Intel to AMD upgrade, re-using 2x24Gb memory?

[exm]

Currently I am running a z790 with the Intel i7 14700k processor. I am expecting my 5090 this week, which means a new PSU, which means.. Why not, let’s upgrade the processor+mobo also to the 9950x3d. But I have a few questions:
- I am running the TeamGroup T-Create Expert 48GB Kit (2 x 24GB) in my Intel setup:
* DDR5-7200 PC5-57600
* CAS Latency 34, Timings 34-42-42-84
From what I learned, AMD really prefers 6000-30 RAM. The memory does not have the Expo profile but I read from a review on Amazon that a user has great luck with this kit and a 7800x3D. Any thoughts on using this RAM? 48Gb is really the sweet spot for me.
- I am planning to get the ASRock X870E Nova WiFi motherboard (I have the Riptide for my Intel and I like ASRock). Any thoughts?

 

[eidairaman1]

Run it

 

[Outback Bronze]

From what I learned, AMD really prefers 6000-30 RAM. The memory does not have the Expo profile

You can just run a custom speed of 6000+ with custom timings if you're interested.

Will summon @ir_cow for you

 

[Countryside]

6000 or 6400 at the best and make sure that bios is updated for better compatibility

 

[FreedomEclipse]

6400@CL30 is good but 6000@CL28 is the holy grail. 6000 being the sweet spot and CL28 being low latency

 

[ratirt]

You can check on the motherboard vendor's site if the ram is compatible. It should work fine though. Bios update might be necessary.

 

[DirtyDingusMcgee]

6400@CL30 is good but 6000@CL28 is the holy grail. 6000 being the sweet spot and CL28 being low latency

Dang. G-skill got some CL26 6000 for cheap right now

 

[FreedomEclipse]

Dang. G-skill got some CL26 6000 for cheap right now

Not if you already have a decent set of ram it ain't.

 

[Cowboystrekk]

That kit is Hynix M-die. One of the best kits for ryzen. Just underclock it to 6000-6400 depending on how good your imc is and tighten up a few timings. At 6000 using xmp voktage that kit should run cl28-rcd35-rp35 without problems. Just set trfc to 500 and trefi to 50000 and you get a great deal

 

[exm]

Thank you all for your replies! Looks like the RAM is re-usable from what I gathered. I did venture in my Intel BIOS a while back and there a ton of options. I assume that the ASRock X870E Nova is similar. Is there a guide (since I don't have the Expo profile to start with) which settings to use as a baseline, and which settings to try to tweak further?

For the record: I am running this at the 7200 XMP Profile on my 14700k

 

[JustBenching]

Currently I am running a z790 with the Intel i7 14700k processor. I am expecting my 5090 this week, which means a new PSU, which means.. Why not, let’s upgrade the processor+mobo also to the 9950x3d. But I have a few questions:
- I am running the TeamGroup T-Create Expert 48GB Kit (2 x 24GB) in my Intel setup:
* DDR5-7200 PC5-57600
* CAS Latency 34, Timings 34-42-42-84
From what I learned, AMD really prefers 6000-30 RAM. The memory does not have the Expo profile but I read from a review on Amazon that a user has great luck with this kit and a 7800x3D. Any thoughts on using this RAM? 48Gb is really the sweet spot for me.
- I am planning to get the ASRock X870E Nova WiFi motherboard (I have the Riptide for my Intel and I like ASRock). Any thoughts?

It will run just fine, you can even enable XMP (I used a 7600c34 kit on my 9800x 3d), the problem is you need to manually tune it to run at 6000c30.

 

[ir_cow]

You can just run a custom speed of 6000+ with custom timings if you're interested.

Will summon @ir_cow for you

Im not worthy of a summon lol

In any case if your using a 9800X3D, buy whatever ram is the prettiest because it sure does not matter for gaming.

1920 x 1080

3840 x2160

Rendering and synthetic benchmarks take a hit when using JEDEC and rendering. Take note that the blender render is already 20 minutes long. Not a big deal.

 

[exm]

Im not worthy of a summon lol

In any case if your using a 9800X3D, buy whatever ram is the prettiest because it sure does not matter for gaming.

Will get the 9950x3d for 4k gaming and Adobe work.

Interesting. Should I even 'downclock' the memory from 7200 to 6000Hz (default 34-42-42-84)? Should I attempt 8000Hz or is that pointless? I'm a bit confused with the overall recommendation to run memory at 6000Hz (again, coming from the Intel 'world').

 

[ir_cow]

If your using a 9950X3D, than it might be beneficial to use lower speed because half the cores doesn't have access to the 3D cache and is basically just a Ryzen 9700X.

The reason everyone is suggesting to run 6000 is because of the IMC to Memory Ratio. The best IMC can only do 6600, with most toppings out at 6400. Once you switch over to the equivalent to Gear 2 (Intel), latency is a killer and will perform worse in gaming until around 8000 MT/s just to take 6000 perf.

But since your gaming at 4K and using the x3D cache, it doesn't matter the memory. In Adobe is does not matter much either as generally work stuff doesn't need fast memory anyways. Just a lot of memory. I rather have 128GB at 4800 instead of 32GB at 6400 for work related activities.

 

[exm]

If your using a 9950X3D, than it might be beneficial to use lower speed because half the cores doesn't have access to the 3D cache and is basically just a Ryzen 9700X.

The reason everyone is suggesting to run 6000 is because of the IMC to Memory Ratio. The best IMC can only do 6600, with most toppings out at 6400. Once you switch over to the equivalent to Gear 2 (Intel), latency is a killer and will perform worse in gaming until around 8000 MT/s just to take 6000 perf.

But since your gaming at 4K and using the x3D cache, it doesn't matter the memory. In Adobe is does not matter much either as generally work stuff doesn't need fast memory anyways. Just a lot of memory. I rather have 128GB at 4800 instead of 32GB at 6400 for work related activities.

Thanks, that makes sense! Guess 6000 with the lowest possible latency it is.

 

[ir_cow]

Honestly I would just enable XMP at 7200 and not worry about it at all.

 

[Cowboystrekk]

Honestly I would just enable XMP at 7200 and not worry about it at all.

2:1 at 7200 would be equal to 5600 1:1 and he will lose several percent performance then. At around 6000 1:1 the io-die and infinity fabric gets to be the bottleneck. You can get some improvements at 6200, 6400 etc, but at 6000 you get the most. 7600 2:1 is about equal to 6000. Write scores still scales above 6000. One important thing is to raise fclk. It defaults to 2000, but all are able to run it at 2100-2233. This will improve read and copy aswell. 2000 maxes out at around 63k, 2200 at 70k so over 10% improvement to bandwith. That matters. As for tuning a 9800X3D ypu can get good gains in certain games.

Examples here: Going from 6000 expo to 7600 tuned with a small allcore oc improves fps in cyberpunk på 13% and rarkos by 12% when cpu bound which you can be if using dlss.

 

[Zach_01]

Thank you all for your replies! Looks like the RAM is re-usable from what I gathered. I did venture in my Intel BIOS a while back and there a ton of options. I assume that the ASRock X870E Nova is similar. Is there a guide (since I don't have the Expo profile to start with) which settings to use as a baseline, and which settings to try to tweak further?

For the record: I am running this at the 7200 XMP Profile on my 14700k

You got yourself a hand-full since the kit does not support EXPO/XMP at 6000-6400.

Not a big deal if this is your game. Just prepare yourself (with loads of patience) for a lot of restarts and memory training.

Keep in mind that X3D variants are a lot more insensitive on DRAM speed and latency.
The extra “close to core die” cache is compensating for the most if not all the caveats that the memory subsystem may have.
1. Just stay on 1:1 on the UCLK:MCLK, what ever that maybe (6000-6400), meaning for 6000 UCLK:MCLK = 3000:3000
2. Stay minimum at 2:3:3 at the FCLK:UCLK:MCLK

Example:
UCLK:MCLK = 3100:3100 (for 6200MT/s)
FCLK should be at least 2066 (can try higher and test)

As I said don’t expect miracles from let’s say 2000:3000:3000 to 2133:3200:3200.
Even if you do see better results on memory benchmarks like AIDA it doesn’t mean you will see similar better results on games.
3DVcache is powerful as it is on apps and games that are getting benefited from it.
But feel free to try for the fun of it.

 

[Wirko]

Just set trfc to 500 and trefi to 50000 and you get a great deal

Why play with refresh timing when there's data integrity at risk? I mean, I would too, also while trying to reach the maximum operating temperature in the case, but in the end I'd set it back to stock.

 

[Cowboystrekk]

Why play with refresh timing when there's data integrity at risk? I mean, I would too, also while trying to reach the maximum operating temperature in the case, but in the end I'd set it back to stock.

I have tuned a lot of systems and trefi at 50k seems to be doable without file corruotion. 55k+ is less safe.

 

[exm]

Thanks all, I decided to pickup the Patriot Viper Venom 64GB (2 x 32GB) DDR5 6000 with 30-40-40-76 timings for a great deal. I think after selling my old kit, I only need to add $25 or so. This gives me memory with the Expo profile, so no hassle and 16 Gb more. Sometimes the easiest solution is to simply upgrade

 

[ir_cow]

$25 for extra 16GB of memory is a pretty good deal.

 

[exm]

$25 for extra 16GB of memory is a pretty good deal.

Plus (hopefully) much easier to get stable!

 

[ir_cow]

7200 wasn't going to be a problem. It would run at UCLK/Memory/2 (Gear 2) by default. Also XMP vs EXPO isn't really a issue either. EXPO has a few extra sub-timings like tFAW. However the Motherboard with XMP will set these same sub values.

If anything 6000 2X64GB will be harder on the CPU IMC. Now you have dual rank memory in 1:1 where as previously it was single rank in 2:1 mode.

 

[exm]

7200 wasn't going to be a problem. It would run at UCLK/Memory/2 (Gear 2) by default. Also XMP vs EXPO isn't really a issue either. EXPO has a few extra sub-timings like tFAW. However the Motherboard with XMP will set these same sub values.

If anything 6000 2X64GB will be harder on the CPU IMC. Now you have dual rank memory in 1:1 where as previously it was single rank in 2:1 mode.

That goes way over my head! Using stock settings on my new memory, what does 'harder on the CPU IMC' mean?