I thought I’d take a moment to share what I have been working on. I figured there would be at least a few people in this section that find it interesting
I have been non existent here in the HW section this past week or so because I went insane and stepped into the world of 8000mhz! Been tweaking my 9950x3D for the past week. OCd my ram to 8000mhz CL38 and Undervolted my 9950x3D with a per CCD negative offset of -30. Saw a MASSIVE improvement in workstation task and a massive improvement in 1% lows in Gaming. I Hardly have any dips now. Just consistent FPS. Even in the Nightfall event. Its smooth as butter in the giant mob of people.
Due to my undervolt and ram OC a boost in both SC and MC benchmarks. Scored 45.1k @ 190w in Cinebench r23 and 2618 in 2024 @ 198w. Never exceeded 70c! Was scoring 41.9k @ 76c stock and with basic 6200mhz Expo. 2.6k in 2024 is just an insane score for not even having PBO turned on!
As for gaming in WoW Specifically? We know wow is a CPU/Ram Speed based Game. Due to my recent changes Frame dips are now non existent pretty much. Evene in populated areas and Large mobs during world events like Nightfall. I also saw a 6% increase in Average FPS in Wow. As for other games I saw an 8% increase in Average FPS. Some games more some less. and a 25-30% increase in my 1% lows which is just bonkers to me!
My New Ram 8000mhz OC timings https://i.imgur.com/1sYBxaf.png
R23 Scores https://i.imgur.com/KGMNYoi.png
2024 Benchmark https://i.imgur.com/NG6qTKC.png
WoW FPS @4k Max settings (Stormrage Prime Time) https://i.imgur.com/Y8Up6zW.jpg
My Setup https://i.imgur.com/WguchSW.jpg
My Rig https://i.imgur.com/yTaTAmO.jpg https://i.imgur.com/dl7VCar.jpg
Share your thoughts Lilybugg. I know you will see this sooner or later
that’s the thing I didnt OC at all. PBO is disabled. Stock boost. But I get it. Undervolting is the new OC. Just wanted to note that Im still under the stock power limit settings are still enforced and so is the stock boost algorithm
If you havent seen my older post. I have done extensive testing with wow and Ram speeds over various platforms. WoW drinks up Ram Speed like water in the desert. It has a HEAVY effect on your FPS. However didnt think that was really true on the X3D CPUs due to their large cache. That is kind of true but kind of is at the same time. I saw a minor FPS increase. Something no one would notice without an FPS counter running. However the 1% lows? Saw a dramatic improvement. So world events, Raids ect ect? Areas with massive people in 1 spot? Are just soo much smoother.
In other Games I saw a boost in Both. More so the 1% lows though. The difference in the 1% lows just blew me away. Thats all. I wasnt expecting suchs a drastic differance. I wasnt expecting such a large difference in the 1% lows
At 1080p? and possibly even 1440p if you have a good enough Ram/GPU combo no you shouldn’t. Even places like the Nightfall event. 4k MAX settings on a server like Stormrage or Area52 is a differant story.
I was commenting for those who play at 4k MAX settings. During the nightfall event at 4k while fighting the boss it would get a bit hairy and Id hover around 30-40fps just while everyone was spamming aoe on the boss. Especially as Melee. This “Dip” in fps is caused by your 1% lows. However now I don’t dip that low EVER. At the worst of the worst times ill dip to about 55-65 FPS and smooth as butter. No “jitteriness” whatsoever. Then after the boss dies I got back up to 100+
This was more about times of heavy congestion where you have a ton of people in one spot. Thats when Frame dips happen. If they are big enough it will start to look “jittery”. The more you drop below 35 FP the more jittery it starts to look. I should have been more clear on that. I apologize
This is true in wow but ONLY in situations like a World Boss fight for example. MASS people in 1 spot spamming spells. In those situations by 1% lows are night and day better
I should have been more clear about that. Sometimes I dont explain things the best or over explain things more than I should and it makes the point more merky
9900X3D shares its cache across two CCDs, though. Overall it may have more, but each chiplet can only access its own cache - which includes the extra L3 cache. So the one chiplet that’s going to be doing the heavy lifting for a gaming workload will have the same 96MB as a 9800X3D, while the other will have just 32MB.
Writing to the cache of the other CCD would incur pretty significant performance losses. Not only due to the latency between the CCDs but also because it would likely stall the other CCD while performing read/write operations to maintain coherency. This would mean the non-X3D CCD being able to stall the X3D CCD as well as vice versa, which would create a scheduling nightmare.
Just to explain a little better here for folks reading this? The V-cache is on one CCD. When you are gaming the 9900x3D and 9950X3D are a single CCD chip. Because you are only using the v-cache CCD. When gaming other than the slightly higher boost frequency it’s exactly the same as the 9800x3D. However the 9900x3D is slightly worse when gaming because it’s only a 6 Core CCD
This is what so many people do not realize. 9950X3D is only a 16 cord chip when you are out of the game otherwise it’s 8. This is why I tell so many people that if you’re just a gamer the 9950X3D is not an upgrade. You are literally trading an Apple for an Apple.
However the 9950x would work just as you describe. Because it doesn’t Park any of its cores when gaming. So therefore it has a cross CCD latency penalty. However even on that chip you can use something like Process Lasso to Game On One CCD while running all your background test and anything else you would have going on on a second monitor on the other CCD. Which removes the cross CCD latency from the equation.
In my case I already use process lasso. So my just my game runs on CCD0 (vcache CCD) and all my background task and anything else I want to do while gaming runs on CCD1 (frequency CCD). Heck even the wow launcher is running on CCD1.
With process lasso if on the 9950X3D or 9950x you essentially have a 9800x3D and a 9700x in the same rig which once again removes cross CCD latency from the equation
100% correct, via review of the CPU “stacked 3D V-Cache memory is only present on the first 8-core CCD, while the other is a regular one with just the 32 MB on-die L3 cache. AMD continues to rely on software level OS scheduler optimizations to ensure that gaming workloads are scheduled to the CCD with 3D V-Cache, while multithreaded productivity workloads are allowed to spread evenly among the two CCDs”
Yes, yes it does. The 9900X3D has a total of 128MB of cache, but it’s shared across the CCDs as I already explained - 96MB on one, 32MB on the other. As I also already explained, however, these CCDs only have access to their individual caches as it’s not shared *between* them. As such the 9900X3D’s additional cache above the 9800X3D is irrelevant to performance (whether the second CCD is being used or not).
If there’s a better word to use than “share” I’m all ears.
It’s good stuff, I just looked at it today as I’m swamped at work (thanks tariffs). We know the game engine will toss as much as it can on one big core and then move as much as it can on a second big core. Faster ram frequency can reduce latency which we know helps the game, surprised it does so well. There is always a question of “is the low FPS issue an upstream or downstream problem” with MMO games. With big boss fights and large PVP battles it seemed to be a strong mix of both but hard to tell the exact split. Looks like it may be more of a downstream issue then once believed thanks to you.
not according to AMD via an interview on a HardwareLuxx
The just-announced Ryzen 9 9900X3D and 9950X3D, however, have two CCDs—the former with six cores per CCD and the latter with 8 cores per CCD. But the 3D V-Cache sits underneath just one of these chiplets, which means the top-end 9950X3D has the same 64 MB of 3D V-Cache as the 9800X3D, and only 8 of its cores (again, like the 9800X3D) access it.
The benefit of this is twofold. First, the cores on the CCD without the chiplet can boost to a higher clock speed. Second, there are more cores, which is great for applications that require lots of multicore performance. Crucially, with the 9950X3D at least, in addition to these two benefits, we should also get similar gaming performance to the 9800X3D, given it still has eight cores with access to 64 MB of 3D V-Cache.
I think what you are trying to say (I may be wrong) is that the CCD have access to certain levels of cache that is shared and that may be. I can’t recall each cache in the CPU but I know the 64 MB of 3D V-Cache can only be accessed by one and is not shared.
Hes right. It can get pretty confusing without an image to look at.
This image should help explain. The 9800x3D and 9950x3D CC0 are the exact same. CCD1 on the 9950x3D has an additional 32MB of regular L3 cache. CCD1 is 100% nonexistent while gaming https://i.imgur.com/lnrBGGk.png
You are 100% right and I am 100% wrong. This is what I get for reading your post before my coffee. Your wording just left it a bit confusing. I MISREAD your post thinking you were explaining it as while gaming the 9950x3D accesses the cache from the other CCD. However you said it DOESNT. Which is 100% correct. I guess I got your post and Aquoriuspost mixed up
There is a LOT of misinformation about the 9950x3D out there. So I didnt want people reading the thread to get confused or misinformed. Edited my comments to make them more clear.
The image above should help people visualize it
There would have been less confusion if you had stated it doesnt “share cache across CCDs while gaming” You did but your wording was a bit murky hence all the confusion.
I had to reread that post because it comes off stating “it’s shared” and then states “it’s not shared”. I think what we all are trying to say is and to slightly change what Asterchades stated the 9950X3D has a combined total of 128MB of cache followed by his "CCDs only have access to their individual caches as it’s not shared ". The line stating “but it’s shared across the CCDs” makes it confusing but I get what he is saying.