAM nice and stable at 6000 and almost 100,000MB/s read and write speeds and got the Ryzen vCache cores and work cores overclocked to a nice 5375/5800 respectively with lots of thermal headroom (I never really see over 60c)

nullI was extremely happy with these results but when I woke up and realized it was the MT update dropping, honestly... Eff apples to apples, who could resist. downloaded it and started playing with it immediately.

The first thing I noticed is that the CPU was now constantly bound according to the data.

That didn't make sense to me as how could spreading the load over more cores when I was GPU limited make me now CPU bound?

On top of this the parked cores go wild regardless of what you do (game bar on/off affinity tricks etc..) The best I could do was disable parking completely in the registry to prevent the potential latency of them hopping around... this improved me a few percentage points.

On top of this, I have a Huey Marianas track that I run to optimize my play. Basically, if I can get acceptable frames on that track, the rest of my play (I usually play campaigns) is a no-brainer. With the 3090, I'd target no lower than 42 fps on the Marianas in the Huey which would give me easily 60-80 fps in something like an su-25 on another map.

When I ran my Huey track, I couldn't see any gains between ST/MT.

On top of this, you are now jumping benchmarks all over the place between before and after and ST/MT and weird core parking and behaviors in MT etc...

As you can see, MT runs were a hot mess... weirdly parked cores, CPU Bound with GPU at 100% sometimes and other times at 70% depending on the run. No problem with the hardware though... nice cool running overclocked 4090... the first comment about this setup was good... its amazing. Its the MT update that started making me loopy.

I have said a lot as I've gathered information to share with other people who are trying to figure it out also, but I stopped it all today and started going through this systematically.

Back to basics. and there are a lot of test to still do.

The focus tonight was only on one thing. Am I seeing gains between ST/MT version of DCS.

Here are some of the conclusions after running two different track files (my Huey and Plazma torture map) in a systematic fashion over way too many hours and logging the data and comparing it.

Conclusions/Comments:

- Core parking works perfectly (game bar on or off) in ST

- Core parking is whacked in MT regardless of dozens of settings (game bar, registry, affinity etc...)

- I'm GPU limited in ST and not sure what the eff I am in MT. I would assume GPU limited also but the cores are doing some pretty weird stuff and later conclusions confirm I'm still GPU limited even though DCS would like to tell me otherwise.

- the 7x3D/4090 combo is killer in DCS

- There is something weird happening with a soft FPS lock of some type... fps should float between whatever numbers the system can get CPU or GPU limited doesn't matter, but all of a sudden it will lock at 60 fps.... then float again, then lock at 90 fps... seems to even lock in the 75 regions sometimes but that might be when its floating there steady. I can't figure this one out. It will lock at 60 fps and CPU nor GPU are working hard, not CPU limited, frame times are basically a flat line at the bottom, not cpu bound and GPU in the 65% range and it will just sit at 60 fps, then start floating freely again and I'll be GPU limited... not sure what is happening here.

and finally:

- MT is absolutely faster on my system than ST. ST seems to look more stable with the charts as MT looks like a hot mess with stuff parking and unparking and CPU and GPU bound but not... not sure what the hell is happening but here are my results:

HUEY Map is mostly what I would consider single-player campaign-type flying. I saw gains on this track file between 1 and 7% while flying it. These gains averaged to about 4 maybe 4.5%. Hey... gains are gains, I'll take it. This translated into 3 or so fps on average.

Plazma's torture map was another story. The low gains on that map were 10% and the high gains were 40 or more. I averaged around 22.63% gains across the entire track file but the real story would be those moments that would normally have brought your smooth-running sim to a crawl will probably start to become eliminated or at least wildly buffered now. Sort of just like ED said they would, so honestly, kudos to @BIGNEWY and @Eagle Dynamics team. They said from the start this wasn't a magic fps bullet (which its not) but hot dang if it doesn't look like it will take care of the worst offending moments in DCS. I can't wait to replay the p-51 bombers dropping bombs mission again.

Next, I want to see how much an effect the vCache has over the higher clock speeds. Now that I've settled the ST/MT I'll probably use the ST version to run on both chips to see where the advantages lie and also disable core parking completely and just open the chip up to let run on whatever cores the scheduler wants without any parked cores.

I'll leave you with two screenshots of the Plazma torture map in ST and MT both same spot.

Single

Multi:

[trevoC]

Once I figure out how much an advantage vCache does or doesn't have I can figure out whether its worth it or not to try and properly park the cores in MT and get it running on the vCache or just leave it be. If I'm losing huge gains by not using vCache I want to figure it out. If I'm not... I'm going to move on.

You can see some of the other items in the bottom left of the above images that I want to benchmark the effect of.

Hope this helps... have a good one.

At some point I'm also just going to call a spade a spade and go back to playing and truly enjoying DCS for the first time in VR. How pretty the game is was really missed with my jump to VR on a 1070 ti originally.... I think finally I've got that back.

[some1]

3 hours ago, trevoC said:

HUEY Map is mostly what I would consider single-player campaign-type flying. I saw gains on this track file between 1 and 7% while flying it. These gains averaged to about 4 maybe 4.5%. Hey... gains are gains, I'll take it. This translated into 3 or so fps on average.

Both your Huey screenshots show GPU pegged at 100% or close to that. No amount of CPU will make a significant change in FPS in that scenario. 

I wouldn't pay too much attention to DCS counter showing "CPU bound". Even if it is not bugged, it's only what the game sees, without VR software overhead. 

As I explained here, FPS is not a good indicator of CPU performance in VR, you should look at CPU frametimes.

 

[Th3ChosenOn3]

23 hours ago, trevoC said:

I'm not alone... looks like vCache is not giving anyone any gains in DCS on 7x3D's

3 hours ago, trevoC said:

Once I figure out how much an advantage vCache does or doesn't have I can figure out whether its worth it or not to try and properly park the cores in MT and get it running on the vCache or just leave it be. If I'm losing huge gains by not using vCache I want to figure it out. If I'm not... I'm going to move on.

You claimed v-cache doesn't give anyone any gains, and now you're saying you don't know if v-cache has a performance improvement or not. It might be beneficial to figure that out first before making such claims, as what you're doing doesn't help anyone.

 

3 hours ago, trevoC said:

Notice I'm GPU limited, all cores are parked nicely and I've really not attempted any optimization for DCS in software yet

3 hours ago, trevoC said:

The first thing I noticed is that the CPU was now constantly bound according to the data.

You shouldn't be testing for a CPU bottleneck in a very GPU bound situation like you're doing on Marianas. I'm also hoping you're using an application to capture data to analyze instead of just looking at a hardware monitor and Nvidia's performance overlay.

Edited March 14, 2023 by Th3ChosenOn3
you're