LuftManu Posted November 13, 2024 Posted November 13, 2024 Hello guys! After my trip the USA I also got one 9800 X3D. Will do the testing this week @chiliwili69 Regarding the 285K, it was a total flop. I DO NOT RECOMMEND THE ULTRA SERIES. Had to return it as also probably was a defective unit. Overpriced and defective. Won't be getting it atgain for tests. Initial tests tho showed a decrease of 30% in Power usage but also a regression. Just clock down your gen 13 / 14. The most interesting part was the E-Core performance. In MSFS I managed to get 50% less W and almost 1/3 of the heat, but at lower FPS too. Not worth it, even at high resolutions. Nothing a undervolted 14 gen can do. And also, 100W while gaming is not that much. I even got those marks while using my 14900KS and getting top performance. PS: Maybe early sensor info, but I didn't like those V peaks, but it was OK in tems I guess! Here is some 3DMark Benching, I ended up 4th with the 4090 and 285K combo after some Ring OC and more tuning, but NOT MUCH improvement in games. Maybe with CU-DIMM and massive work you will be able to have enough performance, but not to justify the price. (And worse than 14 gen or X3D parts). https://www.3dmark.com/search#advanced?test=fs P&cpuId=3356&gpuId=1509&gpuCount=0&gpuType=ALL&deviceType=ALL&storageModel=ALL&showRamDisks=false&memoryChannels=0&country=&scoreType=overallScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock= Comments about Ultra series: Latency is killing it. More than 80ns and even 90 is killing game performance The IHS heat dissipation and new Rl-ILM retention mechanism is tons better than before (and no bending, finally) Still tons of BUGS with bios. Too early to launch Too few in stock. What the...? What to expect? Intel promised to improve performance. Sadly, latency can only be improved so much. This is a design flaw. E-cores maybe with more priority? The P cores are not that good anymore. With a Ultra 7 I also had, testing the E cores only and OCing them was really easy. Up to 5ghz. New gen in 2025 but... other socket? Not bueno, Intel. Whole platform change for 1 gen. If 300 Series are launched, they have to reduce latency. If Latency and Core layout is fixed, IPC is there to compete against X3D. 1
chiliwili69 Posted November 13, 2024 Author Posted November 13, 2024 7 hours ago, DBCOOPER011 said: Frames: 10960 - Time: 60000ms - Avg: 182.667 - Min: 159 - Max: 231 2024-11-12 23:31:59 - Il-2 Frames: 10723 - Time: 60000ms - Avg: 178.717 - Min: 157 - Max: 221 Thanks for this test. Wow, +22fps by just changing the chip. Was all RAM/Mobo the same?
=SFG=BoostedStig Posted November 13, 2024 Posted November 13, 2024 11 hours ago, chiliwili69 said: Many thanks for this test. It gives your +16fps with respect your previous Ryzen9 9850X3D at 5.7GHz. Not bad at all. And perhaps you also save some watts in your electricity bill. Yeah maybe! Although the 4090 does the bulk of the electricity wasting around here haha
=SFG=BoostedStig Posted November 13, 2024 Posted November 13, 2024 18 hours ago, =SFG=BoostedStig said: Got a 9800X3D on release. So far I've only had a chance to test the CPU benchmark Motherboard: Asus ROG Strix X670E-F CPU: AMD R7 9800X3D CPU Freq: 5.4 Ghz (PBO enabled +200mhz offset) Cores: 8 Threads: 16 RAM size: 64GB (2x32GB) RAM Freq: 6200 MHz (DDR5) FCLK: 2066 MHz RAM timings: 30-36-35-40 GPU: Gigabyte Gaming OC 4090 Windows 10 - 23H2 - 22631.4391 - Most recent version that isn't 24H2 or any early release/access build I'm pretty sure. 1080P CPU Test: 2024-11-12 19:52:14 - Il-2 Frames: 10118 - Time: 60000ms - Avg: 168.633 - Min: 147 - Max: 215 Making a small update to my findings here. I realized that I had run that test with some other things in the background (Discord, a Chrome window) so I re-ran them today. Didn't change the avg significantly. 2024-11-13 14:06:59 - Il-2 Frames: 10165 - Time: 60000ms - Avg: 169.417 - Min: 147 - Max: 229
=SFG=BoostedStig Posted November 13, 2024 Posted November 13, 2024 12 hours ago, DBCOOPER011 said: Just got my 9800X3D in today and did a couple runs with PBO +200 and -30 CO at XMP settings and got the below. Still need to tweak my ram settings since it didnt take with my 7800X3D setup. Seems like there is a lot of potential with this chip. My last 7800X3D at -30 CO below also.. 9800X3D: 2024-11-12 23:29:55 - Il-2 Frames: 10960 - Time: 60000ms - Avg: 182.667 - Min: 159 - Max: 231 2024-11-12 23:31:59 - Il-2 Frames: 10723 - Time: 60000ms - Avg: 178.717 - Min: 157 - Max: 221 7800X3D: 2024-11-08 05:41:58 - Il-2 Frames: 9490 - Time: 60000ms - Avg: 158.167 - Min: 135 - Max: 203 -30 Is this Windows 24H2 or 23H2? I've seen some testing in racing sims report 24H2 is slightly better and I'm wondering if that's the difference in our minimums.
=SFG=BoostedStig Posted November 14, 2024 Posted November 14, 2024 20 hours ago, =SFG=BoostedStig said: Making a small update to my findings here. I realized that I had run that test with some other things in the background (Discord, a Chrome window) so I re-ran them today. Didn't change the avg significantly. 2024-11-13 14:06:59 - Il-2 Frames: 10165 - Time: 60000ms - Avg: 169.417 - Min: 147 - Max: 229 Okay yet ANOTHER small update. After seeing @DBCOOPER011's runs vs my own, I figured something had to be different. I had heard and seen evidence that W11 24H2 brings performance updates, but I was not yet on it. So I went and updated to W11 24H2, and am now on what I believe is the most current build 26100.2314 This new result brings me and DB way more in line with each other now. Silly how an OS build can make such a large difference, but it's repeatable, and the results track with what I've seen other trusted sources post for games such as iRacing. 2024-11-13 20:26:25 - Il-2 Frames: 10500 - Time: 60000ms - Avg: 175.000 - Min: 153 - Max: 226
chiliwili69 Posted November 14, 2024 Author Posted November 14, 2024 5 hours ago, =SFG=BoostedStig said: Silly how an OS build can make such a large difference, but it's repeatable, and the results track with what I've seen other trusted sources post for games such as iRacing. Nice finding! I didn´t know that the OS played such a role, but good to know. I put it in the Notes of the spreadsheet.
DBCOOPER011 Posted November 14, 2024 Posted November 14, 2024 23 hours ago, chiliwili69 said: Thanks for this test. Wow, +22fps by just changing the chip. Was all RAM/Mobo the same? Its on an X870 Tomahawk that I was running my 7800X3D with and 6400mhz ram. Was running it with a -30 curve optimizer but dropped it down to -20 for all core stability.. 11 hours ago, =SFG=BoostedStig said: Okay yet ANOTHER small update. After seeing @DBCOOPER011's runs vs my own, I figured something had to be different. I had heard and seen evidence that W11 24H2 brings performance updates, but I was not yet on it. So I went and updated to W11 24H2, and am now on what I believe is the most current build 26100.2314 This new result brings me and DB way more in line with each other now. Silly how an OS build can make such a large difference, but it's repeatable, and the results track with what I've seen other trusted sources post for games such as iRacing. 2024-11-13 20:26:25 - Il-2 Frames: 10500 - Time: 60000ms - Avg: 175.000 - Min: 153 - Max: 226 Yea, Thats pretty much what I'm getting as well. Between 175-180..
=SFG=BoostedStig Posted November 14, 2024 Posted November 14, 2024 6 hours ago, chiliwili69 said: Nice finding! I didn´t know that the OS played such a role, but good to know. I put it in the Notes of the spreadsheet. Yeah, This is a little off topic, but in trying to keep it short: The whole debacle with Ryzen 9000 launch performance not being great, Youtube reviewers getting worse results and then AMD saying "oh we tested in an Admin environment" lead the broader gaming community to figure out that there was a process and core scheduling patch in the newest version of windows (24h2) that at that time had yet to be released. The patch affected AMD performance positively across the board, but the Intel stuff remained largely the same. Gaming YT covered it, but deep in the sim-racing content space, they had discovered that this made a non-trivial difference. At the time, I would have had to be on the insider program to get onto 24H2 and I didn't want to dip my toes into that realm. Now 24H2 is public release and they have been slowly rolling it out to windows users. But you can force install the final release version via the ISO from Microsoft. That's what I decided to do last night.
DBCOOPER011 Posted November 15, 2024 Posted November 15, 2024 (edited) Here are my 9800X3D findings after messing with it some today. I'm using curve shaper now and not curve optimizer. I was able to get higher CBP test scores, but decided not in order to stay well under the 1.4v vcore limit. I'm averaging around 1.3v in single core. I ran the VR3 test a few times since it didnt match up with my 7800X3D score, but the settings are accurate. Motherboard: MSI X870 Tomahawk CPU: Ryzen 9800X3D CPU Freq: 4.7 Ghz Cores: 8 Threads: 16 RAM size: 32Gb (2x16GB) RAM Freq: 6200 MHz NB Freq: 3100 MHz RAM timings: 30-36-36-48 GPU: 4090 CPU Test: 2024-11-14 21:35:55 - Il-2 Frames: 10805 - Time: 60000ms - Avg: 180.083 - Min: 155 - Max: 229 2024-11-14 21:38:05 - Il-2 Frames: 10777 - Time: 60000ms - Avg: 179.617 - Min: 156 - Max: 227 2024-11-14 21:40:31 - Il-2 Frames: 10737 - Time: 60000ms - Avg: 178.950 - Min: 157 - Max: 228 VR2 Test: 2024-11-14 21:58:58 - Il-2 Frames: 5373 - Time: 60000ms - Avg: 89.550 - Min: 85 - Max: 91 VR3 Test: 2024-11-14 22:03:53 - Il-2 Frames: 5395 - Time: 60000ms - Avg: 89.917 - Min: 87 - Max: 91 2024-11-14 22:20:42 - Il-2 Frames: 5382 - Time: 60000ms - Avg: 89.700 - Min: 83 - Max: 91 Edited November 15, 2024 by DBCOOPER011
chiliwili69 Posted November 15, 2024 Author Posted November 15, 2024 1 hour ago, DBCOOPER011 said: VR2 Test: 2024-11-14 21:58:58 - Il-2 Frames: 5373 - Time: 60000ms - Avg: 89.550 - Min: 85 - Max: 91 VR3 Test: 2024-11-14 22:03:53 - Il-2 Frames: 5395 - Time: 60000ms - Avg: 89.917 - Min: 87 - Max: 91 Thank you for running again the VR tests. Performance is top. I assume you used the G2? (or Crystal?) Just two comments: - It looks like current top CPUs are touching the ceiling of the 90fps for VR1 and VR2 tests. It means that we don´t measure really anything if from now all top CPU delivers 90fps in the VR tests 1 and 2. Perhaps we could raise the bar and do a more demanding set of tests. This is something to think. - It really surprised me your VR3 test. As you know the simple difference between VR2 and VR3 is just activate the MSAAx8 and Clouds to Extreme. And the MSAAx8 is really demanding a lot of GPU load, that´s why most of the previous tests were around 50´s.
FTC_ChilliBalls Posted November 17, 2024 Posted November 17, 2024 (edited) I hope you guys don‘t mind me asking here, but does the 5800X3D benefit from faster Ram speeds in this game? I know that in most games the performance of X3D chips is relatively hardware agnostic, but some, like Star Citizen, benefit noticeably from faster RAM and tighter CL timings Edited November 17, 2024 by FTC_ChilliBalls
DBCOOPER011 Posted November 17, 2024 Posted November 17, 2024 (edited) On 11/15/2024 at 12:10 AM, chiliwili69 said: Thank you for running again the VR tests. Performance is top. I assume you used the G2? (or Crystal?) Just two comments: - It really surprised me your VR3 test. As you know the simple difference between VR2 and VR3 is just activate the MSAAx8 and Clouds to Extreme. And the MSAAx8 is really demanding a lot of GPU load, that´s why most of the previous tests were around 50´s. No problem. I used the crystal and lowered the resolution in steamvr to 2860x3384 (44%) per the instructions. I just ran the test again at 90hz, verifying 8X MSAA/clouds extreme with the other CPU bench settings and got the result below. Maybe this isnt taxing the crystal enough since it runs native 1.0 at 4312x5100? 2024-11-16 19:15:14 - Il-2 Frames: 5370 - Time: 60000ms - Avg: 89.500 - Min: 84 - Max: 91 3 hours ago, FTC_ChilliBalls said: I hope you guys don‘t mind me asking here, but does the 5800X3D benefit from faster Ram speeds in this game? I know that in most games the performance of X3D chips is relatively hardware agnostic, but some, like Star Citizen, benefit noticeably from faster RAM and tighter CL timings I have not had the 5800X3D, but with this 9800X3D and my 7800X3d, it seems the lowest latency possible benefit the most in IL2 VR for me. I use to run 6400Mhz DDR5, but now running 6000Mhz with the lowest timing I can get with m-die DDR5 sticks. I'm getting low 180 FPS scores in the CPU test with this and using curve shaper in the bios on the 9800X3D.. Edited November 17, 2024 by DBCOOPER011 1 1
RAAF492SQNOz_Steve Posted November 17, 2024 Posted November 17, 2024 (edited) 7 hours ago, FTC_ChilliBalls said: I hope you guys don‘t mind me asking here, but does the 5800X3D benefit from faster Ram speeds in this game? I know that in most games the performance of X3D chips is relatively hardware agnostic, but some, like Star Citizen, benefit noticeably from faster RAM and tighter CL timings My personal experience is no, do not waste your money doing a RAM upgrade when running the 5800X3D unless you have a really slow set of RAM modules installed. I did an upgrade several years ago from a 3600 MHz 16 Gb (2x8 Gb) memory kit to a 4000 MHz 32 Gb (2 x 16Gb) with some really tight timings and it made zero difference to IL2 performance. Did not see any difference in game performance for IL2. My comments are specifically related to IL2 and flying in VR with a Varjo Aero at 39ppd (highest setting). From very distant recollections you could get some improved game performance with faster memory with the vanilla version AMD 5800X CPU (that I initially had installed in my PC) but the 5800X3D blew the 5800X out of the water for the games that I played at the time. As an aside, after doing lots of memory tuning during winter and achieving the, at the time, magical 4K Mhz with 2K infinity fabric I discovered that during summer that my PC became unstable so had to back things off a bit. These are now my stable all year round DDR4 settings. Edited November 17, 2024 by RAAF492SQNOz_Steve Provided current DDR4 memory timings. VR comment. 2
FTC_ChilliBalls Posted November 17, 2024 Posted November 17, 2024 4 hours ago, RAAF492SQNOz_Steve said: My personal experience is no, do not waste your money doing a RAM upgrade when running the 5800X3D unless you have a really slow set of RAM modules installed. I did an upgrade several years ago from a 3600 MHz 16 Gb (2x8 Gb) memory kit to a 4000 MHz 32 Gb (2 x 16Gb) with some really tight timings and it made zero difference to IL2 performance. Did not see any difference in game performance for IL2. My comments are specifically related to IL2 and flying in VR with a Varjo Aero at 39ppd (highest setting). From very distant recollections you could get some improved game performance with faster memory with the vanilla version AMD 5800X CPU (that I initially had installed in my PC) but the 5800X3D blew the 5800X out of the water for the games that I played at the time. As an aside, after doing lots of memory tuning during winter and achieving the, at the time, magical 4K Mhz with 2K infinity fabric I discovered that during summer that my PC became unstable so had to back things off a bit. These are now my stable all year round DDR4 settings. The question was rather about whether I should bother to do a manual RAM OC to see how far I could push it. From what I‘ve read in your and @DBCOOPER011's post is that apparently the game doesn‘t gain performance from higher clock speeds but perhaps from tighter timings. We‘ll see^^
=SFG=BoostedStig Posted November 17, 2024 Posted November 17, 2024 (edited) @FTC_ChilliBalls I agree with @DBCOOPER011 that games tend to care most about latency. To some degree, speed and latency are connected to each other because the timings are measured in clock cycles. So the faster the memory is, the shorter amount of actual time each clock cycle lasts. If you take the same memory timings and only bump the speed up 200MT/s or whatever, you inherently get a small bonus because you've shortened the length of each cycle. I've attached some photos here of some testing I've done. The photos are in order (I think) like this. I've highlighted the speed and latency measured in Aida64 in each one, and then for the ones I manually touched I've highlighted what I changed in relation to the photo before it. I wouldn't pay too too much attention to the raw latency number cause it can be affected by background tasks running, your OS in general. The super hardcore run a fully stripped OS to do these tests because it will report lower numbers. This is just my everyday computer I don't care THAT much lol *The sticks are 16gb Hynix A-die dual rank. 2x32GB configuration* Stock Full Auto motherboard settings 4800CL40 - 88.9ns latency EXPO turned on - 6000CL30 - 72ns latency EXPO with manual timings - 64ns 6200cl30 - 63.4ns 6200cl28 - 60.8ns What we notice here is the difference between board auto and EXPO is massive, 16.9ns, nearly 20% reduction in latency. But notably we also get a large jump between EXPO profile and manually tuning the timings. Same clock speed, 8ns latency improvement, another 11%. It quickly gets nonsensical and the returns diminish fast. The difference between 6000 with manual timings and 6200 tuned up is a marginal ~3ns at best and it takes a lot of effort and time, and also yields functionally zero performance improvement in games. TLDR; Look at the benchmark scores between myself and DB. We have largely the same computer hardware, but for some reason his score is higher than mine. My guess is his CPU is slightly better and maybe it boosts to an extra 50mhz or something, or his curve shaper settings are giving him more aggressive clock speed ramping? Who knows. It's close enough know where I think it's all margin of error. Edited November 17, 2024 by =SFG=BoostedStig 1
DBCOOPER011 Posted November 18, 2024 Posted November 18, 2024 Did 3 IL2 CPU benchmarks with 3 different DDR5 speeds in CapFrameX. All are stable in testmem5 and it appears to be a wash regarding any significant FPS gain in IL2. Bios settings below: PBO-Advanced PBO limits- Motherboard PBO Scaler- Manual/10X CPU Boost Override- 200Mhz Positive Curve Shaper: Min/Low- Negative 28 Medium- Negative 26 High- Negative 22 Max- Negative 20 2
LuftManu Posted December 1, 2024 Posted December 1, 2024 Dear friends, first Core ultra here: 285K with an Z890 Taichi OCF and 8600MTs memory. New 113 microcode, new PPM from Intel. Everything stock sans: - Ecores from 4,6 to 4,8 - Undervolt applied from -85Mv to -55Mv on worse cores - D2D and NGU to 3200Mhz for RAM and Latency. 2024-12-01 20:49:38 - Il-2 Frames: 10907 - Time: 60000ms - Avg: 181.783 - Min: 4 - Max: 242 There is a hiccup when FRAPs start, but It resumes back again. Early times for this Intel release, many strange things. Can't beat my 14900KS OCed yet... I'm trying to reach 9600 Mhz and so far I was able to get it stable with huge-active cooling and 1.6 but throwing a lot of errors. Happy running it here with the Undervolt, fast ram and max of 70ºC all core So "daily" and realistic config for the benchmark better than suicide runs. I was surprised, I thought it was going to run worse on Il-2. I am also getting really good results on Cyberpunk Benchmark. So far I was really dissapointed with this launch (and I also have a 9800X3D) but it's improving. If we have healthy competition, it benefits the users Kind regards,
chiliwili69 Posted December 2, 2024 Author Posted December 2, 2024 12 hours ago, LuftManu said: 285K with an Z890 Taichi OCF and 8600MTs memory. Many thanks for running the bench with this new CPU. I have seen to max p-cores frequency is 5.4 Ghz, but that CPU can reach 5.7 with out-of-the-box Turbo mode. Did you try to run the Passmark single-threaded test? https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+Ultra+9+285K&id=6296
LuftManu Posted December 2, 2024 Posted December 2, 2024 1 hour ago, chiliwili69 said: Many thanks for running the bench with this new CPU. I have seen to max p-cores frequency is 5.4 Ghz, but that CPU can reach 5.7 with out-of-the-box Turbo mode. Did you try to run the Passmark single-threaded test? https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+Ultra+9+285K&id=6296 HI! Yes, it was 5.7 in two cores in the test, as it was stock. The picture was after a full-core load only to check voltages and temps Also I was using W11 I will retry too using OC and tuning more the memory 2 ns less. These CPUs are really really fun to tweak. It takes me back to the times of 9900K. Sadly, there was no generational performance uplift, but it's miles ahead in stability and temps from 13-14th gen. And finally, you can overclock and tweak things! With the 13-14 gen you are almost limited to undervolt and control the temps. Kind regards,
von_Tom Posted December 21, 2024 Posted December 21, 2024 On 12/2/2024 at 9:42 AM, LuftManu said: HI! You speak in tongues and I have no idea what all of it meant! In very simple terms - from youyr experiences would you go Ultra 285k or 9800X3D for VR? Or see what the 9950X3D brings? von Tom
LuftManu Posted December 21, 2024 Posted December 21, 2024 6 hours ago, von_Tom said: You speak in tongues and I have no idea what all of it meant! In very simple terms - from youyr experiences would you go Ultra 285k or 9800X3D for VR? Or see what the 9950X3D brings? von Tom Hello! I would just go for the 9800X3D. I have this one as my main Sim RIG. The new Korea engine will use up to 5 cores, so we are served for the future. Same with other sims. The AMD performs much better out of the box and probably outperform even a fine-tuned 285K. Don't bother with the Intel Ultra family. For getting the same performance as the 9800X3D (Not even PBO) with the 285K, you need to buy expensive memory, pray for the silicon gods to have a good sample and expend hours tweaking and overclocking. (The 9950X3D does not bring dual cache and we are yet to know how it performs with the single CCD 3D cache. It might perform a little less than the 9800X3D in games, but it will have more cores for a more "multipurpouse" computer build). 1
LuftManu Posted December 23, 2024 Posted December 23, 2024 Benchmarks with 5080 should be up by 17th January on the table. New AMD gen won't bring anything better than 7900 XT or 7900XTX. (9000 series) So those two will remain leaders in AMD cards.
LuftManu Posted February 2 Posted February 2 Hi! First test with 5080 on Stock, OC and also Smooth motion (Frame generation) 2025-02-02 17:55:17 - Il-2 Frames: 9410 - Time: 60000ms - Avg: 156.833 - Min: 131 - Max: 193 - Stock 2025-02-02 17:59:05 - Il-2 Frames: 10719 - Time: 60000ms - Avg: 178.650 - Min: 149 - Max: 215 - OC. (+350 Core) 3150 mhz 15375 memory (+375 memory) 2025-02-02 18:02:15 - Il-2 Frames: 31623 - Time: 60000ms - Avg: 527.050 - Min: 449 - Max: 609 - FrameGeneration (Smooth) Stock Yes, this is real. It works on flat screen. No artffacts. Had one crash tho. But damm! Works great so far. Only works on 5000 Series so far. If this worked on VR... If there is need for proof I can record via mobile or something. (Smooth motion doesn't like overlays but Fraps manages to catch the fps) Here is a third party review: Rig: Motherboard: MSI X870E Carbon Wifi CPU: AMD R7 9800X3D CPU Freq: 5.25 Ghz PBO enabled Cores: 8 Threads: 16 RAM size: 48GB (2x48GB) RAM Freq: 6200 MHz (DDR5) FCLK: 2177 MHz RAM timings: 30-36-36-28-68 GPU: MSI 5080 Vanguard Special Launch edition Latest windows and release drivers. https://www.3dmark.com/3dm/125131692 A 3DMark test I did with OC (Matches the 4090) PS: DCS with the Frame generation on too. From 70 to 230-250 fps Spoiler 1
chiliwili69 Posted February 7 Author Posted February 7 On 2/2/2025 at 6:25 PM, LuftManu said: First test with 5080 on Stock, OC and also Smooth motion (Frame generation) Many thanks for this, but what benchmark test did you run? was it the CPU Test? or the GPU Test? Did you run the same tests with your previous card (4090?) in that PC?
LuftManu Posted February 16 Posted February 16 On 2/7/2025 at 9:04 PM, chiliwili69 said: Many thanks for this, but what benchmark test did you run? was it the CPU Test? or the GPU Test? Did you run the same tests with your previous card (4090?) in that PC? HI! It's another 9800X3D rig. It was CPU test Wanted to try the "unloading" frames from the CPU to GPU with the frame generator. Still, the 4090 is better.
c19580 Posted March 12 Posted March 12 I’d like to get some advice on whether upgrading my CPU would enable me to get a step upgrade in VR performance. My current system is a 4090/13900k/DDR5 6000 and run a Pimax Crystal Light at max resolution and most IL2 settings at max (except clouds/shadows) and get a consistent 72 fps in multiplayer. What Ryzen CPU, if any, would allow me to get up to a consistent 90 fps with the same settings? Trying to figure out if the expense would be worth it since I’d also need to change the mobo and ram.
354thFG_Drewm3i-VR Posted March 13 Posted March 13 11 hours ago, c19580 said: I’d like to get some advice on whether upgrading my CPU would enable me to get a step upgrade in VR performance. My current system is a 4090/13900k/DDR5 6000 and run a Pimax Crystal Light at max resolution and most IL2 settings at max (except clouds/shadows) and get a consistent 72 fps in multiplayer. What Ryzen CPU, if any, would allow me to get up to a consistent 90 fps with the same settings? Trying to figure out if the expense would be worth it since I’d also need to change the mobo and ram. In multiplayer? Nah. What the cpu really helps for is running the AI overhead. You would see better performance, but not worth the money IMO. 1
354thFG_Drewm3i-VR Posted March 13 Posted March 13 I would say turn down/off some settings @c19580 like horizon drawn distance, ssao, msaa, and maybe even the preset to high and go from there. You may get 90 fps with little loss in quality. Have to experiment. You can also get more out of your pc with something like razer cortex to shut off background processes and process lasso to optimize IL-2 priority and core affinity settings (turn off SMT for the game at least, and make sure it runs only on P-cores).
chiliwili69 Posted March 14 Author Posted March 14 I would agree with the advice, your CPU/GPU combo is quite at the top (I suposse the CPU is properly cooled), I don´t think an AMD CPU would give you more fps. The best way to know exactly what is your limiting factor in every scene is you use fpsVR while you play. I tend to think that giving the high number of pixels required to be rendered by the Crystal Light your 4090 could be the limiting factor. 1
Aapje Posted March 14 Posted March 14 The 9800X3D might provide better stability in frame rates, but it's going to be an expensive change for relatively little gain.
=SFG=BoostedStig Posted March 15 Posted March 15 RTX 5090 checking in for the 4K test Largely the same system as before, just with a 5090 now. Of note, I ran it a couple times and quickly realized that usage of the GPU never got above 90%. Power draw was all over the place. I'm fairly certain this is a CPU limit now. Motherboard: Asus ROG Strix X670E-F CPU: AMD R7 9800X3D CPU Freq: 5.4 Ghz (PBO enabled +200mhz offset) Cores: 8 Threads: 16 RAM size: 64GB (2x32GB) RAM Freq: 6000 MHz (DDR5) FCLK: 2000 MHz RAM timings: 30-36-32-30 GPU: MSI RTX 5090 Vanguard SOC Launch Edition OS: Windows 11 - 24H2 - Build 26100.3476 4K Test: 2025-03-14 21:45:35 - Il-2 Frames: 14439 - Time: 60000ms - Avg: 240.650 - Min: 211 - Max: 297
c19580 Posted March 16 Posted March 16 On 3/13/2025 at 2:45 PM, 356thFS_Drewm3i-VR said: I would say turn down/off some settings @c19580 like horizon drawn distance, ssao, msaa, and maybe even the preset to high and go from there. You may get 90 fps with little loss in quality. Have to experiment. You can also get more out of your pc with something like razer cortex to shut off background processes and process lasso to optimize IL-2 priority and core affinity settings (turn off SMT for the game at least, and make sure it runs only on P-cores). thanks i’ll have to look into razer cortex and process lasso. Not familiar with them!
chiliwili69 Posted March 17 Author Posted March 17 On 3/15/2025 at 3:58 AM, =SFG=BoostedStig said: 4K Test: 2025-03-14 21:45:35 - Il-2 Frames: 14439 - Time: 60000ms - Avg: 240.650 - Min: 211 - Max: 297 Many thanks for running the first GPU 4K test with the 5090, it is a good improvement over the 4090 (190-195 fps ). And that, without reaching perhaps the true limit since GPU load was not at 100%. The passmark GPU benchmarks seems to don´t fully reflect this improvement since the 4090 has around 38000 and the 5090 around 39000. Strange. In any case, the important thing is how it performs in IL-2. Rendering 4K at 240 fps is 3840x2160=8.3 Mpixels per 240, that is a load of 2Gpixels/second. And assuming the 90% factor (means 10% more)it means that a 5090 can process a load of rendering 2.2 Gpixels/second. This is only a little bit below the raw GPU performance required by the Meganex8K at default resolution at 72Hz. We still will need advanced techniques (FFR, DFR with eyetracking, upscaling, etc) or reduce default resolution to run the latest 4K VR devices. 1
=SFG=BoostedStig Posted March 28 Posted March 28 On 3/17/2025 at 3:16 AM, chiliwili69 said: Many thanks for running the first GPU 4K test with the 5090, it is a good improvement over the 4090 (190-195 fps ). And that, without reaching perhaps the true limit since GPU load was not at 100%. The passmark GPU benchmarks seems to don´t fully reflect this improvement since the 4090 has around 38000 and the 5090 around 39000. Strange. In any case, the important thing is how it performs in IL-2. Rendering 4K at 240 fps is 3840x2160=8.3 Mpixels per 240, that is a load of 2Gpixels/second. And assuming the 90% factor (means 10% more)it means that a 5090 can process a load of rendering 2.2 Gpixels/second. This is only a little bit below the raw GPU performance required by the Meganex8K at default resolution at 72Hz. We still will need advanced techniques (FFR, DFR with eyetracking, upscaling, etc) or reduce default resolution to run the latest 4K VR devices. There's something suspicious going on with these 5090s in VR. I haven't tested it in IL2 yet, but in iRacing, the gain is way more than I'm expecting. On paper, everything says this should only be maybe 25% better than a 4090, but I am seeing way better than 25% performance improvement in iRacing VR. On my 4090, I was running about 2880x3400 per eye at 90hz, but now I am running 3125x3700 per eye and am able to get 120hz out of it, in iRacing. There is some suspicion that the performance scales better than expected with high resolution VR because the memory bandwidth is nearly 70% higher with the new card. But none of us really know why it's running so much faster than it seems like it should be. I am going to try running the benchmark here shortly.
=SFG=BoostedStig Posted March 28 Posted March 28 54 minutes ago, =SFG=BoostedStig said: There's something suspicious going on with these 5090s in VR. I haven't tested it in IL2 yet, but in iRacing, the gain is way more than I'm expecting. On paper, everything says this should only be maybe 25% better than a 4090, but I am seeing way better than 25% performance improvement in iRacing VR. On my 4090, I was running about 2880x3400 per eye at 90hz, but now I am running 3125x3700 per eye and am able to get 120hz out of it, in iRacing. There is some suspicion that the performance scales better than expected with high resolution VR because the memory bandwidth is nearly 70% higher with the new card. But none of us really know why it's running so much faster than it seems like it should be. I am going to try running the benchmark here shortly. This will probably take more testing than I have time for this evening, and I am going on a trip tomorrow morning. But here's what I've got. Motherboard: Asus ROG Strix X670E-F CPU: AMD R7 9800X3D CPU Freq: 5.4 Ghz (PBO enabled +200mhz offset) Cores: 8 Threads: 16 RAM size: 64GB (2x32GB) RAM Freq: 6000 MHz (DDR5) FCLK: 2000 MHz RAM timings: 30-36-32-30 GPU: MSI RTX 5090 Vanguard SOC Launch Edition OS: Windows 11 - 24H2 - Build 26100.3476 HMD: Pimax Crystal set at 44% SS, as specified. VR Test 2: 2025-03-27 21:23:10 - Il-2 Frames: 5382 - Time: 60000ms - Avg: 89.700 - Min: 86 - Max: 91 The minimum frames here is either just a frame drop somewhere, or a CPU limit somewhere, because GPU usage throughout the benchmark really doesn't go above 50%. VR Test 3: 2025-03-27 20:55:05 - Il-2 Frames: 5317 - Time: 60000ms - Avg: 88.617 - Min: 78 - Max: 91 This one is a LOT heavier, as expected. GPU usage is high, but never really gets to 100%. I see a lot of high 80's in the early parts of the run and then a run up to the high 90's when the camera pans around the passing planes, which is where the brief dip to 78 is at.
chiliwili69 Posted March 29 Author Posted March 29 On 3/28/2025 at 2:31 AM, =SFG=BoostedStig said: but now I am running 3125x3700 per eye and am able to get 120hz out of it, in iRacing Hey, this is indeed much more than expected from the 5090. Running 3125x3700 is 23.1Mpixels at 120Hz, which is like rendering 30.8 Mpixels at 90Hz, which is 2.77 Gpixels/s Your previous IL-2 4K test you achieved 2.2 Gpixels/s rendered. Perhaps iRacing applies some short of FFR or the resolution is internally capped.
chiliwili69 Posted March 29 Author Posted March 29 (edited) On 3/28/2025 at 3:30 AM, =SFG=BoostedStig said: VR Test 2: 2025-03-27 21:23:10 - Il-2 Frames: 5382 - Time: 60000ms - Avg: 89.700 - Min: 86 - Max: 91 Thank you for these VR tests with your Crystal and the 5090. This SYN_VANDER benchmark is perhaps a bit outdated in the VR area since they were designed some years ago when the resolution of the VR devices were not as high as it is now and the GPUs were not as powerful as they are right now. So, perhaps it is the time to raise the bar of the VR tests with just higher resolutions (without using MSAAx4 at all). I could change the VR benchmark to whatever makes sense. They are designed to look for the best CPUs and GPUs to play IL-2 in VR, specially latest VR devices. Firstly, I think I would keep the same frequency of 90Hz since most the devices are able to run at that frequency. Secondly, I would reduce the VR tests to just only two tests, VR test 1 and VR test 2. It should be enough. I would propose to set the new VR Test 1 to the same resolution of the previous VR test 2, which is 19.5 Million pixels at 90Hz, (since this was the default resolution of the discontinued Reverb G2). This resolution is achievable by the 4090 cards as tested by @DBCOOPER011. For the new VR Test 2, I would propose a resolution high enough not achievable by a 5090, for example 35 Million pixels or whatever resolution makes a 5090 perform around 50-60 fps in VR. What you do think? Edited March 29 by chiliwili69
Aapje Posted May 15 Posted May 15 Please don't hijack threads that are on a completely different topic, but make a separate threads. @LukeFF can you please move the last comment to a separate thread?
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now