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

Intel i7-9800x mini review

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
Was kinda hoping for more of a gap in the 2048k results. Will have to crunch the numbers later to see where each scenario sits, but I suspect it may not be as big a gap due to the fast ram in there.
 
Was kinda hoping for more of a gap in the 2048k results. Will have to crunch the numbers later to see where each scenario sits, but I suspect it may not be as big a gap due to the fast ram in there.

Here is a stock run. The ram is running at stock specs (3200MHz 14-14-14-34) and it doesn't look like there is a real AVX offset so the CPU ran the test at ~4.1GHz:
View attachment 202838
 
Comparing the OC and stock P95 results for now. For starters, we're expecting ~5% increase from core clock, and ~17% in ram bandwidth. Looking at 64k 8 core 8 worker, we see near enough the 5% expected due to core clocks, as this is a small size not affected by ram speeds.

Moving to 2048k, I'm not so sure what's happening. 8 core 1 worker went up 11%, and 8 core 8 workers went up 25%. For those not familiar with the workers, see 8c1w as a single multi-thread task. 8c8w are 8 individual single thread tasks. 1w means threads can share cache, 8w each has their own needs. Given the increases are greater than 5%, this can't be due to core clocks alone, and they are ram limited. But 1w is short of expectations if it were purely ram limited, and 8w smashes expectations. If you stack both ram and core increases that comes to 22%, so not far off, but I'm not sure this is the case because if you're ram limited the core shouldn't make much difference.

A single 2048k worker has 16MB working data, and past experience shows performance is usually great if the L3 is at least the same size. But here we're still clearly seeing ram bandwidth effects. There is also some quantity of look up data, which didn't seem to matter but it may do in this case. What cache speeds were used? I see 3000 in the OC screenshot, and 2400 at stock, were these used during the runs?
 
Comparing the OC and stock P95 results for now. For starters, we're expecting ~5% increase from core clock, and ~17% in ram bandwidth. Looking at 64k 8 core 8 worker, we see near enough the 5% expected due to core clocks, as this is a small size not affected by ram speeds.

Moving to 2048k, I'm not so sure what's happening. 8 core 1 worker went up 11%, and 8 core 8 workers went up 25%. For those not familiar with the workers, see 8c1w as a single multi-thread task. 8c8w are 8 individual single thread tasks. 1w means threads can share cache, 8w each has their own needs. Given the increases are greater than 5%, this can't be due to core clocks alone, and they are ram limited. But 1w is short of expectations if it were purely ram limited, and 8w smashes expectations. If you stack both ram and core increases that comes to 22%, so not far off, but I'm not sure this is the case because if you're ram limited the core shouldn't make much difference.

A single 2048k worker has 16MB working data, and past experience shows performance is usually great if the L3 is at least the same size. But here we're still clearly seeing ram bandwidth effects. There is also some quantity of look up data, which didn't seem to matter but it may do in this case. What cache speeds were used? I see 3000 in the OC screenshot, and 2400 at stock, were these used during the runs?

I think cache speed is playing a huge role in the differences. Stock mesh speed is 2400mhz and the overclock run was at 3200mhz I believe. I can do more runs for you hopefully this weekend if you want to see something.
 
Was 3200 the limit for cache? My 7800X would only do 3000 and 3100 was no boot at all. I also noted the stock (auto) mesh clock changed with later mobo bios going from 2000 to 2700.

Tanks again for the testing. I will try to put some perspective on this against my 7800X later.

The other request I had was y-cruncher 25m and 1b if you're familiar with those? Same as used on hwbot. Last time I looked I held the WR for 6 core CPUs. Fancied a go at 8 at some point too :D Note it uses AVX-512.
 
I can try y-cruncher later as well. I haven't tried higher than 3200mhz mesh wise, it just seemed like a good number to shoot for.
 
Bump, I went ahead and ran benchmarks from work while over teamviewer to finish up the "review." I was able to run Y-Cruncher as well for you @mackerel but I had to lower the AVX offset again so really it was running at stock clock speeds with the mesh/memory overclocked.

Over all I'm less than enthusiastic about this CPU. Looking at Reddit there is another person with a 9800x who is "stable" at 4.9GHz and less than 1.3v with seemingly better temps than I on a similar AIO. Perhaps I just got a dud. I regret going from my 7920x to this CPU as I ran the 7920x at 4.4GHz every day with 1.12v and only a -3 AVX offset. Perhaps I will go back that route, I am undecided as of now. If anyone has any suggestions in terms of clocking this CPU, I am all ears however.
 
Y-cruncher uses AVX-512 so it's the other offset. It is pretty punishing. On my 7800X I could just about run it at 4.5 bench stable. The longer 10b test needed to be reduced more.
 
Yeah this 9800x is no where near stable for AX-512 at those speeds without me adding a ton of volts.
 
I never got to ton of volts as I hit the thermal limit before that... AVX-512 is such a beast the test version of Prime95 supporting it already results in scary temps at stock never mind looking at OC.
 
I never got to ton of volts as I hit the thermal limit before that... AVX-512 is such a beast the test version of Prime95 supporting it already results in scary temps at stock never mind looking at OC.

Looks like we'll all have to grow content with hefty AVX offsets if that instruction set is added to Intel's gaming CPUs.
 
Last edited:
Looks like we'll all have to grow content with hefty AVX offsets if that instruction set is added to Intel's gaming CPUs.

What's a gaming CPU? From today's Intel Architecture Day notes, it sounds like it is being added to the range of consumer CPUs out next year. Well, that's only half the story... AVX-512 has multiple implementations, and one part is if the hardware has one or two execution units (per core). Skylake-X has two, which is in part what gives it the massive performance boost. The consumer desktop versions are expected to only have one unit, thus half the performance in that area, but also means it wont hit the power budget so hard. To my uses, a single unit is no better than AVX2 in throughput, but there may be other reasons it can give some advantage in other cases.
 
I am about to exchange my 7800X for a 9800X or 7900X. I suffer from my 7800X temperatures since it is not delidded and I am not going to delid it.

One thing I find quite curious is, why is my ram writing speed way faster than my reading speed?

And which CPU do you guys think is better? 7900X or 9800X? I have excellent cooling (custom loop with 2X 360 radiators).

My results:

View attachment 212196


I didn't feel like creating a new topic since this one is quite useful to be honest, very few 9800X results on the internet, no one reviewed it (Linus, GamersNexus, etc...)
 
Back