6400c30 vs 8000c36 Ryzen 9800X3D

[Cowboystrekk]

I have tested these settings vs each other and they perform quite similar. However 8000 is much better due to one thing: it requires far less soc voltage due to 2:1 so consumption, temps and noise on 8000 is far superior. If you have Hynix A or M 24gb, just do it or aim for 7800 if 8000 is not possible.

8000 needs 890mv soc vs 1160mv soc for 6400. 8000 gets 0.5ns better aida, identical FF: dawntrail and same geekbench 6 as 6400.

 

[Waldorf]

lol, guess you never heard of variations in silicon, especially with ryzen IMC/Soc.
telling folks to universally apply stuff, just because it works for you? priceless.

 

[Cowboystrekk]

lol, guess you never heard of variations in silicon, especially with ryzen IMC/Soc.
telling folks to universally apply stuff, just because it works for you? priceless.

I know there is a lot of variation, but in most scenarios 2:1 at 7800+ will perform similar to 6400, but require quite less soc voltage

 

[Waldorf]

while gimping everything as you have lower bus clocks.
ignoring its wasting more power than going 1:1 (6000 or less) and switching between power profiles (e.g power saving plan limited to 50% max cpu perf).

i have yet to see anyone uv/uc any 5xxx chip where it consumes less power than me running 3600/1800 on IF/ram with 1.1v soc but on power saving, even enough for playing old games, something even stock (as i n jedec) cant do, and compared to other running reduced bus/ram clocks with less soc save around 5-10w, while its ~50w less for me (on power savings plan).

pll need to stop wasting tine to shave off 10 maybe 20w thru stuff like soc V, and take 2 min to tweak two different power plans (low load;game/work load), and reduce power/heat by a lot more, heck you can even find guides to make 2 "shortcuts" to switch between plans for those that dont have a UPS (e.g. no visible battery icon in tray to switch).

 

[Cowboystrekk]

With a power tuned 9800x3d with co and fps cap in games I dont feel I have a need for separate powerplans and have to switch. I prefer an allround profile. I use balanced power profile with a few adjustments. In warcraft 1 remastered for instance cpu uses around 25-30W. In expedition 33 it is usually around 40-50W to maintain 120fps cap. In cinebench r23 it maxes out around 105W holding 5.22GHz allcore. My point of making this tread was that you can get same performance at quite a bit less power, temp and noise. Noise is the main factor for me. At high load I see a drop of over 20W. At idle cpu uses 16-17W vs 26-27W at expo default (1.2v soc and 1.05v vddp).

I tuned my previous 5800x3d with 3800/1900 at 0.98v soc and a 60W cap when mining. It was extremely efficient and temps in low 50s during load, probably not far of your tuning even with various limits.

 

[ir_cow]

And here I am at 1.2-1.3 SoC all the time .

Have you tried any stability test and y-cruncher 2.5B? I want to see 0.890 V in action. I don't think my computer would even boot below 1.1v

 

[freeagent]

And here I am at 1.2-1.3 SoC all the time .

1.285v for me, and don't let me tell you about my high VDDGs

 

[Cowboystrekk]

And here I am at 1.2-1.3 SoC all the time .

Have you tried any stability test and y-cruncher 2.5B? I want to see 0.890 V in action. I don't think my computer would even boot below 1.1v

I ran ycruncher vt3, cinebench 10 min and testmem 5 usmus cfg Friend of mine runs his 9950X3D at 0.96v soc at 8000 totally stable. 8000 generally requires low soc, but some motherboards/cpus cant handle it. 7800 and even 7600 may be viable options

And here I am at 1.2-1.3 SoC all the time .

Have you tried any stability test and y-cruncher 2.5B? I want to see 0.890 V in action. I don't think my computer would even boot below 1.1v

What do you run ram at? I need far higher in 1:1.

1.285v for me, and don't let me tell you about my high VDDGs

I need 0.92v iod and 0.9v ccd at 8000. At 6400 default 0.85v works fine.

 

[freeagent]

I run 2200 IF so the extra IOD helps me out there. 1.065

 

[Sol_Badguy]

DDR5 Memory Performance Scaling with AMD Zen 5

Our AMD Zen 5 Memory Scaling Review examines six different DDR5 memory speeds, including DDR5-8000, DDR5-6400 1:1, DDR5-6400 CL28 and others. We looked into how these speeds impact performance in gaming at four resolutions and a wide range of application workloads.

www.techpowerup.com

 

[ir_cow]

What do you run ram at? I need far higher in 1:1.

Whatever comes to the review table. Last one was 8000 MT/s

 

[cadaveca]

And here I am at 1.2-1.3 SoC all the time .

Have you tried any stability test and y-cruncher 2.5B? I want to see 0.890 V in action. I don't think my computer would even boot below 1.1v

Kits With 5600 JEDEC make me have this low 0.900 SOC voltage by default, while 4800 MHz kits give me 1.05v default. You haven't noticed this? I also run 8000 @ 0.95v, was confused when I got 6000 MHz kit and need so much, can't even push 6400 1:1.

 

[ir_cow]

@cadaveca hmm I have not looked that closely into since I stopped reviewing motherboards. I can load up a few kits and see what I get as default.

 

[Cowboystrekk]

Whatever comes to the review table. Last one was 8000 MT/s

Strange that you needed so much. Uclk usually dicates soc requirement and 2000 uclk is very low.

 

[RootinTootinPootin]

for 1:2 you don't need High SOC voltage, I only use 1.1v for 2200 Fclk..

I'd choose 1:2 as it uses lower SOC voltages which could kill these fragile chips..

@Cowboystrekk
You can try running them GDM off..Nice timings BTW..

 

[Cowboystrekk]

for 1:2 you don't need High SOC voltage, I only use 1.1v for 2200 Fclk..

I'd choose 1:2 as it uses lower SOC voltages which could kill these fragile chips..

@Cowboystrekk
You can try running them GDM off..Nice timings BTW..

Thx! Imc is strong but bin on ram is avg. It does quite nice cl and rfc, but struggles with rcd and rp. Also hates above 1.5v vdd. Old Hynix A from mid 2023. I tried 8200 and I know cpu/mb can handle it as another kit ran it fine. But my current kit struggles with very relaxed timings at 8200 so far. I havent tried gdm off yet, just finish3d finding right voltages, min timings etc. My experience is that gdm off requures atleast 30mv higher vdd. I'm having thermal issues already so uncertain it is doable.

 

[RootinTootinPootin]

Thx! Imc is strong but bin on ram is avg. It does quite nice cl and rfc, but struggles with rcd and rp. Also hates above 1.5v vdd. Old Hynix A from mid 2023. I tried 8200 and I know cpu/mb can handle it as another kit ran it fine. But my current kit struggles with very relaxed timings at 8200 so far. I havent tried gdm off yet, just finish3d finding right voltages, min timings etc. My experience is that gdm off requures atleast 30mv higher vdd. I'm having thermal issues already so uncertain it is doable.

CL36 should be fine, if you're not chasing world records that's already good. spending more money on RAM with AMD is kinda pointless, unless you're going for a higher capacity, that is the only time it would make sense to spend more money on that platform.

 

[Cowboystrekk]

CL36 should be fine, if you're not chasing world records that's already good. spending more money on RAM with AMD is kinda pointless, unless you're going for a higher capacity, that is the only time it would make sense to spend more money on that platform.

Yeah. I got my kit really cheap used for 90usd which is very chrap in Norway. Same kit costs 140usd new here.

 

[Waldorf]

@Cowboystrekk
most ppl use their computer outside of gaming, thus dont need 8-16 C to run at full speed.
using a power plan allows to drop +50w in power consumption easily, without any stability testing, and can be used by virtually everyone on any hw.

my problem is with folks spending days/week and endup saving a few watt, when power plan takes 10 min max saving double most likely more.

 

[Degreco]

I have tested these settings vs each other and they perform quite similar.

And they should. There is not much RAM scaling or other difference (JEDEC or XMP voltage differences aside - as you already said) for the 9800x3D or any other x3D CPU.

Because the whole x3D technology is designed to avoid ram operations (if possible). And that huge chunk of L3 cache is there to make sure that happens more often.

Update:
@AVATARAT : Do you have a hotkey for those emojis, or is it just muscle memory at this point?

Anyway, chew on this and come back with a real response. If it’s just another emoji, save us both some time.

 

[Cowboystrekk]

And they should. There is not much RAM scaling or other difference (JEDEC or XMP voltage differences aside - as you already said) for the 9800x3D or any other x3D CPU.

Because the whole x3D technology is designed to avoid ram operations (if possible). And that huge chunk of L3 cache is there to make sure that happens more often.

Update:
@AVATARAT : Do you have a hotkey for those emojis, or is it just muscle memory at this point?

Anyway, chew on this and come back with a real response. If it’s just another emoji, save us both some time.

View attachment 407796

It depends on the game. Some games get good scaling by tuning/oc ram even on 7800x3d/9800x3d:

up to 14% higher avg fps on 9800x3d. Around 2-4% in that test can be attributed to 200MHz higher clpckspeed on 9800x3d.

 

[Degreco]

Everything above 6400 MT/s like shown in the linced YT video is not running with IF 1:1, introduces higher latencies in a lot of situations and you get stability issues on top. Even if the occasional game benefits a couple of frames, most modern game engines do not.

I quote Igor on his last mem scaling benchmark on AMD from may 2025:

"In gaming, RAM is less relevant with these CPUs because the core components of game engines are relatively small and can be largely buffered with the 96 or 128 MB L3 cache"

If you are talking about applications and huge data requests from RAM then...but you are talking about gaming, where there is not a great difference. I trust the big reviewers like TPU, GN, Techspot/HUB, Igor etc. more on this than some random YT test.

This is the source of the quote - a good read on the topic:
https://www.igorslab.de/en/amd-ryze...-vs-expo-vs-jedec-in-synthetics-and-gaming/8/

This is SOTT with 9800x3D and RTX4090.

Yes, one could argue he should have tested with 720p or 1080p, but this is ela seasoned expert, I trust him on the analysis more than a random YT dude (sorry). As I trust both Steves from GN and HUB, as they make the very same argument: The sweet spot holds.

 

[Cowboystrekk]

Everything above 6400 MT/s like shown in the linced YT video is not running with IF 1:1, introduces higher latencies in a lot of situations and you get stability issues on top. Even if the occasional game benefits a couple of frames, most modern game engines do not.

I quote Igor on his last mem scaling benchmark on AMD from may 2025:

"In gaming, RAM is less relevant with these CPUs because the core components of game engines are relatively small and can be largely buffered with the 96 or 128 MB L3 cache"

If you are talking about applications and huge data requests from RAM then...but you are talking about gaming, where there is not a great difference. I trust the big reviewers like TPU, GN, Techspot/HUB, Igor etc. more on this than some random YT test.

This is the source of the quote - a good read on the topic:
https://www.igorslab.de/en/amd-ryze...-vs-expo-vs-jedec-in-synthetics-and-gaming/8/

This is SOTT with 9800x3D and RTX4090.
View attachment 407820

Yes, one could argue he should have tested with 720p or 1080p, but this is ela seasoned expert, I trust him on the analysis more than a random YT dude (sorry). As I trust both Steves from GN and HUB, as they make the very same argument: The sweet spot holds.

I'm not saying that 8000 is faster than 6400, but ram tuning with lowering rfc and raising refi will be beneficial and in some games give you 10%+ fps. If you can run 8000 stable it will usually be preferable over 6400 due to requiring much lower soc voltage. Similar performance but lower temps.

 

[AVATARAT]

@Cowboystrekk
This is very interesting information, I updated to the latest bios/agesa yesterday and tested it, before I needed 1.275v vSoC for FCLK 2200 (6200MT/s), now it can run with 1.2v (I need to test lower).
So AMD improved it, thanks for the info!

On topic, 6400 offer a bit lower latency, 8000 offer a bit bigger bandwidth.

Some games benefit from higher bandwidth, others from lower latency and it's vice versa in different games/programs, and we're talking about few FPS, so under 0.5-1% difference that's fine for benchmarking but irrelevant for 24/7. Yes, when the benefit is for 0.1% Low, then that's nice.

And they should. There is not much RAM scaling or other difference (JEDEC or XMP voltage differences aside - as you already said) for the 9800x3D or any other x3D CPU.

Because the whole x3D technology is designed to avoid ram operations (if possible). And that huge chunk of L3 cache is there to make sure that happens more often.

Update:
@AVATARAT : Do you have a hotkey for those emojis, or is it just muscle memory at this point?

Anyway, chew on this and come back with a real response. If it’s just another emoji, save us both some time.

Yes, the x3D cache runs with low latency as long as there is data in the cache, if there isn't it adds ~7ns (the time it searches the x3D cache as AMD explains) to the operations with missing data, so X+7, which is slower than CPUs without cache.
If you use Aida 64, you'll see that x3D are ~5-7ns slower than non-x3D with the same timings, so yes, as you said, if possible. There are a few games where the game logic is quite small and can fit in the 3d cache and it boosts FPS by very high values.

Here's an example of those things:

 

[dgianstefani]

"High" SOC is only really an issue if you have to keep the overall CPU package power down for some reason. I basically run 1.3 V 24/7, but I can do that because I have good cooling, and doing so doesn't take away from the core power budget.

What's interesting is going too high will actually reduce stability etc. So there is a sweet spot. Usually between 1.2 and 1.3.