• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

i7-9800x overclock good enough?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

jposch

New Member
Joined
Oct 7, 2024
I can go higher with HT off, but that feels like cheating. lol i may try to lower vcore a bit more as I finally got the bclk at 125 to work. on 100mhz, i couldnt get quite as high. my only concern is that the cache ratio refuses to do anything over x27, is it that important? mainly for gaming, otherwise just light duty desktop work.
420mm AIO, Arctic Freezer III. this was not a drop in cooling solution. I mcgyvered the hardware from the 115x kit and the normal included hardware. under load, this stays around 80C after a cinebench run. bios varies in terminology from many guides. ROG x299-e. coming from haswell, this has been a learning curve.

9800x validated.PNG
 
It's been a while but I think 4.9 GHz is about the limit, but I can't remember if that is with or without AVX2 code which might require lower. AVX-512 will almost certainly crash it without hefty offsets.

I didn't play with bus, does that affect the mesh cache speed too? I don't recall above 3 GHz working for me on cache. It helps get data around within the CPU but I don't recall if it affected gaming much.
 
It's been a while but I think 4.9 GHz is about the limit, but I can't remember if that is with or without AVX2 code which might require lower. AVX-512 will almost certainly crash it without hefty offsets.

I didn't play with bus, does that affect the mesh cache speed too? I don't recall above 3 GHz working for me on cache. It helps get data around within the CPU but I don't recall if it affected gaming much.
Avx2 and avx512 are offset-10 from the AI overclock which i started with before going to manual mode, so i left them. I'm u sure what AVX instruction set is even used for beyond prime95, which is do not use. 1. It's an unrealisticly heavy load,(for what I use my PC for) and 2. I find that stopping the stress test is clunky and I just use task manager to force stop it.
I don't want to put any voltage beyond the limits of what will remain stable for 5 years or so. After nearly 10 years on 4790k, either the motherboard, psu or cpu seemed to have degraded and I couldn't maintain 5ghz anymore. Cpu was delidded and lapped, but no liquid metal. Didn't seem to be a thermal limit as it stayed below 70C under full load.
This cpu was just installed as-is due to using solder TIM and the risk for damaging surface mounted components.
I'm far from an expert in overclocking.
I was just hoping for tips from someone experienced with the x299 platform and see if there is anything concerning i should change.
 
The AVX offset is applied to the ratio. If you're running 125 bus, that's 1.25 GHz lower when the CPU gets AVX code. The possible problem there is that AVX is getting used ever more including games. Monitor the CPU clock under use. If you see around your set 4875 then that suggests no AVX, but 3625 could indicate AVX is going on. What version of Cinebench was used? R15 was the last one that didn't use AVX. R20 and newer ones do. If you do find it dropping to the lower clock, you could try manually increasing the AVX2 ratio and test that for stability where you see the drop. You can probably ignore AVX-512 or even totally disable it in bios to make sure it doesn't get used. It might get more wider use since AMD are on their 2nd generation with it, but Intel did a U turn and don't have it in recent consumer products.
 
The AVX offset is applied to the ratio. If you're running 125 bus, that's 1.25 GHz lower when the CPU gets AVX code. The possible problem there is that AVX is getting used ever more including games. Monitor the CPU clock under use. If you see around your set 4875 then that suggests no AVX, but 3625 could indicate AVX is going on. What version of Cinebench was used? R15 was the last one that didn't use AVX. R20 and newer ones do. If you do find it dropping to the lower clock, you could try manually increasing the AVX2 ratio and test that for stability where you see the drop. You can probably ignore AVX-512 or even totally disable it in bios to make sure it doesn't get used. It might get more wider use since AMD are on their 2nd generation with it, but Intel did a U turn and don't have it in recent consumer products.
I'm using CB r23. Hwm and cpu-z show full frequency, but CB just says 3.8ghz. That's the base out of box frequency, so I assume that's all that is displaying.
Xtu shows 3.9ghz max P core. Not sure who to believe. Lol
Post magically merged:

Task manager shows 3.85ghz. I'll have to try in a game.
 
XTU or CPUz...Hwinfo... ;)

All p-cores on this hcip... no need to discern with something that doesn't have the e-cores.
 
3.8 Ghz is the base speed, believe Hwinfo.
Should I believe windows over XTU? Turning SVID on or off makes no difference.
@mackerel i didn't change the avx offsets to -6 if that's worth anything. Lol
Didn't see any difference, but xtu as avx stress tests I'll mess with more.
 
You should believe the apps listed above. :)

Have hwinfo and CB in the window at the same time... note clocks at idle from value column... start CB, note clocks. Those are your clockspeeds. Same thing with XTU and CPUz. Not sure what's up with task manager.
 
3.8 GHz is the CPU base speed, and that is what Windows/CB might be reporting. In task manager the top line where the CPU model is, will be the base speed. Below the core charts could be real time clock. The other tools will show the clock at the time. Double check what the clock is while CB is running.

If I want to check AVX clocks specifically, I use Prime95 since it has the option to disable AVX-512/AVX2/AVX for the stress test.

Also double check what the AVX2/AVX-512 offsets are. I don't know if that is shown in XTU or if you have to go to bios. I ask because -10 was mentioned earlier, now -6?
 
You should believe the apps listed above. :)

Have hwinfo and CB in the window at the same time... note clocks at idle from value column... start CB, note clocks. Those are your clockspeeds. Same thing with XTU and CPUz. Not sure what's up with task manager.
Hw monitor does clock up and down. Not as nice as if like. Instead of stepping, it seems to just jump between 1200mhz and the 4885mhz.
Xtu goes from the 12mhz up to 3900mhz.
I'll just look at hardware monitor and assume it's correct. With the avx offset higher, hwm did bounce around more on frequency is seemed. The changes happen so fast, it's hard to see what is going on.

Hardware info64 shows much much more data and looks correct.
I suspect the 125 bclk is messing with the xtu reading.
Post magically merged:

3.8 GHz is the CPU base speed, and that is what Windows/CB might be reporting. In task manager the top line where the CPU model is, will be the base speed. Below the core charts could be real time clock. The other tools will show the clock at the time. Double check what the clock is while CB is running.

If I want to check AVX clocks specifically, I use Prime95 since it has the option to disable AVX-512/AVX2/AVX for the stress test.

Also double check what the AVX2/AVX-512 offsets are. I don't know if that is shown in XTU or if you have to go to bios. I ask because -10 was mentioned earlier, now -6?
Bios and xtu both show them. I had left it at default-10, but have been changing it to downclock less, but perhaps I'll leave it be. 🤷🏻‍♂️
Seems stable, cool enough and plenty fast enough. I was hoping to get the cpu to download more at idle, but doesn't seem to do that well. I'm sure I'm missing some setting. Maybe disable turbo boost?
 
Bios and xtu both show them. I had left it at default-10, but have been changing it to downclock less, but perhaps I'll leave it be. 🤷🏻‍♂️
Seems stable, cool enough and plenty fast enough. I was hoping to get the cpu to download more at idle, but doesn't seem to do that well. I'm sure I'm missing some setting. Maybe disable turbo boost?
If it works, it works. I think you have to keep turbo on for overclocks to work, at least if you want to go above base multiplier. Again, not sure on this since I haven't done bus overclock in a very long time.

Just to check my sanity, I'm quickly running some things on my 7980XE stock:

Cinebench R15: 3.2 GHz-ish
Cinebench R20: 3.2 GHz-ish
Cinebench R23: 3.15 GHz-ish

P95 AVX-512: 2.6 GHz-ish
P95 AVX2: 2.8 GHz-ish
P95 AVX: 2.9 GHz-ish
P95 no AVX: 3.2 GHz

With P95, use small FFT, and turn off HT for more consistency. Bigger FFTs are ram bandwidth limited and will stress a lot less.

If you wonder why all the "-ish", the clocks were not stable on one value, and typically moved up or down 100 MHz steps. R23 did clock a bit lower than the other two.

Ok, so Cinebench R20/R23 with AVX usage still clocks the same as non-AVX workloads. Interesting. Maybe I'm mis-remembering. It is known that Cinebench doesn't really make heavy use of AVX in those versions, as I've ran it before on CPUs not supporting AVX and it hardly made any difference compared to those that did in the same generation. AVX used to be disabled on Pentium/Celeron tiers, and also got effectively disabled if you did non-K overclock on Skylake. I thought the presence of AVX would be enough to downclock.
 
Last edited:
So, to get close to apples to apples, I downloaded passmark and ran thier cpumark and compared the results to the averaged out scores for the 9800x on passmark website. This should keep any manufacturer/fan boy bias insignificant. Looks like about a 22% uplift over those averages, when averaging each test protocol. I'll take it, and be happy. No more dicking around besides maybe drop the voltages a bit.
 
Back