Thanks for this clip. My 9950x runs with R23 46900, after max pbo and very important: oc the dram. Sync-Mode with the sweet-Spot 6400Mhz, 30,38,38. curve-shaper all core 12 to 15. thats it
On the 9950x -35 would be no where near stable. On the lesser core cpus like the 9600x that runs no problem, but flat setting -35 on the 9950x lead to crashing.
@@JohnnyRage303 WTF? so maybe i have a golden? -35 curve: 16h on prime95 small FFT, Cinebench stable, OCCT large and small Extreme 2h, karhu for the memory for 9h, and AIDA64 stability system test 4h, no problem, no stuttering in games no audio cuts, nothing. My SP rating on asus says 118, so i dont know. I did test a CPU from a friend and it was -30 stable, didnt bother to test -35 but mmm i dont know man maybe its your chip only that cant do it? at least -30 should be doable.
@@tacoterito22 well maybe you got way more lucky then me, but to your original question of -35 vs curve shaper + optimizer: ultimately regardless of strategy your limited by that FMAX so if you +200 it would be 5.9ghz. You are just trying to reach a given voltage to frequency sweet spot where it is stable and you maintain the FMAX of 5.9ghz. Mine couldn't do it all core, but if you have all core -35 set I would be willing to bet you're never going to get any better than that. all these other tricks are to stabilize a CPU under different conditions to try to maintain that FMAX frequency.
coming from intel i gotta say i really don't enjoy the overclocking on these amd chips, i'm on 9950x rn too and it feels like the only way to make it clock close to 5,8 or higher in games is to direct die it. currently been using the heatkiller iv block with noctua nth2 paste on a setup with 4 radiators 17 phanteks t30 and 2 d5 pumps and it still doesn't clock higher than 5,6-5,7 in games on ccd0 i don't get how these chips run so hot while pulling below 150watts in games
There still isn't a really compelling reason to upgrade to Zen 5 for most people with a mid-high end previous generation CPU, the overall performance improvement isn't massive and upgrading is relatively costly going from DDR4 to DDR5, so it isn't too surprising they're struggling to sell them, maybe when the Nvidia 50 series comes out there will be more of a reason to upgrade.
Yeah, I have done a lot of video card testing and for most cases, any CPU in the last few gens are really hard to bottleneck. if you AM4 and have the 5700x3d available there's not really a reason to build new gen. If you have an older am4 with lower ends parts it might be time to full upgrade but the 7000 series is a 95% solution with 70% of the cost. I'm not surprised they haven't sold like crazy but all rumors indicate intel will be in the same spot. All that aside I have had fun learning to tune the new system. Thanks for watching / commenting.
@@beerussama2297 for extreme budgets am4 is still good but you're right people preach like the 5600 is all you need, when in reality the 7600 is on par with the 5800x3d and it only goes up from there on the AM5. If you are already on AM5 the upgrade to 9000 is hard. I think the biggest upgraders will be the 7600 to 9800x3d once people get the taste for speed.
Have you tried to get higher Uclk speeds? You are probably the only person I've seen do the right thing with the 8000mhz ram & higher divider. It was obvious that the 2000mhz Fclk infinity fabric would be a bottleneck for 8000mhz even with the divider/Sync purpose. Since bandwidth would override latency.
On my 9600x I have a 6000mhz kit. I oc'd it to 6400 cl 30, and ran uclk 3200. Anything beyond that I couldn't get 1:1. At one point I had it booting with 2400 fclk, and I was like zen 5 is gonna slap. But once I applied all my settings it was never able to boot at 2400 again. Both the 9600 and 9950x have no problem at 2200 for me. I'm 2 out of 2 for immediately setting fclk 2200, where zen 4 it seemed like one in a hundred would hit 2200.
Yes, because at 8000MT ram speed you have 4000mhz DDR5, the uclk can either be = memclk or 1/2 memclk (4000mhz in this case) Uclk will not run 1:1 at 4000mhz. the highest I have gotten is 3200mhz on ddr5 6400. so if you are forced memclk/2 you get 2000mhz. now with dual ccd cpus there are 2 seperate infinity fabric lanes one to each ccd. So my initial testing running Fclk faster improved my cinebench scores. All zen 4 and on CPUs have desync'd frequencies, where AM4 you needed to run 1:1:1 or you would take penalty latency. this is not true for zen 4-5. People generally just run 2:3 because this ratio was recommended initially by AMD and it works well 2000 fclk 3000 uclk/memclk 6000mt/s ram. its on my list of things to do benchmark 9950x 2000 flkck, 2200 fclk, and my ddr5 6000 kit to see what does the best.
@@kevinerbs2778 You can t raise ulck. If you run ram at 8000, it will go 4000 mlck and 2000 ulck because it is ulclk=mlck/2. If you want higher ulck, you have to run 1:1 mod so uclk=mlck, in this case for 6000 ram you ll have 3000 ulck and 3000 mlck with 2000 flck default. For 6400 ram 1:1 mode, 3200 and 3200 and so on. Rare chips can do 6600 1:1.
@@Eusebiugh This doesn't apply there a newer divider on AM5 which were not on AM4. He answered my question already, he is running Fclk above Uclk so your whole comment is wrong.
My stock performance on my 9950x is 44900 in R23 ? I keep seeing people tweaking their settings and getting 44000 to 45000 (at best) as their improvement. I've limited the temperature to 88 C as well because I was worried about it jumping up to 95 degrees immediately when running multicore benches. Shall I leave it as it is, or should I play with it and see if I can squeeze some more out of it? I'm running Asus Crosshair Hair x870e, Tiny Win11, 96GB Dominator Titanium 6400Mhz, 9950x with a Lian Li AIO water cooler.. 2 x MP700 Gen 5 SSD's in RAID-0 but I can only get a maximum read speed of 12GB/s and Write speed of 20GB/s and I don't know why. I've tried everything. This PC isn't for gaming so I don't even have a graphics card in it so I'm not worried about the GEN5 PCIe lanes being used up. I'm so confused because I thought my READ speed would might be maxing out a Gen 4 PCIe bus - but then the Write speed negates that theory completely.
That's a great score, I would be willing to to bet you are on the higher end of scores due to the lite version of windows installed. If you are happy with the results run it as is. stock is always stable. but you can always play around to see if you can get a little more out of it. at least enable PBO with a modest negative curve optimizer should be no problem. As far as the SSDs those are pretty fast transfer speeds. The reason you might not see the PCIE limit in a lot of cases is because they lanes are shared. You should just look up your motherboard specs, if the PCIE lanes are shared then they will be limited on bandwidth with your GPU. if they are not shared then you should have your full bandwidth. The Crosshair is a good board but I don't know its PCIE specs off the top of my head.
@@JohnnyRage303 Thank you for your reply, I really appreciate it. The debloated OS helped a bit. To get the best scores, it also helps going to task manager and closing down as many services as look unncessary as you can. You can gain 1000+ points from closing down useless services. I thought the same about the PCIe lanes. My mobo is supposed to have the NVME drives directly wired to the CPU in Gen5 mode. But I checked the settings and I also don't have a GPU as its a workstation, so I'm just using the iGFX. I tried everything including messing with lane bifurcation settings and focing Gen5 instead of auto, but nothing changed. I've seen these SSDs get 24GB/s read using Windows RAID. I think that settings up an array on the motherboard is just bad in general. Wish I could get some better general timings for my memory without that weird flickering. The 90ns latency is a bit weird too. Not great. Do you have any experience in locking the CPU at a fixed frequency? I heard it can make windows feel snappier as there is a little latency in ramping up the clock speed from idle.
@MOSMASTERING generally locking the cpu is a traditional overclock. You set 1 multiplier and a fixed voltage. I did this a ton back in the day on intel cpu's because you could pickup a real 20% performance increase. On the newer amd cpus you can still do it. Most people don't. The pros are better multithreaded performance, set speed. The cons are higher energy use, lower performance in lesser thread count tasks. Most people prefer PBO because it's relatively efficient at managing the cpu for good performance but there are always specific use cases where a locked freq could be a benefit.
@@JohnnyRage303 After all my tweaks and benching, I just want to leave it all alone and use it to run Cubase - that is the only application I'll be installing. So long as it runs that smoothly along with all its plugins and effects, I'll be happy. Real-time audio processing and when entering midi data using a keyboard, it all benefits from low latency. So that's my primary goal - lowest latency with best performance I can get. That's why I thik ram tweaking might help the most in that respect.
I have 9950x with 360mm aio and i accidentally overvolted to constant fixed 1.35 volts and underclocked to fixed 5.1ghz bcuz im just testing things learning on my own tuning. I benchmarked in cinebench 23 and hit 103c with average 96c. The max safe temp advertised for this cpu was 95c. Now im panicking over my cpu already having degradation by the first week using it. I undid everything and now my cpu doesnt go above 75c stress test with auto voltage and clockspeed on and pbo on and curve optimizer and shape manually tuned with 150mhz+. Is my cpu damaged and how can i tell if it is? Even after undoing my terrible tuning settings, the cpu is getting above 1.4 volts while idle automatically although very good temps now. Is this also bad for cpu lifespan? I want to prevent further degradation if so. Ive tried searching many forums but nobody has clear answers. Please help on what i should do thanks❤
I wouldn't worry about a one time temp spike as long as you didn't leave it there. The tjmax is designed to be ran at continously there's some cushion built in there. Usually most amd cpus benefit the most from a good pbo setup the all core overclocks generally only improve very particular all core workloads. I would just run it pbo, and I'd bet you never see a problem
@JohnnyRage303 yes thanks i figured out my problem with pbo, i had the overdrive scalar to 10x and my voltage was spiking high more frequently but turned it to 1x and then auto and now its all normal. Im never doing fixed clockspeed & voltage ever again lol that was a nightmare accidentally sending my cpu to the desert at noon😅
curve shaper isn't what I thought it would be, seems like a extention of curve optimizer, should be no reason they can't do this for zen 4 or even zen 3
@@kevinerbs2778 i don't think that would prevent them from doing so, unless the sensors they are using are wildly different. especially for zen 4, zen 3 maybe but i don't see why they couldn't do this for zen 4.
It's a nice touch because you get squeeze out that extra bit after finding your stable optimizer setting. I doubt they will offer it backwards to zen 4. As mentioned I think there's additional temp sensor info.
@@JohnnyRage303 if the extra sensor information is whats enabling the curve shaper feature that is interesting at best. but I understand now why curve shaper on the 9600X isn't worth it at all.
Thanks for this clip. My 9950x runs with R23 46900, after max pbo and very important: oc the dram. Sync-Mode with the sweet-Spot 6400Mhz, 30,38,38. curve-shaper all core 12 to 15. thats it
Great results, I run 6400 cl30 on my 9600x, the 9950x has 8000 right now. I haven't fully tuned it yet. its sitting at cl38 which is the XMP profile.
Great video!
Thanks!
We got JohnnyRage without a mustache before GTA6
The mustache was in your head, you saw what you wanted to see.
Great video mate, it would be nice to test, Curve optimizer -35 vs Curve optimizer -15 + Curve shaper!!
On the 9950x -35 would be no where near stable. On the lesser core cpus like the 9600x that runs no problem, but flat setting -35 on the 9950x lead to crashing.
@@JohnnyRage303 WTF? so maybe i have a golden? -35 curve: 16h on prime95 small FFT, Cinebench stable, OCCT large and small Extreme 2h, karhu for the memory for 9h, and AIDA64 stability system test 4h, no problem, no stuttering in games no audio cuts, nothing. My SP rating on asus says 118, so i dont know. I did test a CPU from a friend and it was -30 stable, didnt bother to test -35 but mmm i dont know man maybe its your chip only that cant do it? at least -30 should be doable.
@@tacoterito22 did you raise the fmax +200 as well?
@@JohnnyRage303Yes i did
@@tacoterito22 well maybe you got way more lucky then me, but to your original question of -35 vs curve shaper + optimizer: ultimately regardless of strategy your limited by that FMAX so if you +200 it would be 5.9ghz. You are just trying to reach a given voltage to frequency sweet spot where it is stable and you maintain the FMAX of 5.9ghz. Mine couldn't do it all core, but if you have all core -35 set I would be willing to bet you're never going to get any better than that. all these other tricks are to stabilize a CPU under different conditions to try to maintain that FMAX frequency.
coming from intel i gotta say i really don't enjoy the overclocking on these amd chips, i'm on 9950x rn too and it feels like the only way to make it clock close to 5,8 or higher in games is to direct die it. currently been using the heatkiller iv block with noctua nth2 paste on a setup with 4 radiators 17 phanteks t30 and 2 d5 pumps and it still doesn't clock higher than 5,6-5,7 in games on ccd0
i don't get how these chips run so hot while pulling below 150watts in games
There still isn't a really compelling reason to upgrade to Zen 5 for most people with a mid-high end previous generation CPU, the overall performance improvement isn't massive and upgrading is relatively costly going from DDR4 to DDR5, so it isn't too surprising they're struggling to sell them, maybe when the Nvidia 50 series comes out there will be more of a reason to upgrade.
Yeah, I have done a lot of video card testing and for most cases, any CPU in the last few gens are really hard to bottleneck. if you AM4 and have the 5700x3d available there's not really a reason to build new gen. If you have an older am4 with lower ends parts it might be time to full upgrade but the 7000 series is a 95% solution with 70% of the cost.
I'm not surprised they haven't sold like crazy but all rumors indicate intel will be in the same spot. All that aside I have had fun learning to tune the new system. Thanks for watching / commenting.
I kindly disagree. From 7000 to 9000 maybe. But 5000 definetly reaches it's limits with a 4090.
@@beerussama2297 I mean if you're needing a 4090 sure, you have the money anyway, it's more at the mid range it doesn't make much sense.
@@beerussama2297 for extreme budgets am4 is still good but you're right people preach like the 5600 is all you need, when in reality the 7600 is on par with the 5800x3d and it only goes up from there on the AM5. If you are already on AM5 the upgrade to 9000 is hard. I think the biggest upgraders will be the 7600 to 9800x3d once people get the taste for speed.
Have you tried to get higher Uclk speeds?
You are probably the only person I've seen do the right thing with the 8000mhz ram & higher divider. It was obvious that the 2000mhz Fclk infinity fabric would be a bottleneck for 8000mhz even with the divider/Sync purpose. Since bandwidth would override latency.
On my 9600x I have a 6000mhz kit. I oc'd it to 6400 cl 30, and ran uclk 3200. Anything beyond that I couldn't get 1:1. At one point I had it booting with 2400 fclk, and I was like zen 5 is gonna slap. But once I applied all my settings it was never able to boot at 2400 again. Both the 9600 and 9950x have no problem at 2200 for me. I'm 2 out of 2 for immediately setting fclk 2200, where zen 4 it seemed like one in a hundred would hit 2200.
@@JohnnyRage303 is your uclk only 2000mhz on the 9950x?
while the infinity fabric is 2200mhz?
Yes, because at 8000MT ram speed you have 4000mhz DDR5, the uclk can either be = memclk or 1/2 memclk (4000mhz in this case) Uclk will not run 1:1 at 4000mhz. the highest I have gotten is 3200mhz on ddr5 6400. so if you are forced memclk/2 you get 2000mhz. now with dual ccd cpus there are 2 seperate infinity fabric lanes one to each ccd. So my initial testing running Fclk faster improved my cinebench scores. All zen 4 and on CPUs have desync'd frequencies, where AM4 you needed to run 1:1:1 or you would take penalty latency. this is not true for zen 4-5. People generally just run 2:3 because this ratio was recommended initially by AMD and it works well 2000 fclk 3000 uclk/memclk 6000mt/s ram. its on my list of things to do benchmark 9950x 2000 flkck, 2200 fclk, and my ddr5 6000 kit to see what does the best.
@@kevinerbs2778 You can t raise ulck. If you run ram at 8000, it will go 4000 mlck and 2000 ulck because it is ulclk=mlck/2. If you want higher ulck, you have to run 1:1 mod so uclk=mlck, in this case for 6000 ram you ll have 3000 ulck and 3000 mlck with 2000 flck default. For 6400 ram 1:1 mode, 3200 and 3200 and so on. Rare chips can do 6600 1:1.
@@Eusebiugh This doesn't apply there a newer divider on AM5 which were not on AM4. He answered my question already, he is running Fclk above Uclk so your whole comment is wrong.
My stock performance on my 9950x is 44900 in R23 ?
I keep seeing people tweaking their settings and getting 44000 to 45000 (at best) as their improvement.
I've limited the temperature to 88 C as well because I was worried about it jumping up to 95 degrees immediately when running multicore benches.
Shall I leave it as it is, or should I play with it and see if I can squeeze some more out of it?
I'm running Asus Crosshair Hair x870e, Tiny Win11, 96GB Dominator Titanium 6400Mhz, 9950x with a Lian Li AIO water cooler..
2 x MP700 Gen 5 SSD's in RAID-0 but I can only get a maximum read speed of 12GB/s and Write speed of 20GB/s and I don't know why. I've tried everything. This PC isn't for gaming so I don't even have a graphics card in it so I'm not worried about the GEN5 PCIe lanes being used up. I'm so confused because I thought my READ speed would might be maxing out a Gen 4 PCIe bus - but then the Write speed negates that theory completely.
Normely the drive c:\ has bad results in speed tests. Don‘t worry about that.
That's a great score, I would be willing to to bet you are on the higher end of scores due to the lite version of windows installed. If you are happy with the results run it as is. stock is always stable. but you can always play around to see if you can get a little more out of it. at least enable PBO with a modest negative curve optimizer should be no problem. As far as the SSDs those are pretty fast transfer speeds. The reason you might not see the PCIE limit in a lot of cases is because they lanes are shared. You should just look up your motherboard specs, if the PCIE lanes are shared then they will be limited on bandwidth with your GPU. if they are not shared then you should have your full bandwidth. The Crosshair is a good board but I don't know its PCIE specs off the top of my head.
@@JohnnyRage303 Thank you for your reply, I really appreciate it. The debloated OS helped a bit. To get the best scores, it also helps going to task manager and closing down as many services as look unncessary as you can. You can gain 1000+ points from closing down useless services.
I thought the same about the PCIe lanes. My mobo is supposed to have the NVME drives directly wired to the CPU in Gen5 mode. But I checked the settings and I also don't have a GPU as its a workstation, so I'm just using the iGFX.
I tried everything including messing with lane bifurcation settings and focing Gen5 instead of auto, but nothing changed.
I've seen these SSDs get 24GB/s read using Windows RAID. I think that settings up an array on the motherboard is just bad in general.
Wish I could get some better general timings for my memory without that weird flickering. The 90ns latency is a bit weird too. Not great.
Do you have any experience in locking the CPU at a fixed frequency? I heard it can make windows feel snappier as there is a little latency in ramping up the clock speed from idle.
@MOSMASTERING generally locking the cpu is a traditional overclock. You set 1 multiplier and a fixed voltage. I did this a ton back in the day on intel cpu's because you could pickup a real 20% performance increase. On the newer amd cpus you can still do it. Most people don't. The pros are better multithreaded performance, set speed. The cons are higher energy use, lower performance in lesser thread count tasks. Most people prefer PBO because it's relatively efficient at managing the cpu for good performance but there are always specific use cases where a locked freq could be a benefit.
@@JohnnyRage303 After all my tweaks and benching, I just want to leave it all alone and use it to run Cubase - that is the only application I'll be installing. So long as it runs that smoothly along with all its plugins and effects, I'll be happy. Real-time audio processing and when entering midi data using a keyboard, it all benefits from low latency. So that's my primary goal - lowest latency with best performance I can get. That's why I thik ram tweaking might help the most in that respect.
I have 9950x with 360mm aio and i accidentally overvolted to constant fixed 1.35 volts and underclocked to fixed 5.1ghz bcuz im just testing things learning on my own tuning. I benchmarked in cinebench 23 and hit 103c with average 96c. The max safe temp advertised for this cpu was 95c. Now im panicking over my cpu already having degradation by the first week using it. I undid everything and now my cpu doesnt go above 75c stress test with auto voltage and clockspeed on and pbo on and curve optimizer and shape manually tuned with 150mhz+. Is my cpu damaged and how can i tell if it is? Even after undoing my terrible tuning settings, the cpu is getting above 1.4 volts while idle automatically although very good temps now. Is this also bad for cpu lifespan? I want to prevent further degradation if so. Ive tried searching many forums but nobody has clear answers. Please help on what i should do thanks❤
I wouldn't worry about a one time temp spike as long as you didn't leave it there. The tjmax is designed to be ran at continously there's some cushion built in there. Usually most amd cpus benefit the most from a good pbo setup the all core overclocks generally only improve very particular all core workloads. I would just run it pbo, and I'd bet you never see a problem
@JohnnyRage303 yes thanks i figured out my problem with pbo, i had the overdrive scalar to 10x and my voltage was spiking high more frequently but turned it to 1x and then auto and now its all normal. Im never doing fixed clockspeed & voltage ever again lol that was a nightmare accidentally sending my cpu to the desert at noon😅
curve shaper isn't what I thought it would be, seems like a extention of curve optimizer, should be no reason they can't do this for zen 4 or even zen 3
They can't because they calibrated it for the lower temperature sensors of Zen 5.
@@kevinerbs2778 i don't think that would prevent them from doing so, unless the sensors they are using are wildly different. especially for zen 4, zen 3 maybe but i don't see why they couldn't do this for zen 4.
It's a nice touch because you get squeeze out that extra bit after finding your stable optimizer setting. I doubt they will offer it backwards to zen 4. As mentioned I think there's additional temp sensor info.
@@JohnnyRage303 if the extra sensor information is whats enabling the curve shaper feature that is interesting at best. but I understand now why curve shaper on the 9600X isn't worth it at all.