Overclock.net banner
21 - 40 of 9,494 Posts
So what's up with this result from HUB? I can see the cheap boards dropping clocks, but not this much variation in the higher end boards. Differences aren't huge, but still I would think they would be the same?

Azure Font Rectangle Screenshot Parallel


This is interesting too. He only ran 1 hour of P95, but gives a decent indication of which boards may currently be capable.

Product Azure Font Rectangle Aqua


 
Discussion starter · #26 ·
Very interesting. Never considered what ‘processor level’ really even meant. Is 9 just considered the highest tier? By what standards/metrics? And if 9800x3d is 9, what will 9900/9950x3d be?
9 as well I'd say. They will also most likely some time next year release lower tier X3D parts. Like the done with the 5000X3D series chips.
 
Implies the part is unlocked, but little else.

I would expect the most aggressive X3D "turbo mode" to max out boost clock override, apply a significant negative CO, loosen power limits, push FCLK to ~2200, and set a bunch of aggressive memory timings...which could amount to a 35% increase in some cherry picked tasks.
But in most cases just crash randomly, I suspect.
 
Implies the part is unlocked, but little else.

I would expect the most aggressive X3D "turbo mode" to max out boost clock override, apply a significant negative CO, loosen power limits, push FCLK to ~2200, and set a bunch of aggressive memory timings...which could amount to a 35% increase in some cherry picked tasks.
Pretty sure they pulled that down. Probably for good reason.
 
This 9800X3D chip should be a beast, can’t wait to see what it does to the 285K. I haven’t ran an AMD system since the olden days, that just might change. I’d probably be more inclined towards the 9950X3D personally, but either one is going to be great.
 
  • Rep+
Reactions: bmagnien
Discussion starter · #33 ·
Kinda have the feeling Gigabyte was trying to give the impression they discovered a secret tweak for their motherboards that up'ed gaming performance, while in reality it comes directly from AMD (?)
Anyway, will be fun to test :)

My guess is that they reused the same mechanism that, on X3D models, forces a videogame to run on the single X3D CCD (that is, an automatic "game mode" power profile signal that gets sent to the firmware that changes the CPPC availability for that specific process only, when in foreground) to disable SMT and force the task execution to a single CCD for non-X3D models too.

That would reliably provide "some" performance improvement to a few videogames that don't need more than 6-8 cores at the same time. But a 35% delta may hardly be reached in that way (the average game fps may improve at most by a 5% only with SMT off on a 7800X3D) so they may have also added some further power-states optimisations thanks to the lower power requirements when disabling SMT.

I tend to agree with Melan that if they went down the generic automatic ram/fclk/cpu freq oc path instead, it would just cause instability to random cpu bins and be counterproductive to their brand, so I don't think they did that.

Actually, if you have a 9950X, you can validate my guess on this improvement by running some videogames benchmarks while using ProcessLasso to force execution on a single CCD only, and also disable the SMT-adjacent core on that CCD.
 
Discussion starter · #37 · (Edited)
This is a 30% improvement over the 19.5k that I’d consider edge of stability CO oc for the 7800x3d in cb23 mt (without BCLK oc). That is…awesome?

+ the hwinfo SS from that same run just to keep everything in one place
Table Window Building Wood Chair
 
This is a 30% improvement over the 19.5k that I’d consider edge of stability CO oc for the 7800x3d in cb23 mt (without BCLK oc). That is…awesome?

+ the hwinfo SS from that same run just to keep everything in one place
View attachment 2677661
That Dawntrail result is higher than anything I've seen as well. Maybe nobody with 7000x3d has made a serious run at it though. Best I can do with my 14900k is 53k. And that's at 6.4Ghz.

Somebody needs to run this with a good 7000x3d rig. :)

Or just wait for reviews in a week or 2...
 
Discussion starter · #40 · (Edited)
That Dawntrail result is higher than anything I've seen as well. Maybe nobody with 7000x3d has made a serious run at it though. Best I can do with my 14900k is 53k. And that's at 6.4Ghz.

Somebody needs to run this with a good 7000x3d rig. :)

Or just wait for reviews in a week or 2...
this is fairly tuned 7800x3d + 4090, -20 co, 8000c34 2x24gb. scored over 30k in Port Royal just before running this bench so about the most I've been able to squeeze out of this hardware combo. fresh install of 24h2 latest LCU. i think i matched the settings? just chose 1280x720 for resolution and only changed first drop down in graphics settings to the standard (laptop) preset, didn't touch anything else. that 9800x3d score is a 28.8% improvement, so like spot on to the other comparables (cb23 mt is 28.9% higher than my best, cheesed 7800x3d score of 19594 at -40 CO) . This is actually insane

Font Screenshot Multimedia Electronic device Display device


Product Font Computer Screenshot Software


 
21 - 40 of 9,494 Posts