if you have a rtx 3080 as you say you do . you should check your pc then .. the game runs at 30 fps on a ryzen 1200 8 gigs and a gtx 1050 ti .
Hows performance after patch 11 ?
- Thread starter [VR] Garfs
- Start date
For me, pre-U11 was solid everywhere (60fps as I limit to that) with the exception of fires in settlements. Then I started getting stuttering and drops in rates. Post U11, that seems to be all OK now.
One caveat: intermittently, when I exit the concourse to go back to my ship in thne hangar, I'll see a 50% loss in frame rate.
(sys specs in sig)
One caveat: intermittently, when I exit the concourse to go back to my ship in thne hangar, I'll see a 50% loss in frame rate.
(sys specs in sig)
EDO is proving more dependent on memory performance than I was expecting. Past tests showed a non-linear scaling with CPU clocks (in non-GPU limited scenarios), so I was anticipating some improvement, but the new memory I've put in my main system is helping enough to see/feel without having to benchmark it (though the difference is readily apparent in benchmarks) and my old memory was not slow. Makes me think even more that a good LGA-1700 DDR5 setup or the soon to be released 5800X3D could be the system to have for this game.
I don't think this is new to U11, just an observation in general.
I don't think this is new to U11, just an observation in general.
If I look at my memory usage during Odyssey, it does seem to tend to dump a lot of it when for instance you go from a settlement to supercruise or when you log out to the main menu. Now I'd guess most games do this to some extent, but this game seems to be pretty extreme. Is all the additional I/O really necessary and are there some gains to be made there as well?
On a sidenote, U11 or the hotfix introduced a new(?) bug where the various interactive vendors are absent from surface ports (haven't observed this in stations yet). All the other generic NPCs are still present, but somehow the FPS is higher (10-20 FPS increase depending on the location) without the handful of static vendors. And occasionally active settlements are completely abandoned as well. A relog fixes both.
Memory utilization, in the capacity sense, isn't a very good indicator of what the game is doing with that memory or how the performance of the physical memory will influence things.
What I'd like is to be able to measure the cache hit rates, but I can't find any useful Window's performance counters for this on most modern CPUs.
The alternative is to not aggressively evict assets. This might improve loading times in some areas, but would probably balloon main and graphics memory consumption. FDev probably decided that the loading pauses were a worthwhile trade-off.
I guess, but just feels a bit unnecessary to dump 2-3 GBs of settlement assets when I'm moving from settlement to settlement, especially if 90% of those assets are completely identical.
But very necessary for those GPUs with limited VRAM - (not me, I have 16GB, but others may not be so fortunate) - I suppose the code isn't written to take full advantage of assets.
16GB of VRAM? Bery nice! What GPU? Out of curiousity...
Radeon 6900XT. I replaced my GPU in December and that one was being offered for a 'reasonable ' price at the time (is currently about £50 cheaper on some etailer sites)
As it happens, this morning I was mulling over the Alder Lake options and the DDR4/DDR5 decision (to upgrade from a i7 4790k).
DDR5 seems to be a lot more available now, so within the next month or so I may have a new system - any recommendations for the DDR5 ram? The 4800MHz variety is at a reasonable price point, so not sure if it's worth going any higher (bearing in mind that the mobo will be mid-range).
Big caches and videocards having their own memory means it's unlikely RAM performance will give you anything real for extra dollars. The only thing I've seen RAM performance make a tangible difference to on its own is igpu performance. And in that case, it's like 10%.
But if it's just about upgrade room, I get it.
But if it's just about upgrade room, I get it.
I don't have enough experience with DDR5 to know much about the properties of the various ICs, PCBs, and binnings, but the performance sweet spot seems to be around 5200MT/s. Even mid-range boards support DDR5-6400+.
I've noticed the dip in performance when powering up settlements. It's quite significant, even without fires.
Anything refering to CPU or memory performance is refering to non-GPU-limited scenarios, which in my case, is anything up to and including 1440p ultra (at those settings my 6800 XT tend to bounce around 40-70% utilization in a high intensity CZ) that features populated settlements.
I went from this:
To this:
And gained a solid ~5% to average frame rates in the process, with a more noticeable improvement at the low-end. It was about the same difference as 400MHz+ of CPU clock speed on a Vermeer (Zen 3) part, which is a considerably larger differential than you'll see in the gaming clocks between, say, a 5600X and a 5950X.
The game still scales well with CPU clock increases, but improving memory performance seems to overtake CPU clock in importance after about 4.4GHz, on this platform. And this is all at the same FCLK, UCLK, and MEMCLK. The difference is correspondingly larger with a larger difference in the performance of the memory subsystem.
Going from some cheap 2400-3200MT/s, for example, to even budget 3600 stuff, then spending a little effort tuning it, would probably be one of the more economical upgrades that could be done for this game on AM4, if one already has a Vermeer. I suspect the situation is similar for any system that isn't GPU limited where a reasonably fast CPU has been paired with mediocre memory.
Last edited:
I was playing around with settings today. At one time I had really crappy performance <30 fps at a planet, so I opened up task manager, and found 11% CPU and 71% GPU. After some tinkering I actually found a new setting that gave me a pretty solid 45 fps in stations etc and 90 fps in space, in full res without any upscaling.
(Reverb G2, RTX 3070 OC, Ryz 3950x and 3200 DDR4)
(Reverb G2, RTX 3070 OC, Ryz 3950x and 3200 DDR4)
I don't have enough experience with DDR5 to know much about the properties of the various ICs, PCBs, and binnings, but the performance sweet spot seems to be around 5200MT/s. Even mid-range boards support DDR5-6400+.
I've noticed the dip in performance when powering up settlements. It's quite significant, even without fires.
Anything refering to CPU or memory performance is refering to non-GPU-limited scenarios, which in my case, is anything up to and including 1440p ultra (at those settings my 6800 XT tend to bounce around 40-70% utilization in a high intensity CZ) that features populated settlements.
I went from this:
To this:
And gained a solid ~5% to average frame rates in the process, with a more noticeable improvement at the low-end. It was about the same difference as 400MHz+ of CPU clock speed on a Vermeer (Zen 3) part, which is a considerably larger differential than you'll see in the gaming clocks between, say, a 5600X and a 5950X.
The game still scales well with CPU clock increases, but improving memory performance seems to overtake CPU clock in importance after about 4.4GHz, on this platform. And this is all at the same FCLK, UCLK, and MEMCLK. The difference is correspondingly larger with a larger difference in the performance of the memory subsystem.
Going from some cheap 2400-3200MT/s, for example, to even budget 3600 stuff, then spending a little effort tuning it, would probably be one of the more economical upgrades that could be done for this game on AM4, if one already has a Vermeer. I suspect the situation is similar for any system that isn't GPU limited where a reasonably fast CPU has been paired with mediocre memory.
I would not have believed you'd get as much as 5% from a memory overclock, even if it is a bit aggressive. I would have expected 2-3%. In the context of choosing a motherboard and ram over it and paying possible premiums for such choices, I still don't believe it's meaningful. The fact that RAM performance yields such significant numbers probably means something for the game. I dismissed the idea of attempting any RAM overclocking for Elite but you've convinced me.
Depends on what your budget is and what your goals are.
It doesn't take an expensive board to support fast memory, or facilitate memory tuning. Indeed, ITX boards, by virtue of almost always having only two DIMM slots (meaning point-to-point topology and short trace lengths), have a very strong tendency to handle memory well, and are rarely priced beyond the mid-range.
When it comes to the memory itself, it's certainly possible to spend way too much and there are extreme diminishing returns at the high-end. However, there is usually a price-performance balance point where cheaping out further won't be worthwhile either.
Then you get to the CPU and GPU. Most CPU lines plateau in per-core performance in the mid-range; unless one knows for a fact that their games can use an unusually large number of cores, spending more for extra cores is simply a waste, for gaming performance. It's not an entirely dissimilar story with GPU upgrades; most games at higher settings are indeed GPU limited (and if I'm building a gaming box, I won't hesitate to put upwards of half the entire budget in to the GPU itself...one of my systems has an $80 motherboard, $100 RAM, and a $1500 video card), but there are enough exceptions that blindly upgrading the GPU without looking at what's going on can easily turn into a waste...as we've repeatedly seen with Odyssey (I'm pretty sure I could dredge up a half-dozen posts on this forum that are some permutation of "I bought an RTX 3090 for my six year old system and my CZ frame rate still sucks!").
At this point, DDR5 is too expensive for me to seriously consider outside the extreme high-end, but if one is going to be getting a top GPU and CPU anyway, spending the extra on the DDR5 board and some fast-enough DDR5, that will only add ~10% to the total budget, is probably a wise move.
I was never a fan of buying small-form-factor boards or of skimping out on the motherboard in general. I despise RGB bells and whistles (more power to you if that's your thing) and 2 foot tall chunks of aluminum, but I want to know my VRM is handling things.
I agree DDR5 might be the right move if we're talking about a new build, that's going to be a platform going forward, but that's really the only thing that makes it relevant right now imo. I'd be real skeptical to pay more, even it's not-all-that-much-more comparative to a GPU or CPU, to help out with Elite because I couldn't get the GPU I wanted right now type of thing.
To what end though? the issues with odyssey performance are not low hanging fruit that's just a matter of proper caching or overzealous management of texture data. That kind of stuff would have been resolved by now if it was. They've been looking for such gimme fixes for 9 months.
what's being seen is likely a symptom of a problem that can't be corrected directly, regardless of identifying it as a problem behavior or a sign of things being done improperly. That kind of detailed identification of weird / nonstandard practices was made at release by analyzing the rendering frame call by call and it didn't seem to do any good.
Do you suppose this has something to do with why my game runs more poorly after being on for several hours? My FPS starts out perfect, but gradually declines until eventually I have to restart.