7900x3D/4090 Water Cooled vs. 5900x/3090 Water Cooled HP Reverb G2 VR

[trevoC]

Just starting a thread to post my results.

Old Build:

5900x Water Cooled (liquid arctic cooler)
Asus Hero MB w/ 64GB DDR4
3090 (not liquid cooled)
2 x 2TB Gen 3 NVME
EVGA 1200 watt PSU

New Build:

AMD 7900x3D Water Cooled (Arctic liquid 360mm rad)
Asus ROG Crosshair X670E Hero
64GB DC DDR5 6400 Ram
MSI Suprim RTX 4090 Liquid X (Liquid Cooled)
4TB Gen 4 NVME (Kingston Fury)
2x2TB Gen 3 NVME
Corsair HX1500i PSU
NZXT H7 Flow

I'll do some tests on the old system and post a screenshot of my settings so I can attempt a similar setup in the new system.

All results are in VR on the HP G2 Reverb (Gen 1)

Just waiting on the arctic liquid cooler before I can start my build, will run older system tests in the meantime. Not attempting a perfect scientific method comparison but will be as thorough as I have the patience to be to offer the community some semblance of comparison.

For context, my 3090 build is severely CPU bound which is why I didn't attempt to get a 4090 for it. Most of the time the 3090 is not over 89% usage and many times is in the 60% usage area while the CPU is full tilt (20% OC'd)

Not sure if the 7900x3D was the perfect buy over the 7950 but we'll see. Initial results show the two of them almost head to head, less than 1% difference (7950 edging it by a bit) but no DCS tests I've seen.

The reason for the 7900x3D over the 7950x3D is the higher base clock speeds which should result in better single core performance and because its a 6+6 chiplet design over an 8+8 design there should be some extra headroom (in regards to heat) in the 6 chip design to squeeze more Over Clocking out of it as the 8 chip design will generate more heat and DCS won't use them anyway.

Is my thinking correct on this, who knows, we'll find out soon enough. In the 5900x vs 5950x (I've run both) the 5900x was faster in my builds. Same was true in the 3900x generation. Either way, the difference would probably be insignificant. The real gains for me will be the extra CPU headroom over the 5900x for more draw calls to the 4090 which I'm hoping I can take much more advantage of than the 5900x did with the 3090.

Excited to build this, but not sure when I'll be able to get around to it. Too many projects on the go and this build requires me to build 4 pcs essentially as I have a rotation of parts as I build new machines. (my old machine some parts to my son, and my sons old machine 3900x back to me for the office and my office machine back to one of my daughters etc.. etc..) I love upgrade time, but it will be busy on the workbench for a bit.

Some parts sleeping under my desk until the cooler arrives.null

[EightyDuce]

Should be an awesome build. One thing to take care of is RAM, being 6400mt/s, you may have to dial it down to 6000 or 6200 as very few memory controllers on Zen4 can reliably hit 6400. Also, the XMP timings will likely be VERY conservative/loose.

Take a look at this video, your kit is likely Hynix A die (maybe M.die) so you should have no problems copying secondary, tertiary and Tras timings for a significant boost.

 

 

Note: only do that after you establish that the system works as it should stock config (minus dialing down the RAM from 6400 to 6000).

 

As for 7950X3D vs 7900X3D, I would personally would have opted for the 7950X3D simply because thermal headroom will be insignificantly different due to the IHS (thicc) and you still retain the same scheduling "obstacles" but you lose out on 2 extra cores for both v-cached and non cached CCDs, base clock is irrelevant as you looking to boost as much as possible and 7950 will boost higher; 7950X3D is extremely efficient with 120w tdp. $100 difference in price is arguably negligible at that point as well.

All that being said, I don't think you'll be disappointed with the 7900X3D and should be great build when done.  Just make sure you update your BIOS and chipset drivers to the latest offering as you incur a significant performance penalty otherwise.

Edited March 2, 2023 by EightyDuce
EDIT: Make sure to update your MB bios and chipset drivers!!!

[trevoC]

Great advice, thanks for your time.

7950 was also sold out.... they went quick but I figured it would be likey insignificant differences in DCS. Only other applications on this machine are MSFS and iRacing.

[dureiken]

can't wait to see your result, specially in CPU time in VR !

[Greekbull]

Nice I'm just about to head to bed after doing part of my new build.  I'm going from a 5950X, 4090(used to have a 3090), 64GB of 3600 DDR4, to a new 7950X3D, X670E Crosshair Hero, 4090 FE, 64GB of Trident Z CL30 6000 AMD EXPO RAM.  Hope to finish things up tomorrow and work on tuning it through the weekend.

[trevoC]

6 hours ago, Greekbull said:

Nice I'm just about to head to bed after doing part of my new build.  I'm going from a 5950X, 4090(used to have a 3090), 64GB of 3600 DDR4, to a new 7950X3D, X670E Crosshair Hero, 4090 FE, 64GB of Trident Z CL30 6000 AMD EXPO RAM.  Hope to finish things up tomorrow and work on tuning it through the weekend.

Would be interesting to do a test run with identical settings to see where we land. Would probably tell one of us if something was wildly off if one of us had a much higher fps.

Edited March 3, 2023 by trevoC

[dureiken]

DO you have F16 ?

[trevoC]

I do.

[dureiken]

I have this little track, where GPU and CPU are loaded, specially due to MFD and some units around.

If you could test ? maybe someone have better, but we need a track with CPU load

DCS F16 Perf test Syria SA6.trk

[trevoC]

Alright, I'll take a look this weekend on my 5900x machine and run a baseline with that track.

The baseline I've always used is the Huey at Marianas which nets me around 43 fps with my current settings near the docks... This is about half the fps I could expect from being in the clouds in an fc3 aircraft for example and around 60 fps in general play with full-fidelity aircraft. So many variables but I usually test at Marianas with the Huey and find that if it runs acceptably there it will run acceptably 99% of the time.

No problem running your track though.

[dureiken]

What is really important is to use something (FPSVR/OXRTK) to monitor gpu and CPU frame time and to save the datas !

[trevoC]

I was just going to record the now built in fps box in DCS.

i believe it shows CPU also.... at work and can't remember

[Th3ChosenOn3]

7 hours ago, dureiken said:

I have this little track, where GPU and CPU are loaded, specially due to MFD and some units around.

If you could test ? maybe someone have better, but we need a track with CPU load

DCS F16 Perf test Syria SA6.trk 93.85 kB · 1 download

I've been benchmarking my 5800X and RX 7900 XTX on Plazma's Torture Map, in preparation for comparing it to my 7950X3D next week.

The results are quite consistent from my experience. It's very heavy on the CPU, so you might see your GPU hover around 30-60% usage depending on area, settings and resolution. I've been doing the Benchmark with the High Preset on 1920x1080, 2560x1440 and 3440x1440 and they all end with this result due to a CPU bottleneck:

 

Plazma Torture Map - Level Flight.trk

[trevoC]

Alright, I'm here, downloading both tracks and will record with my own test (marianas and huey because thats what I'm used to) and will put a video together.

All parts have arrive except the arctic liquid cooler  so looks like it will arrive sometime this week (wed me thinks). Should give me plenty of time to get my files ready. Will run the tracks right now and put a video together of them.

[trevoC]

Alright,

Here's the benchmark on the 5900x | Asus Hero VIII | 64GB RAM | 3090 | Arctic Liquid Cooler | OC'd I think around 15% (CPU)

See settings at beginning. I don't always run the same settings, and these settings are definitely not optimized for the worst case scenarios. Usually if I was flying Huey on Marianas or a heavy map I'd bump viewing distance from Ultra down to High and visibility a bit lower (50% or something). This will usually net me 45 fps on Marianas (or close to) and around 60-90 fps in a standard ingress scenario at angels 20 for example.

Very frustrating to be CPU bound. Maybe I'll test the 3090 in the new rig to see if it can take advantage of the 3090 in full before popping the 4090 in. No promises as I really don't care, it would just be interesting to see if the 7000 series is enough for DCS to use a 3090.

If you are CPU bound its really important to set your baseline on a tough map or scenario that would represent the worst case scenario for you as if you test your fps in an instant action in a fighter at 30,000 feet you are bound to start pumping up all the settings when your hitting 100 fps until it looks pretty. First time missles and bombs start flying (looking at you reflected... just kidding, love your stuff) you begin a slideshow.

In these cases, these are pretty much as bad as it gets for me. If I set the Marianas map to hit 42 fps (my target is above 40) and then jump into an su-25 campaign I'm hitting 70-90fps most of the time and dropping to mid 40's when the action goes nuts.

Posted 3am EST, if you don't see 4k its probably not ready until 5am according to youtube.

I'll do as close to exactly the same video for the upgrade and then overlay them to see the specs side by side.

[dureiken]

NIce thanks for the work

 

maybe you could use OXR toolkit to have both gpu and CPU frame time and the best would be to do an .CSV to compare both CPU on excel !

 

thanks

[ISantus]

I am very much looking forward to these results. I am lucky in that the system I currently run is very clsoe to your baseline system (5900x, 3090, 32gb ram, G2 using OpenXR). I recently picked up the 4090 but haven't been blown away by the outcomes. Looking next to the 7...X3d upgrades and if they will be worth it. My hope is that either the x3d systems make a huge difference or the multithreading comes out before the results of the 7800x3d come out and they show drastic improvement (save me the money). Eventual system I am thinking is 7800x3d, 64gb ram DDR5 6000, 4090, OpenXR.

Thanks for your work on this. 

[trevoC]

7 hours ago, dureiken said:

NIce thanks for the work

 

maybe you could use OXR toolkit to have both gpu and CPU frame time and the best would be to do an .CSV to compare both CPU on excel !

 

thanks

They were running on advanced (openxr stats). To my surprise, they don't show up on the screen (where this is mirrored) only in the headset. OpenXR menu doesn't show up either as I went in there to show the settings.

[Hoirtel]

On 3/2/2023 at 10:51 PM, EightyDuce said:

Should be an awesome build. One thing to take care of is RAM, being 6400mt/s, you may have to dial it down to 6000 or 6200 as very few memory controllers on Zen4 can reliably hit 6400. Also, the XMP timings will likely be VERY conservative/loose.

Take a look at this video, your kit is likely Hynix A die (maybe M.die) so you should have no problems copying secondary, tertiary and Tras timings for a significant boost.

 

 

Note: only do that after you establish that the system works as it should stock config (minus dialing down the RAM from 6400 to 6000).

 

As for 7950X3D vs 7900X3D, I would personally would have opted for the 7950X3D simply because thermal headroom will be insignificantly different due to the IHS (thicc) and you still retain the same scheduling "obstacles" but you lose out on 2 extra cores for both v-cached and non cached CCDs, base clock is irrelevant as you looking to boost as much as possible and 7950 will boost higher; 7950X3D is extremely efficient with 120w tdp. $100 difference in price is arguably negligible at that point as well.

All that being said, I don't think you'll be disappointed with the 7900X3D and should be great build when done.  Just make sure you update your BIOS and chipset drivers to the latest offering as you incur a significant performance penalty otherwise.

 

I have been looking at this video, have you tried this on 64Gb kit? this video says its for single rank. All 64Gb kits will be dual rank

[evanswillo]

Thankyou for sharing your build, really interested to see your VR results, please could you tell us exactly what memory you have installed.

[Thinder]

1 hour ago, Hoirtel said:

I have been looking at this video, have you tried this on 64Gb kit? this video says its for single rank. All 64Gb kits will be dual rank

True, same for DDR4 but at 64GB, interleaving becomes near irrelevant, there is enough capacity per stick to compensate and the controllers can manage pretty much the same way...

[ironhard]

How much RAM can DCS use with 64GB under Win11?

I'm contemplating an upgrade from 32gb, but reluctant to upgrade because last gen tech.

Edited March 6, 2023 by ironhard

[some1]

1 hour ago, ironhard said:

How much RAM can DCS use with 64GB under Win11?

It depends. Best to check DCS memory usage on your system while playing the kind of missions you normally play.

On the extreme end, I have an "airshow" type mission with all different aircraft parked in one spot, and it barely loads with 64 GB. 

[EightyDuce]

10 hours ago, Hoirtel said:

I have been looking at this video, have you tried this on 64Gb kit? this video says its for single rank. All 64Gb kits will be dual rank

Yeah, I have a 2x32Gb Gskill 6000 32-38-38-38-96 kit I'm using. I started with his settings and now working from there. Below are two screenshots, one stock XMP and the other with BZ much improved timings. Overnight stable on TestMem5 anat777 extreme.

 

Last one is my current settings.

 

Edited March 6, 2023 by EightyDuce
added current settings

[trevoC]

For reference to above questions.

Old system (one video was taken on) is 4x16gb (low cl like 14 or 15 I think will have to check when I pull it apart to build sons computer)

New system is the following RAM but the build isn't compete and the above video was not done on the new system, it was just my benchmark to compare to.

https://www.newegg.ca/g-skill-64gb/p/N82E16820374432?Item=N82E16820374432