DDR5 7600 with 9800x3d

[Dreg]

Hi all,

 

I currently have a 13900KS with the TeamGroup DDR5 7600 CL36, https://www.newegg.ca/team-group-32gb-ddr5-7600/p/N82E16820331930

 

Am planning to get a 9800x3d and was wondering how memory works with Ryzen 9000 series.

 

As far as the motherboard, am planning to go with the Asus Strix X890E.

 

Would I have to run the kit at a lower speed or 7600 would be fine? 

Thanks

[RONOTHAN##]

54 minutes ago, Dreg said:

Would I have to run the kit at a lower speed or 7600 would be fine? 

Yes and no. Yes 7600 should work, but you would want to run it at a lower speed. For reference, AM5 has two memory controller modes, 1:1 mode (similar to Gear 1 on Intel) and 2:1 mode (similar to Gear 2), where 2:1 sacrifices memory latency for increase clock speed. The memory latency penalty of 2:1 mode is big enough that until you get to DDR5 8000, you're usually better off running 1:1 6000 CL30 unless all you run is Y-Cruncher. 2:1 mode is also open to weird inconsistent behavior, so outside of screenshots and competitive benchmarking, I don't really see the point in running it over a decent 1:1 setup. 

 

Besides, you're going for a 9800X3D, the entire point of the X3D chips is to make memory performance irrelevant, so even if the higher speeds were faster it's not like you'd actually notice. 

[Dreg]

12 hours ago, RONOTHAN## said:

Yes and no. Yes 7600 should work, but you would want to run it at a lower speed. For reference, AM5 has two memory controller modes, 1:1 mode (similar to Gear 1 on Intel) and 2:1 mode (similar to Gear 2), where 2:1 sacrifices memory latency for increase clock speed. The memory latency penalty of 2:1 mode is big enough that until you get to DDR5 8000, you're usually better off running 1:1 6000 CL30 unless all you run is Y-Cruncher. 2:1 mode is also open to weird inconsistent behavior, so outside of screenshots and competitive benchmarking, I don't really see the point in running it over a decent 1:1 setup. 

 

Besides, you're going for a 9800X3D, the entire point of the X3D chips is to make memory performance irrelevant, so even if the higher speeds were faster it's not like you'd actually notice. 

Thanks. So you are saying it would make more sense to lower the 7600 Kit to run at 6000 CL30? 

 

What is the top achievable 1:1 ratio for most 9000 series CPUs? Would it be 6000, 6200 or 6400? 

[RONOTHAN##]

2 hours ago, Dreg said:

So you are saying it would make more sense to lower the 7600 Kit to run at 6000 CL30? 

Yup, lowering it down to a speed where you can use 1:1 mode is going to result in better performance. 

 

2 hours ago, Dreg said:

What is the top achievable 1:1 ratio for most 9000 series CPUs? Would it be 6000, 6200 or 6400? 

It depends. AMD says that every chip should do 6400 1:1, but at least at the launch there were CPUs that struggled to do even 5600MT/s. There are newer BIOSes that have supposedly fixed most of those issues, though I don't know how much I'd personally trust that "every chip should do 6400" statement from the beginning even if most should. 

[Dreg]

41 minutes ago, RONOTHAN## said:

Yup, lowering it down to a speed where you can use 1:1 mode is going to result in better performance. 

 

It depends. AMD says that every chip should do 6400 1:1, but at least at the launch there were CPUs that struggled to do even 5600MT/s. There are newer BIOSes that have supposedly fixed most of those issues, though I don't know how much I'd personally trust that "every chip should do 6400" statement from the beginning even if most should. 

Thanks for the info, I mean worse case 6200 should be guaranteed, and the difference between 6200 and 6400 is pretty unsignificant. 

Last thing to figure out is which timings I would need to choose for a 6200 or 6400 Mhz to work with my DDR5 7600 (which is rated Timing 36-46-46-84, CAS Latency 36). 

 

Am not sure how to know if for example my kit would be able to do 30 or 32 timing at 6200 for example? 

 

 

[--SID--]

Worst case it doesn't work at all or with a lot of tweaking since this kit is optimized for Intel with Intel XMP profiles only.

[RONOTHAN##]

30 minutes ago, Dreg said:

Am not sure how to know if for example my kit would be able to do 30 or 32 timing at 6200 for example? 

So the timings are just a given amount of time measured in clock cycles, so you can convert the rated timings to something of a lower speed. The easy way is to just multiply the timing by the new frequency then divide by the old frequency, so the 7600 CL36-46-46-84 would become DDR5 6400 CL30-39-39-71 when rounding to the nearest value. You can also manually tune the timings, since odds are the kit has some headroom in it (I'd be surprised if you couldn't do CL30-37-37-30 at 1.43V, for instance), but those timings should work out if I quick did the math correctly. 

[Dreg]

46 minutes ago, --SID-- said:

Worst case it doesn't work at all or with a lot of tweaking since this kit is optimized for Intel with Intel XMP profiles only.

Is there a kit from TeamGroup that works very well with AM5, which is around 6200-6400? 

 

I like TeamGroup because the temps on the modules are extremely better the Gskill 6000 kit I had before it. 

[Dreg]

22 minutes ago, RONOTHAN## said:

So the timings are just a given amount of time measured in clock cycles, so you can convert the rated timings to something of a lower speed. The easy way is to just multiply the timing by the new frequency then divide by the old frequency, so the 7600 CL36-46-46-84 would become DDR5 6400 CL30-39-39-71 when rounding to the nearest value. You can also manually tune the timings, since odds are the kit has some headroom in it (I'd be surprised if you couldn't do CL30-37-37-30 at 1.43V, for instance), but those timings should work out if I quick did the math correctly. 

Am guessing with 6000 or 6200, the timings of DDR5 6400 CL30-39-39-71 should be even easier to achieve? 

[RONOTHAN##]

1 minute ago, Dreg said:

Am guessing with 6000 or 6200, the timings of DDR5 6400 CL30-39-39-71 should be even easier to achieve? 

yeah, you could just redo the math with DDR5 6000 or 6200 if you need to, it's just (timings)*(new frequency)/(old frequency), but it should come out to something around there since the difference between 6400 and 6000 is pretty small. 

[--SID--]

1 hour ago, Dreg said:

Is there a kit from TeamGroup that works very well with AM5, which is around 6200-6400?

There's a 6000C30 kit with EXPO profiles: Team Group Delta FF7D532G6000HC30DC01

[Dreg]

1 hour ago, --SID-- said:

There's a 6000C30 kit with EXPO profiles: Team Group Delta FF7D532G6000HC30DC01

Ya I saw that one, and am guessing the performance difference between 6000, 6200 and 6400 for 1440p gaming is pretty much very little.... 

 

Let's see what the 9800x3d will do in terms of DDR5 memory

[Dreg]

On 11/3/2024 at 3:38 PM, RONOTHAN## said:

So the timings are just a given amount of time measured in clock cycles, so you can convert the rated timings to something of a lower speed. The easy way is to just multiply the timing by the new frequency then divide by the old frequency, so the 7600 CL36-46-46-84 would become DDR5 6400 CL30-39-39-71 when rounding to the nearest value. You can also manually tune the timings, since odds are the kit has some headroom in it (I'd be surprised if you couldn't do CL30-37-37-30 at 1.43V, for instance), but those timings should work out if I quick did the math correctly. 

Did some more research, is this pretty much the lowest latency kit for Ryzen 7000/9000 series? https://www.newegg.ca/g-skill-32gb-ddr5-6000/p/N82E16820374635

 

The GSkill can basically run 6200 @28-36-36-96. 

 

Would there be big differences between the  DDR5 6400 @ CL30-39-39-71 and let's say DDR5 6200 @28-36-36-96? Not sure if the 7600 kit would handle 6200 with 28 timing. 

[RONOTHAN##]

1 hour ago, Dreg said:

Did some more research, is this pretty much the lowest latency kit for Ryzen 7000/9000 series? https://www.newegg.ca/g-skill-32gb-ddr5-6000/p/N82E16820374635

Just about, though if you can get 6400 CL32 to run that is just faster, at least according to TechPowerUp's testing. 

 

1 hour ago, Dreg said:

Would there be big differences between the  DDR5 6400 @ CL30-39-39-71 and let's say DDR5 6200 @28-36-36-96?

Not really, and remember that the entire point of the X3D chips is to mask memory latency, so the differences will be even more minimal. 

 

1 hour ago, Dreg said:

Not sure if the 7600 kit would handle 6200 with 28 timing. 

I'd be surprised if it couldn't, you'd just need to increase the memory voltage. I've gotten very basic kits to run 6200 CL26, it just requires 1.5V+ to work. 

[Dreg]

24 minutes ago, RONOTHAN## said:

Just about, though if you can get 6400 CL32 to run that is just faster, at least according to TechPowerUp's testing. 

 

Not really, and remember that the entire point of the X3D chips is to mask memory latency, so the differences will be even more minimal. 

 

I'd be surprised if it couldn't, you'd just need to increase the memory voltage. I've gotten very basic kits to run 6200 CL26, it just requires 1.5V+ to work. 

I saw some stats (am not sure if am allowed to post links to other vids here), where the difference between 6200 CL34 and CL28 is a massive 40 FPS for the 0.1% lows (in 1 game) and like 10 FPS for other games, on the same CPU (7800x3d). That sounds like a lot if valid. 

 

But I think 6200 at CL28 is probably  easier to achieve than say 6400 at CL30, unless someone has a good FCLK sample of a CPU. Who knows, the CL28 might also be slightly faster?  

[RONOTHAN##]

26 minutes ago, Dreg said:

am not sure if am allowed to post links to other vids here

It shouldn't be against the rules.

 

27 minutes ago, Dreg said:

where the difference between 6200 CL34 and CL28 is a massive 40 FPS for the 0.1% lows (in 1 game) and like 10 FPS for other games, on the same CPU (7800x3d). That sounds like a lot if valid.

I'd be curous about the other timings used, since if it's something like 6200 CL34-44-44-108 vs. 6200 CL28-36-36-76, that's a much bigger difference than something like 6400 CL30-37-37-71 vs. 6200 CL28-36-36-76, though even then that's a bigger difference than what I'd expect, especially with a 7800X3D. I'd almost wonder if it's in 2:1 vs. 1:1 for the tests. 

[Dreg]

25 minutes ago, RONOTHAN## said:

It shouldn't be against the rules.

 

I'd be curous about the other timings used, since if it's something like 6200 CL34-44-44-108 vs. 6200 CL28-36-36-76, that's a much bigger difference than something like 6400 CL30-37-37-71 vs. 6200 CL28-36-36-76, though even then that's a bigger difference than what I'd expect, especially with a 7800X3D. I'd almost wonder if it's in 2:1 vs. 1:1 for the tests. 

Here is the vid, let me know what do you think:  

 

[RONOTHAN##]

On 11/5/2024 at 1:46 PM, Dreg said:

Here is the vid, let me know what do you think:  

 

Sorry I didn't respond yesterday, I've been busy and didn't get the chance to look at this video till now. First thing I saw was that it was CL34-44-44 vs CL28-36-36, which will be a pretty big difference on its own, so I wouldn't put too much standing in it vs. comparing it to a 6400 CL30-37-37 setup like you'd probably be running. The numbers you'll get will be much closer to the CL28 values than the CL34 values. 

[Dreg]

11 hours ago, RONOTHAN## said:

Sorry I didn't respond yesterday, I've been busy and didn't get the chance to look at this video till now. First thing I saw was that it was CL34-44-44 vs CL28-36-36, which will be a pretty big difference on its own, so I wouldn't put too much standing in it vs. comparing it to a 6400 CL30-37-37 setup like you'd probably be running. The numbers you'll get will be much closer to the CL28 values than the CL34 values. 

 

With the 9800x3d now out, am pretty much set on it seeing the results, but not many tested DDR5 mem on it yet. 

 

So to conclude, it sounds like 6000 or 6200 at CL30 should be really good, and if I can push them to C28 then that would be even better. 

 

Now comes choosing the motherboard...

[Eigenvektor]

54 minutes ago, Dreg said:

With the 9800x3d now out, am pretty much set on it seeing the results, but not many tested DDR5 mem on it yet.

HuB did a test of memory scaling: https://www.youtube.com/watch?v=BcYixjMMHFk&t=1472s

 

tl;dw: the basic recommendation is to stick to DDR5 6000 CL30

[RONOTHAN##]

2 hours ago, Dreg said:

Now comes choosing the motherboard...

This part is even easier. Figure out what features you need, then just get the cheapest board that has them. For AM5 in particular, the motherboard really doesn't affect much outside of how many USB ports you have and what the PCIe layout is, so getting somethign dirt cheap is more than adequate. 

[Dreg]

4 minutes ago, RONOTHAN## said:

This part is even easier. Figure out what features you need, then just get the cheapest board that has them. For AM5 in particular, the motherboard really doesn't affect much outside of how many USB ports you have and what the PCIe layout is, so getting somethign dirt cheap is more than adequate. 

First thing to decide is x670E or x870E ($$ aside), but am eyeing the Strix or Hero 870/670 rn 

[RONOTHAN##]

Just now, Dreg said:

First thing to decide is x670E or x870E ($$ aside), but am eyeing the Strix or Hero 870/670 rn 

Unless you have a specific reason why, I'd honestly stick more to B650. The performance is the exact same, and unless you need USB4 or a ton of M.2 devices, it'll be more than adequate. 

[GuiltySpark_]

1 minute ago, Dreg said:

First thing to decide is x670E or x870E

Particular reason why you’re looking at those  and not B650?

[PDifolco]

On 11/3/2024 at 4:22 AM, Dreg said:

Hi all,

 

I currently have a 13900KS with the TeamGroup DDR5 7600 CL36, https://www.newegg.ca/team-group-32gb-ddr5-7600/p/N82E16820331930

 

Am planning to get a 9800x3d and was wondering how memory works with Ryzen 9000 series.

 

As far as the motherboard, am planning to go with the Asus Strix X890E.

 

Would I have to run the kit at a lower speed or 7600 would be fine? 

Thanks

I'd do this in steps

First try to increase IF clock to 2133MHz

If it works, run RAM at 6400CL36, shoud be ok, then try to decrease CL to 34 then 32

That's the best you can get in 1:1 mode

If either don't work, let IF at 2000MHz and set RAM at 6000CL36, should work fine, and decrease down to 34 then 32 then 30