Tightening the tertiary timings could probably push latency to near 55ns, but something I haven’t bothered doing yet. Some of the high end kits running DDR5 7800 likely have significantly improved latency, but getting anything to run above 7000 Mhz is difficult on a four DIMM slot Z790 motherboard. The memory controller on Raptor Lake also starts having issues above DDR5 7400.
They’re definitely improving. I bought this Gskill 2x32 GB DDR5 6400 CL32 kit in November for nearly $500 - it’s now down to near $350 last time I saw. Some of the ulta high-speed DDR5 kits are also now a lot more affordable, but still at a hefty premium.
With the design of the CPU, I’m thinking about getting a shield at the very least, simply for cleanliness. The anti-benching bracket for 13th Intel’s does seem to have a real effect in also reducing CPU temps, so I’m curious if the AM5 version would help in that regard too. Many Amazon reviews claim a about a 5 degree temp reduction, but I often take those reviews with a bucket of salt.
Since AMD moved to the LGA socket design as well, this looks like it’s the same concept as the LGA1700 contact frame. I don’t personally have one, but I know most people are claiming a 5-10 degree improvement. I would expect something similar for AM5.
I have another 7950x3d data point for those looking for more ‘real world’ performance graphs in the sim.
This is a 5 minute capture in the default A320, departing London City airport, westbound towards Heathrow, flying right up the Thames at 1000 feet and 300 knots. When I run the sim in pancake mode (3440x1440) I lock the frame rate to my monitor’s actual refresh (59.94 hz) using RTSS. It eliminates almost all vsync mismatch micro stutters. I’m not interested in super high frame rates as much as smooth and stutter free flying.
All settings maxed except TLOD at 200, and no AI traffic (too unpredictable). Clouds on ultra and using broken clouds preset. Nvidia 4090.
You can see the first small and very brief FPS dip from loading Canary Wharf, the second around the houses of parliament / London Eye, and the final one approaching Heathrow. The worst drop was to 56 fps for a fraction of a second. The rest was remarkably smooth, especially for a near worst case low altitude photogrammetry plus airliner scenario like this.
@DeltaPrime0223 RivaTuner Statistics Server. It allows for frame rate locking down to 3 decimal places and offers many other refresh rate fine tuning settings. It’s the most precise frame rate capping utility there is… that’s easy to use anyway, hehe.
Is anyone running the 7950x3d on a 240mm AIO, or (better yet) a mid-grade tower cooler? I ran across this video, and sounds like the cooling requirements aren’t that bad at all.
I had been wanting to wait for the 7800x3d, but I can’t deny being a bit envious of those already on this CPU. I don’t think I’d need the extra cores right now, but who knows if I’ll end up with extra stuff to run later on down the road to make use of them.
FWIW, a German PC Gamer publication did a MSFS benchmark showing a simulated 7800X3D and (unannounced) 7600X3D, by turning off the high frequency CCX in 7950X3D and 7900X3D.
Of course, this is to be taken with a grain of salt given 7800X3D is not exactly the higher end chip with one CCX turned off, and RL results can vary depending on selected settings, etc…
@kido007dz Those results line up exactly with my observations in the sim.
In my EGLC test scenario, sitting on the runway facing west, low resolution and frame rate unlocked, so CPU limited…
CCX 0 only: 86 fps (balanced power plan)
CCX 0 and 1: 93 fps (high performance power plan)
When FS is allowed to use all cores it still goes straight to the vcache ones first, but it also uses 2 cores from the other CCX. Whatever it’s doing there makes a difference. No difference in temperatures observed either.
I have also tested running the sim with SMT turned off and I can see no appreciable difference in frame rate performance, maybe 1 or 2 fps at most, within margin of error, and no difference in smoothness/stutters. What’s strange is that disabling SMT on my 7700X gave around a 10% performance boost and it ran noticeably smoother. Temps on both CPUs in the sim were about the same, around 50 - 55 deg C, so there was no thermal throttling.
These X3Ds are strange but excellent beasts! Also this is still the first iteration of the chipset drivers and scheduler mods. I expect we’ll see some mild performance improvements over time as they get refined.
SMT off won’t increase performance much if at all on newer PCs but assuming you aren’t crowding out your cores with addon and apps it will certainy help with the lows on non XD3 rigs. Faster caches on X3D will do that anyway and I doubt you would notice too much so just leave SMT enabled.
@DensestSnail693 Yup, that makes sense. Historically FS hasn’t played very nicely with logical core scheduling in the stutter department, on either intel or amd, but on the X3D it seems the extra cache is smoothing out any collisions, or maybe the new scheduler is just a lot better designed, or both.
Welp, I did it. I decided to pull the trigger on a new “money is no object” build with a 7950x3D and a MSI Suprim 4090. This is going to be a pretty major step up from my current rig.
That’s odd, I found a video on Youtube and the “7800x3d” performs better than the 7950x3d. And he’s using the 6000mhz ram. Also hardware unboxed tested on 12 games and the average of the “7800x3d” was also better.
Disregard the language, I didn’t find any in english until now:
Again, “simulated” 7800x3D doesn’t means it will performs like this. The most probable thing is that the scheduler still needs some refinement (as we can see in some titles were the CCD is wrongly associated). As well as the AMD 3D V-cache CCD controller.
Also we don’t know in what situation the reviewers were running their benchmarks. Chipset updates and particularly release version of BIOS updates weren’t released until 7950x3d/7900x3d launch day in many motherboards.
Apart from that, 4% performance difference on average in favour of the “simulated” 7800x3D, could be treated as marginal.
IMHO Is not that hard to just wait for the real life 7800x3D reviews instead of promoting those reviewers that literally just clickbait their videos. I’m not saying this one you mentioned does but, there are many out there that do.
Exactly it’s just conjecture. Let’s wait and see with the real one. For me the 7800X3D is enough of a saving that makes it interesting if it is close to the others. That said if the 7950X3D turns out to be that much better then I’ll opt for that. We’ll know soon enough.
My prediction is that the 7950x3d will outperform the 7800x3d by about 3-5% on average in most games, and 6-9% total in MSFS specifically, under ideal conditions.
As far as I’m aware the max single core boost frequency for the vcache ccx on the 7950x3d is 5.2 ghz, and on the 7800 it’s 5 ghz. It’s a small difference, but still there. On my system running as a “simulated” 7800 produces 86 fps in my cpu limited test, and allowing it access to both CCXes produces 93 fps. FS consistently uses about 40% of the first two cores on the second die if allowed, with noticeable effect.
No idea yet how curve optimizer tweaking will play out between them. I don’t think thermal throttling is going to be a factor for either though as even in the most demanding scenery I have yet to see any temp higher than 65 C.
You don’t need to be an engineer to realize the 7800x3d and 7950x3d have similar features in common. Hence why it is “simulated” disabling the CCD1. That said, it is a conjeture, not a reality. So time will tell. Both probably will be nice options for MSFS once able to compare.
]
Have a nice day 
