Poll: Ryzen 7950X3D, 7900X3D, 7800X3D

[Audioboxer]

The sweetspot for the 7800X3D from what I've read is 6000MHz with CL30.

I'd say 6200, with 6400 1:1 being up for grabs if you are patient and have a decent CPU IMC.

6200 should be achievable for mostly everyone who puts in a little time. 6600 1:1 is sort of entering unicorn territory, highly IMC (internal memory controller) dependent and will likely need a fair bit of time to tune/tweak. Above 6400 you're just best aiming for 8000 2:1, but that tends to need a 2 DIMM board like the x670 GENE.

You don't need to buy XMP kits at those frequencies, just aim for a-die (unless you want 2x24GB, then get m-die rev 2). There are a-die 6000 XMP kits. They will OC really well, especially just going to 6200~6400.

So tldr; aiming for 6200 should be a goal, then progressing onto 6400 if your CPU is up for it.

I just delidded my 7950x3D and I'm using the thermal grizzly direct die block, it's a damn decent temp drop with liquid metal

That's an all core AVX2 load and then the one that usually gets people near/at thermal limits, an all core SSE load.

You can really see when you delid how the 3D cache CCDs really do run hotter than frequency only CCDs. Hence their frequency restrictions.

 

[LtMatt]

I'd say 6200, with 6400 1:1 being up for grabs if you are patient and have a decent CPU IMC.

6200 should be achievable for mostly everyone who puts in a little time. 6600 1:1 is sort of entering unicorn territory, highly IMC (internal memory controller) dependent and will likely need a fair bit of time to tune/tweak. Above 6400 you're just best aiming for 8000 2:1, but that tends to need a 2 DIMM board like the x670 GENE.

You don't need to buy XMP kits at those frequencies, just aim for a-die (unless you want 2x24GB, then get m-die rev 2). There are a-die 6000 XMP kits. They will OC really well, especially just going to 6200~6400.

So tldr; aiming for 6200 should be a goal, then progressing onto 6400 if your CPU is up for it.

Spot on.

 

[Wing-Man]

with out pulling stick to bits how do you know what die / rank type you have? all i know is i have A die(from cpuz) and and it will not OC in any way unless im not dumping the right voltage settings

 

[Dreadi]

with out pulling stick to bits how do you know what die / rank type you have? all i know is i have A die(from cpuz) and and it will not OC in any way unless im not dumping the right voltage settings

Thaiphoon (sp?) burner, hwinfo and aida64 may give you more info on chips but as long as it's A-die you should be good to go.
If it's 2x16gb they will be single rank

My g.skill runs 6400mhz cl28 with fclk at 2133mhz as does my friends my Corsair A die albeit his runs a fair bit warmer

 

[Noxia]

Id be interested to know just how much improvement comes from tweaking memory.
I am a lazy overclocker, I appreciate some just do it for fun but I wonder how useful this is really. However, I also change components at times almost for the sake of it so not saying things are always done because they make sense!

I'm actually changing my memory just to see if I can enable EXPO (for whatever reason it's called DOCP or something on my mobo).

Old ram is not QVL and doesn't work with expo 1 or 2 it was:
Trident Z5 RGB DDR5-6000MHz CL36-36-36-96 1.35V 32GB (2x16GB)

My new ram is:
CORSAIR DOMINATOR TITANIUM RGB DDR5 RAM 64GB (2x32GB) DDR5 6000MHz CL30.
But confused on timings, site says that tested latency is 30-36-36-76 but SPD latency is 40-40-40-77.

I am really clueless on memory to be honest, it's never something I've really focused on. The reason I'm getting 64 instead of 32 is to make it feel like more of an upgrade too.
I'm not actually expecting any noticeable benefits, it's more because I want to rule out the RAM as the issue for not being able to boot with expo enabled (by expo I mean DOCP).

 

[LtMatt]

Id be interested to know just how much improvement comes from tweaking memory.
I am a lazy overclocker, I appreciate some just do it for fun but I wonder how useful this is really. However, I also change components at times almost for the sake of it so not saying things are always done because they make sense!

I'm actually changing my memory just to see if I can enable EXPO (for whatever reason it's called DOCP or something on my mobo).

Old ram is not QVL and doesn't work with expo 1 or 2 it was:
Trident Z5 RGB DDR5-6000MHz CL36-36-36-96 1.35V 32GB (2x16GB)

My new ram is:
CORSAIR DOMINATOR TITANIUM RGB DDR5 RAM 64GB (2x32GB) DDR5 6000MHz CL30.
But confused on timings, site says that tested latency is 30-36-36-76 but SPD latency is 40-40-40-77.

I am really clueless on memory to be honest, it's never something I've really focused on. The reason I'm getting 64 instead of 32 is to make it feel like more of an upgrade too.
I'm not actually expecting any noticeable benefits, it's more because I want to rule out the RAM as the issue for not being able to boot with expo enabled (by expo I mean DOCP).

Cyberpunk 720P + Low Settings Preset (no RT)
7950X3D + 2200 FCLK
7900 XTX

6000C36 EXPO/XMP timings (primary only)


Vs

8000C34 all timings tuned


12% better 1% low FPS
7.5% better average FPS

Might be a little more difference if I had my 4090 in the system at the time of testing since its faster than XTX in this game.

The biggest difference will usually be in the 0.1% and 1% low FPS from faster memory.

 

[kute]

Cyberpunk 720P + Low Settings Preset (no RT)
7950X3D + 2200 FCLK
7900 XTX

6000C30 EXPO/XMP timings (primary only)


Vs

8000C34 all timings tuned


12% better 1% low FPS
7.5% better average FPS

Might be a little more difference if I had my 4090 in the system at the time of testing since its faster than XTX in this game.

The biggest difference will usually be in the 0.1% and 1% low FPS from faster memory.

That is a legitimate difference. I'm in the poorman's camp of 7800X3D, and I finally upgraded the generic 5600/CL36 sticks from the first month of AM5, to modest 6000/30 corsairs, and I've noticed a nice smoothing out of some stuttery tracks/interiors

I'd be curious to see your difference in 1440p and dare I say it, 4k

Also, man, there's so many fields to change anytime I've had a look at the ram screens in bios. It's intimidating. Are you all Hynix employees or something? I'd love to do some proper settings up beyond expo lol

 

[Noxia]

That is a legitimate difference. I'm in the poorman's camp of 7800X3D, and I finally upgraded the generic 5600/CL36 sticks from the first month of AM5, to modest 6000/30 corsairs, and I've noticed a nice smoothing out of some stuttery tracks/interiors

I'd be curious to see your difference in 1440p and dare I say it, 4k

Also, man, there's so many fields to change anytime I've had a look at the ram screens in bios. It's intimidating. Are you all Hynix employees or something? I'd love to do some proper settings up beyond expo lol

Same. I only play at 4K and this is generally where things seem to have little effect if any at all.

 

[LtMatt]

That is a legitimate difference. I'm in the poorman's camp of 7800X3D, and I finally upgraded the generic 5600/CL36 sticks from the first month of AM5, to modest 6000/30 corsairs, and I've noticed a nice smoothing out of some stuttery tracks/interiors

I'd be curious to see your difference in 1440p and dare I say it, 4k

Also, man, there's so many fields to change anytime I've had a look at the ram screens in bios. It's intimidating. Are you all Hynix employees or something? I'd love to do some proper settings up beyond expo lol

If you are GPU limited, the difference is going to be much less significant. At 4K I expect it to be almost no difference.

I've got the 4090 in my system atm, so I'll re-run the tests at 1080P, etc and let you know.

I'll try a min/max settings difference too, since that should show that at 4K max settings, the difference should be close to nill.

If you can share a zen timings screenshot with me, I can suggest some manual timing improvements. Providing your kit is Hynix and not Samsung.

 

[LtMatt]

Okay just did a few quick tests with the 4090 on Cyberpunk at low vs max settings at 720P/2160P. Details below.

Cyberpunk 720P + Low Settings Preset (no RT) + FSR UP
7950X3D + 4090

6000C30 EXPO/XMP timings (primary only 30-36-36-76) + FCLK 2000Mhz


8000C34 All timings tuned + FCLK 2200Mhz


Summary:
720P
1% Low FPS difference 7.2%
Avg FPS difference 5.3%

----

Cyberpunk 2160P + Max Settings Preset Path Tracing + Ray Reconstruction + DLSS Ultra Performance + Frame Gen = Off
7950X3D + 4090

6000C30 EXPO/XMP timings (primary only 30-36-36-76) + FCLK 2000Mhz


8000C34 All timings tuned + FCLK 2200Mhz


Summary:
2160P
1% Low FPS difference 3.2%
Avg FPS difference 1.2%

Cyberpunk at 4K, even with DLSS Ultra Performance mode and in the benchmark the FPS are mostly under 120 FPS, so it's completely GPU limited. Faster ram makes very little difference in this scenario. Faster ram helps if your GPU is not fully utilised and you are CPU limited. Which happens more at lower resolutions.

 

[Audioboxer]

Wasting tens of hours tuning memory is all part of the fun though

Some games can be more memory bandwidth/latency sensitive than others. 1% lows are especially where you find the smoothness from a good tuning.

XMP is just not something I'd recommend anyone runs. At least not if you care about maximising what your money has bought you. I understand why the JDEC standards exist, but mobo secondary timings on AUTO are just disgusting!

At this point seeing GDM enabled just triggers me as well. Get that nonsense off and stable LOL.

 

[stooeh]

Okay just did a few quick tests with the 4090 on Cyberpunk at low vs max settings at 720P/2160P. Details below.

Cyberpunk 720P + Low Settings Preset (no RT) + FSR UP
7950X3D + 4090

6000C30 EXPO/XMP timings (primary only 30-36-36-76) + FCLK 2000Mhz


8000C34 All timings tuned + FCLK 2200Mhz


Summary:
720P
1% Low FPS difference 7.2%
Avg FPS difference 5.3%

----

Cyberpunk 2160P + Max Settings Preset Path Tracing + Ray Reconstruction + DLSS Ultra Performance + Frame Gen = Off
7950X3D + 4090

6000C30 EXPO/XMP timings (primary only 30-36-36-76) + FCLK 2000Mhz


8000C34 All timings tuned + FCLK 2200Mhz


Summary:
2160P
1% Low FPS difference 3.2%
Avg FPS difference 1.2%

Cyberpunk at 4K, even with DLSS Ultra Performance mode and in the benchmark the FPS are mostly under 120 FPS, so it's completely GPU limited. Faster ram makes very little difference in this scenario. Faster ram helps if your GPU is not fully utilised and you are CPU limited. Which happens more at lower resolutions.

Are you benching Low with Frame Gen on ?



Also why do the developers think it is a good idea for presets in this game to have upscaling set to Auto so making comparisons is less useful!

 

[Eddie99]

Okay just did a few quick tests with the 4090 on Cyberpunk at low vs max settings at 720P/2160P. Details below.

Cyberpunk 720P + Low Settings Preset (no RT) + FSR UP
7950X3D + 4090

6000C30 EXPO/XMP timings (primary only 30-36-36-76) + FCLK 2000Mhz


8000C34 All timings tuned + FCLK 2200Mhz


Summary:
720P
1% Low FPS difference 7.2%
Avg FPS difference 5.3%

----

Cyberpunk 2160P + Max Settings Preset Path Tracing + Ray Reconstruction + DLSS Ultra Performance + Frame Gen = Off
7950X3D + 4090

6000C30 EXPO/XMP timings (primary only 30-36-36-76) + FCLK 2000Mhz


8000C34 All timings tuned + FCLK 2200Mhz


Summary:
2160P
1% Low FPS difference 3.2%
Avg FPS difference 1.2%

Cyberpunk at 4K, even with DLSS Ultra Performance mode and in the benchmark the FPS are mostly under 120 FPS, so it's completely GPU limited. Faster ram makes very little difference in this scenario. Faster ram helps if your GPU is not fully utilised and you are CPU limited. Which happens more at lower resolutions.

Is there much performance difference from 6000C30 with tuned timings and 8000C34 with tuned timings? I remember getting a good performance increase just from 6000C30 EXPO to 6000C30 with buildzoid's easy Hynix timings.

 

[LtMatt]

Are you benching Low with Frame Gen on ?



Also why do the developers think it is a good idea for presets in this game to have upscaling set to Auto so making comparisons is less useful!

I’m sure I turned it off, but I thought the numbers looked a little high. I’ll go and double check it’s still set to off.

Edit - Double checked, it’s the low preset (no RT) so no option to enable FG @stooeh.

 

[LtMatt]

Is there much performance difference from 6000C30 with tuned timings and 8000C34 with tuned timings? I remember getting a good performance increase just from 6000C30 EXPO to 6000C30 with buildzoid's easy Hynix timings.

Really depends on the game and if you are CPU limited. It doesn't hurt put it that way. If you are running 4K, max settings on everything, the difference is going to be small at best. For 1080P/1440P or in any games where GPU utilisation is not pegged at 99/100%, the benefits might be better up to 5%+.

 

[LtMatt]

Are you benching Low with Frame Gen on ?



Also why do the developers think it is a good idea for presets in this game to have upscaling set to Auto so making comparisons is less useful!

Spotted your error, you are running FSR Auto. Set to FSR Ultra Performance!

 

[stooeh]

Spotted your error, you are running FSR Auto. Set to FSR Ultra Performance!

Not an error, deliberate so it does not appear as 'Custom'

I’m sure I turned it off, but I thought the numbers looked a little high. I’ll go and double check it’s still set to off.

Edit - Double checked, it’s the low preset (no RT) so no option to enable FG @stooeh.

Well it shows FG on in your screenshots and Custom

 

[LtMatt]

Not an error, deliberate so it does not appear as 'Custom'


Well it shows FG on in your screenshots and Custom

Ah fair point. Stupid game must have forced it on.

Edit - Confirmed, the game enabled FG for some reason. Damn.

 

[Audioboxer]

Is there much performance difference from 6000C30 with tuned timings and 8000C34 with tuned timings? I remember getting a good performance increase just from 6000C30 EXPO to 6000C30 with buildzoid's easy Hynix timings.

2CCD processors love 8000.

If you're on a 7800x3D I would aim for 6200~6400 before worrying about 8000. Although, 8000 is less intensive on the IMC than 6400 because of 2:1. It's just booting 8000 can be a bit of a ******** on 4 DIMM mobos if you have one of them.

 

[Dicehunter]

Should be interesting

EDIT

Well turns out it was £7 well spent, It dropped temps on average of 3-4'c on a 7800X3D.