Any of us running 48GB (2x24GB) high frequency on Z790 platforms?

[xDiVolatilX]

Hey fellas,

Any of us running 48GB (2x24GB) high frequency on Z790 platforms?

If so what speed? Brand of RAM? Brand of motherboard?

I want to see what the faster RAM kits are generally working with XMP out the box for most Z790 boards?

I know every CPU has IMC lottery and the board has variance also. I'm trying to see what is the fastest kits enthusiasts are having the best results with?

For example I'm looking at a 7600 48GB kit from teamgroup
TEAMGROUP T-Force Delta RGB DDR5 Ram 48GB (2x24GB) 7600MHz PC5-60800 CL36 M-DIE Desktop Memory Module Ram for 600 700 Series Chipset XMP 3.0 Ready Black - FF3D548G7600HC36EDC01 https://a.co/d/7JEZEB8

Or a 7200 48GB kit like this one
TEAMGROUP T-Force Delta RGB DDR5 Ram 48GB (2x24GB) 7200MHz PC5-57600 CL34 M-DIE Desktop Memory Module Ram for 600 700 Series Chipset XMP 3.0 Ready Black - FF3D548G7200HC34ADC01 https://a.co/d/jiwCR26

What are you guys running on Z790? Not interested in 32GB kits though only 48GB and looking to see which kit runs out the box on XMP no tweaking.

 

[Mad Maxx]

My rig has a 48GB kit (specs in sig). Everything runs at stock speeds. Tweaking/overclocking bores me. Haven't had the slightest problem with this build.

 

[xDiVolatilX]

My rig has a 48GB kit (specs in sig). Everything runs at stock speeds. Tweaking/overclocking bores me. Haven't had the slightest problem with this build.

Sweet. That board is a very nice-looking board. Is that a normal Z790 board? or a "refresh" Z790 board? I'm not familiar with every different brand "refresh" boards.

For example, on Gigabyte they have added an "X" to all existing Aorus Master and Tachyon boards.

So, when you say you are running at stock speeds do you mean the default 4800? or 5600? or the overclocked XMP 7200?

Have you stability tested it? Prime95 or Ycruncher? Has it ever crashed?

 

[Mad Maxx]

Sweet. That board is a very nice-looking board. Is that a normal Z790 board? or a "refresh" Z790 board? I'm not familiar with every different brand "refresh" boards.

For example, on Gigabyte they have added an "X" to all existing Aorus Master and Tachyon boards.

So, when you say you are running at stock speeds do you mean the default 4800? or 5600? or the overclocked XMP 7200?

Have you stability tested it? Prime95 or Ycruncher? Has it ever crashed?

MSI added the TI to their refreshed 790 boards. I set the profile to 7200 when I installed the RAM and haven’t looked at it in over 6 months. No crashes or other problems so I’m just leaving everything alone. I haven’t run any tests, either.

 

[xDiVolatilX]

So I ended up buying this kit
TEAMGROUP T-Force Delta RGB DDR5 Ram 48GB (2x24GB) 7600MHz PC5-60800 CL36 M-DIE Desktop Memory Module Ram for 600 700 Series Chipset XMP 3.0 Ready Black - FF3D548G7600HC36EDC01
https://a.co/d/7JEZEB8

XMP may or may not be stable (i will do a lot more testing) but I already accepted I might need to faffle with tuning and so I have started to learn tuning memory my damn self lol

It is stable at XMP so far, I will keep reporting back on any findings here.

Also for an undervolt I have got it stable at 7400MHz @ 1.325V (Default XMP is 1.4) at C36 latency which is a nice undervolt for the sacrifice of of 200MHz with a few passes on Ycruncher and 10 mins on prime 95. I haven't had time to do more in depth testing but I will post here results for those interested.

I am on a 13900KS (soon to be 14900KS) on a Aorus Master Z790 board. I would say for me being relatively new to the RAM OC tuning game I consider this a huge success for my first try.

Again this was only a few tests nothing overnight or anything although I wouldn't leave my CPU crunching PI or running Prime at 80C all night anyway lol. I'll do more testing when I have free time but off to a good start!

 

[Captain Newmackwa]

I got a 48GB (2x24GB) Corsair Vengeance DDR5 6400MT/s CL36 kit last month. It's stable at XMP II but unfortunately, it came with Micron/SpecTek chips. I guess that explains the rather loose 36-48-48-104 timings. I changed that to a 64GB (2x32GB) Corsair Vengeance DDR5 6000MT/s CL30 30-36-36-76 kit. That one has SK Hynix A-Die revision chips. In practice, I couldn't tell the difference though.

 

[xDiVolatilX]

I got a 48GB (2x24GB) Corsair Vengeance DDR5 6400MT/s CL36 kit last month. It's stable at XMP II but unfortunately, it came with Micron/SpecTek chips. I guess that explains the rather loose 36-48-48-104 timings. I changed that to a 64GB (2x32GB) Corsair Vengeance DDR5 6000MT/s CL30 30-36-36-76 kit. That one has SK Hynix A-Die revision chips. In practice, I couldn't tell the difference though.

What CPU and motherboard is this? You could have gotten faster speeds if you are on Intel. Any interest in going faster than 7000?

 

[xDiVolatilX]

I just revisited running XMP 7600 C36 (default xmp 1) and it passed a few tests of Ycruncher & prime 95 for 10 mins.
I'll keep running test and games to see if it is stable at xmp default if I need to see about 7400 but so far so good.
I think the newest bios updates are actually improving ddr5 stability on Z790 boards because honestly, I wasn't expecting to get as
good results as I am. It may be a combination of the bios updates and maybe my 13900KS is a good cpu and the Aorus Master
is a good board also even if it's 4 dimm so far it's holding xmp 7600 'c36
I will keep playing with this setup and report back.

 

[Captain Newmackwa]

What CPU and motherboard is this? You could have gotten faster speeds if you are on Intel. Any interest in going faster than 7000?

14900K and Asus Z790 ProArt. I guess I just wanted to save a few bucks so I settled for those speeds. I pretty much only game on this PC so ram speeds don't really matter much from what I've read. As long as the DDR5 is at least 10ns, I'm good.

 

[Sycraft]

In practice, I couldn't tell the difference though.

I feel like that's the reality with high speed RAM and tight timings these days. I see a lot of stress over it, but for gaming at least I've never seen evidence that it matter. Ever benchmark I've come across for games shows a tiny difference, if any, between 5600 and anything faster, and they are usually testing at 1080 to try and emphasize differences. If it can't do at least 5ish percent at 1080, enough that I'd be convinced there is an actual different and not just measurement error, then I see very little reason to fret over it.

Personally like you I have 64GB of 6000mhz RAM because it was:

1) On the QVL list for my board which means it was probably going to work.
2) Available when I bought the components (6400mhz is also on the QVL list for this board at 64GB but was out of stock even if I wanted it).
3) Not too expensive.

I don't sweat it, and I won't unless someone can show me clear evidence of a non-trivial performance improvement in games at 4k.

 

[wareyore]

I was curious and picked up 48G sticks last week. The 32G were 7800 but I had to run them at 7600. I just didn't want to spend the time tweaking them. The 48G were 7600 and testing them there was 2.5% difference in memory specific testing. Basically margin of error and the closer real world performance indiscernible.

To the OP's point, XMP didn't work out of the box for either of these GSKILL kits on my ROG Z790 Hero. But, starting there and setting the speed to 7600 was the simple, no fuss method that did work. I am pretty sure the 48G kit I purchased was on the QVL list but may be wrong.

 

[xDiVolatilX]

This test was at 1

I feel like that's the reality with high speed RAM and tight timings these days. I see a lot of stress over it, but for gaming at least I've never seen evidence that it matter. Ever benchmark I've come across for games shows a tiny difference, if any, between 5600 and anything faster, and they are usually testing at 1080 to try and emphasize differences. If it can't do at least 5ish percent at 1080, enough that I'd be convinced there is an actual different and not just measurement error, then I see very little reason to fret over it.

Personally like you I have 64GB of 6000mhz RAM because it was:

1) On the QVL list for my board which means it was probably going to work.
2) Available when I bought the components (6400mhz is also on the QVL list for this board at 64GB but was out of stock even if I wanted it).
3) Not too expensive.

I don't sweat it, and I won't unless someone can show me clear evidence of a non-trivial performance improvement in games at 4k.

I think if your GPU is maxed out at 4k then it doesn't matter what the resolution is as in if it's maxed out at 1440p or 1080p it would be the same result right? It's just that 4k is the hardest resolution to run natively without upscaling and frame generation so it taxes the GPU.

Sometimes the cpu doesn't have enough steam to even max the GPU though, and sometimes faster RAM allows the CPU to push the GPU harder. Like if a game is more CPU and RAM intensive.

For example on Spiderman Remastered it doesn't matter if I turn the graphical settings all the way down to low and ray tracing off and frame gen off and still get the exact same FPS performance lol. I would assume I am CPU bound at that point. I want to be at 144 but it's at 90-115. What about newer games that release that are more CPU and RAM needy.

Faster CPU and Faster RAM do matter. That's why I want to max out the board since I plan on keeping it for a while.

This guy tests Shadow of the tomb Raider at 1080p which an older game and the known to be a CPU intensive game and the results were a 25% increase from 4800 to 7200. Also If a game isn't GPU bound the wider the margin could be if the system can't keep up with the notoriously fast 4090.


View: https://youtu.be/SI1JQ_dhIWo?si=ZO7NrLvQwZ9NJJYj

 

[xDiVolatilX]

I was curious and picked up 48G sticks last week. The 32G were 7800 but I had to run them at 7600. I just didn't want to spend the time tweaking them. The 48G were 7600 and testing them there was 2.5% difference in memory specific testing. Basically margin of error and the closer real world performance indiscernible.

To the OP's point, XMP didn't work out of the box for either of these GSKILL kits on my ROG Z790 Hero. But, starting there and setting the speed to 7600 was the simple, no fuss method that did work. I am pretty sure the 48G kit I purchased was on the QVL list but may be wrong.

What CPU are you using? Also is your board updated to the latest bios? Oh and what is the max speed of your motherboard for the ram support?
For example on my z790 Aorus Master it claims to hold ram speed up to 8200 MHz and me having a 13900 KS CPU I figured I had a pretty good chance of hitting 7600 or higher.

The new 48 GB kits in my opinion are much better than the 64 GB kits which I have a few of. I also have a feeling that these new BIOS updates have really helped for high speed ddr5 stability because the posts I read from months ago people were not able to achieve these higher speeds on boards that are able to today with the newest BIOS updates and the newest 48 GB kits. I don't think there's any difference from the 32 GB kits to the 48 GB kits but they're definitely is one from the 64 GB kits, and of course the CPU memory controller matters also because of how much voltage it can send to the memory controller. And the motherboard is plain to see what the maximum rated claims should be telling of that also.

It's nice to see that you got 7600 XMP just like I did because from what I read and gather it seems to be now the new highest one click achievable speed whereas before it used to be 7,200 I feel like the newer ram kits and the BIOS updates have bumped it up from 7200 to 7600 XMP one click and that is rather luxurious LOL

 

[Sycraft]

I think if your GPU is maxed out at 4k then it doesn't matter what the resolution is as in if it's maxed out at 1440p or 1080p it would be the same result right? It's just that 4k is the hardest resolution to run natively without upscaling and frame generation so it taxes the GPU.

Right, but it is more likely to be GPU limited at 4k (or scanout limited at any resolution) and that's the resolution I game at, and it isn't likely to go down in the future. The point is that running 1920x1080 on a high spec GPU is kinda an edge case. If that is an edge case you do, for whatever reason, then ok, but to me, not relevant.

I think it is a fairly narrow amount of gamers that care about maximum FPS, and at lower resolution. If that's you, no problem, but it is a limited subset.

This guy tests Shadow of the tomb Raider at 1080p which an older game and the known to be a CPU intensive game and the results were a 25% increase from 4800 to 7200. Also If a game isn't GPU bound the wider the margin could be if the system can't keep up with the notoriously fast 4090.

Again, this is the problem, because this is the kind of thing I see when people are trying to convince me and others that fast RAM is a must: A single old game and a contrived test. It really isn't a great test because, resolution aside, there's two issues:

1) This is a single game, and older at that. If it is an anomaly then that isn't a big deal.

2) He is testing RAM that is too slow vs fast RAM. I'd like to see 5600mhz vs faster, as that is the speed Intel recommends. It has long been true that you can see performance issues with memory slower than they recommend. The question is how much, if any, advantage do you see for overclocked/XMP RAM.


A better test, in my opinion, was done by Hardware Unboxed and this shows me a few things:

1) The difference between RAM speeds is usually nothing, and minimal at best. Their 7-game average of 5600 vs 7200 was 213 fps vs 217. That's less than 2%, within the margin of error for testing.

2) The biggest difference still wasn't big, about 6.7% in Spiderman Remastered. That's not nothing, and is outside the margin for error, but is hardly something I'd get worked up about.

3) A bigger difference was made with RAM tuning by turning down the tRAS setting. That had 6000mhz RAM outperforming untuned 7200mhz RAM. However, as they note, not all systems can handle something like that 100% stable and requires lots of manual tuning of many settings.

4) Going below 5600mhz did incur a non-trivial performance penalty.


So while I'm not hating on people who want to play with it, again I don't think it is something worth getting worked up over. I'd rather have 64GB of RAM then have to drop to lower amounts of RAM to increase speeds or have to buy a more expensive board.

 

[xDiVolatilX]

Right, but it is more likely to be GPU limited at 4k (or scanout limited at any resolution) and that's the resolution I game at, and it isn't likely to go down in the future. The point is that running 1920x1080 on a high spec GPU is kinda an edge case. If that is an edge case you do, for whatever reason, then ok, but to me, not relevant.

I think it is a fairly narrow amount of gamers that care about maximum FPS, and at lower resolution. If that's you, no problem, but it is a limited subset.


Again, this is the problem, because this is the kind of thing I see when people are trying to convince me and others that fast RAM is a must: A single old game and a contrived test. It really isn't a great test because, resolution aside, there's two issues:

1) This is a single game, and older at that. If it is an anomaly then that isn't a big deal.

2) He is testing RAM that is too slow vs fast RAM. I'd like to see 5600mhz vs faster, as that is the speed Intel recommends. It has long been true that you can see performance issues with memory slower than they recommend. The question is how much, if any, advantage do you see for overclocked/XMP RAM.


A better test, in my opinion, was done by Hardware Unboxed and this shows me a few things:

1) The difference between RAM speeds is usually nothing, and minimal at best. Their 7-game average of 5600 vs 7200 was 213 fps vs 217. That's less than 2%, within the margin of error for testing.

2) The biggest difference still wasn't big, about 6.7% in Spiderman Remastered. That's not nothing, and is outside the margin for error, but is hardly something I'd get worked up about.

3) A bigger difference was made with RAM tuning by turning down the tRAS setting. That had 6000mhz RAM outperforming untuned 7200mhz RAM. However, as they note, not all systems can handle something like that 100% stable and requires lots of manual tuning of many settings.

4) Going below 5600mhz did incur a non-trivial performance penalty.


So while I'm not hating on people who want to play with it, again I don't think it is something worth getting worked up over. I'd rather have 64GB of RAM then have to drop to lower amounts of RAM to increase speeds or have to buy a more expensive board.

Well if we're talking price to performance, FPS gain % vs RAM price % it's not exactly accurate.

We would need to talk about the entire PCs cost % vs the upgraded PC % cost

According to hardware unboxed he's not taking into account the whole PCs build cost.

If you have a $600 CPU like a 13900k and a $1700 GPU like a 4090 and a $1000 dollar monitor and etc etc etc you likely have a $2500-$3000 PC setup approximately.

So assuming overall build cost of the PC setup you can spend an extra 10-20% of money to achieve 10-20% of performance. Turning your $3000 PC into a $3300 PC for faster CPU and/or RAM, not that big of a deal lol.

Of course the law of diminishing returns applies to not only PCs but to everything luxurious in life from belongings to cars to clothes etc. That last 25% more or less of the upper end of anything almost always never is price to performance but more for luxury or laughs or fun or hobby lol.

 

[Sycraft]

For me money is only part of it, it is more tradeoffs and stability. The cost wouldn't just be RAM, but a different board likely. My board only lists 6400mhz RAM for 2x32GB configurations. There are more expensive boards that list 7000mhz in that configuration, but not mine, and they are twice the price. Thus it isn't just the RAM cost, but board costs.

Or I could try it with my existing board, I mean there's nothing stopping me from setting it, it is just ASUS doesn't think it'll work. That sounds like a recipe for instability and headaches, if I can even get it to boot.

Or I could drop down to 2x16GB of RAM. fine for games, not so great for Nuendo.

...or I could do what I did.

If it was a 10-20% performance gain across the board, I might look at it. However it looks more like it is at most a 7% performance gain in a game I don't play, and 0% in most games. Thus, to me at least, not worth stressing.

 

[chameleoneel]

I was curious and picked up 48G sticks last week. The 32G were 7800 but I had to run them at 7600.~ The 48G were 7600 and testing them there was 2.5% difference in memory specific testing. Basically margin of error and the closer real world performance indiscernible.

Are you talking 48GB per stick or 2x24GB?

24GB sticks are still single rank. You don't get dual rank sticks until 32GB per stick. Or, you have to run 4 single rank sticks, to gain dual rank. But you usually can't run such high speeds with 4 sticks. So, a 2x24GB set won't perform any better than a 2x32GB.

32GB sticks and 48GB sticks are dual rank. And since they are both dual rank, a 2x32GB set and a 2x48GB set will also be the same performance as eachother. But, dual rank will give a pretty good boost over a single rank set (2x16GB and 2x24GB).

The big jumps are using high speeds and/or optimized sub timings. And then going to a 2 stick set which supports dual rank. Which means 2x32GB or 2x48GB.

 

[xDiVolatilX]

Are you talking 48GB per stick or 2x24GB?

24GB sticks are still single rank. You don't get dual rank sticks until 32GB per stick. Or, you have to run 4 single rank sticks, to gain dual rank. But you usually can't run such high speeds with 4 sticks. So, a 2x24GB set won't perform any better than a 2x32GB.

32GB sticks and 48GB sticks are dual rank. And since they are both dual rank, a 2x32GB set and a 2x48GB set will also be the same performance as eachother. But, dual rank will give a pretty good boost over a single rank set (2x16GB and 2x24GB).

The big jumps are using high speeds and/or optimized sub timings. And then going to a 2 stick set which supports dual rank. Which means 2x32GB or 2x48GB.

On a side note, Overclockers UK on YouTube was able to hit 6000 on 4 sticks. Although he was using a Z790 and a 14900K.

View: https://youtu.be/Znos34vPlFs?si=GfeMXPyKfwnkLWVi
The video is mainly about dealing with the 14900K , but interestingly enough, I noticed him running 6000 on 4 modules.

 

[xDiVolatilX]

Are you talking 48GB per stick or 2x24GB?

24GB sticks are still single rank. You don't get dual rank sticks until 32GB per stick. Or, you have to run 4 single rank sticks, to gain dual rank. But you usually can't run such high speeds with 4 sticks. So, a 2x24GB set won't perform any better than a 2x32GB.

32GB sticks and 48GB sticks are dual rank. And since they are both dual rank, a 2x32GB set and a 2x48GB set will also be the same performance as eachother. But, dual rank will give a pretty good boost over a single rank set (2x16GB and 2x24GB).

The big jumps are using high speeds and/or optimized sub timings. And then going to a 2 stick set which supports dual rank. Which means 2x32GB or 2x48GB.

The max I've seen Buildzoid hit on 32GB modules was 6800 pulling his hair out for hours per usual lol
I must admit his trials and tribulations entertain me because in the end he's done all the dirty work for us lol

 

[xDiVolatilX]

So it's been 30 mins prime stable running the 7600MHz T-Force Delta modules at undervolted by -50 at 1.350V stable max 38C.
I have the CPU to thermal throttle at 80c and it's in 12 core mode so nevermind the throttling of the CPU.
The RAM is loaded to 100% all available RAM used in the prime 95 test so full load on RAM and 12 CPU cores.
By the time I finished this post close to 45 mins stable.
So 7600 is absolutely a reality on 4 DIMM boards if your CPU/RAM/MB is high quality. This myth is busted. Update your BIOS for best results.