z490 and populating ALL 4 DIMM Slots, which mem layout is preferred?

newls1

Supreme [H]ardness
Joined
Sep 8, 2003
Messages
4,607
Ill be populating all 4 dimm slots on (whatever) Z490 board I choose... which layout will provide me with the best chance to hit 4133+ Speeds, T-Topology, or Daisy Chain? Appreciate any feedback.
 
You been here since 2003 and don't even give 24hrs before replying to yourself? Generally you can't have crazy high memory speed and populate all the DIMMs.
 
Daisy chain for 2 DIMMs and T-topology for 4 DIMMs will typically yield best results. If you're going after 32GB I wouldn't do 4 x 8GB single rank, 2 x 16GB dual rank will give you potentially higher clocks and slightly better performance when compared to single rank at the exact same frequency/timings. I don't know anything about those newfangled 32GB DIMMs that are starting to pop up.
 
Daisy chain for 2 DIMMs and T-topology for 4 DIMMs will typically yield best results. If you're going after 32GB I wouldn't do 4 x 8GB single rank, 2 x 16GB dual rank will give you potentially higher clocks and slightly better performance when compared to single rank at the exact same frequency/timings. I don't know anything about those newfangled 32GB DIMMs that are starting to pop up.
Ok, so even though I am not a noob, I guess in my ignorance I thought you just got a memory kit and populated the memory stick slots? so what T-Topology? Daisy Chain? If this was USB, I think I might understand. But memory sticks?
 
Ok, so even though I am not a noob, I guess in my ignorance I thought you just got a memory kit and populated the memory stick slots? so what T-Topology? Daisy Chain? If this was USB, I think I might understand. But memory sticks?
Grab some popcorn and a good beverage:

 
Ill be populating all 4 dimm slots on (whatever) Z490 board I choose... which layout will provide me with the best chance to hit 4133+ Speeds, T-Topology, or Daisy Chain? Appreciate any feedback.

Since Z490 isn't out, your jumping the gun. The truth is, we don't know how it will be when it comes to memory performance.

Generally you can't have crazy high memory speed and populate all the DIMMs.

Generally, you can with Intel CPU's and motherboard chipsets. I've done it many times on Z390.
 
Fair, it's been a while since I done Z. The opposite has been my experience with running 8 16GB DIMM's in hedt. It seems like with all these ITX freaks running around these days with 2 DIMM slots seem to have the best memory overclocking.
 
Fair, it's been a while since I done Z. The opposite has been my experience with running 8 16GB DIMM's in hedt. It seems like with all these ITX freaks running around these days with 2 DIMM slots seem to have the best memory overclocking.

Two DIMM slots is the way to go for ultra-high speed RAM. That said, most people will probably never push RAM clocks further than you can generally achieve on Z390. I've gone as high as 4,133MHz on Z390 using 4x4GB modules. I've also gone to 3866MHz using 4x8GB modules. That was B-Die, so I might have been able to go further with it if I had more time on it. Now, hitting speeds well beyond those I've seen personally may be another story. But most people don't have RAM that can clock like that in the first place. Garden variety modules can easily achieve DDR4 3800MHz or thereabouts on Z390 with standard XMP profiles and zero tuning. I've got some Hynix stuff that's rated for DDR4 3600MHz, and it will do that on Z390 or even X570 if you've got a good board.
 
Two DIMM slots is the way to go for ultra-high speed RAM. That said, most people will probably never push RAM clocks further than you can generally achieve on Z390. I've gone as high as 4,133MHz on Z390 using 4x4GB modules. I've also gone to 3866MHz using 4x8GB modules. That was B-Die, so I might have been able to go further with it if I had more time on it. Now, hitting speeds well beyond those I've seen personally may be another story. But most people don't have RAM that can clock like that in the first place. Garden variety modules can easily achieve DDR4 3800MHz or thereabouts on Z390 with standard XMP profiles and zero tuning. I've got some Hynix stuff that's rated for DDR4 3600MHz, and it will do that on Z390 or even X570 if you've got a good board.
I'm new to this topic, so I have what may be a noob question. My new system has 2 x 16 GB RAM sticks. I have 3600 speed RAM, and since this is a new rig, I want to run at stock settings for a few weeks before I try any overclocking. But when I do, let's say with some work i can get the RAM to run at 4000. How much will that speed up Windows or major software like MS Word? Or a memory hog like Photoshop?

If I add in two more 16 GB sticks, for a total of 64 GB, won't Windows run faster, even if I don't overclock the RAM?

Please don't misunderstand. I'm all for overclocking and I'm all for fast RAM. That's why I paid extra for 3600 RAM instead of 3200 or something slower.
 
Daisy chain for 2 DIMMs and T-topology for 4 DIMMs will typically yield best results. If you're going after 32GB I wouldn't do 4 x 8GB single rank, 2 x 16GB dual rank will give you potentially higher clocks and slightly better performance when compared to single rank at the exact same frequency/timings. I don't know anything about those newfangled 32GB DIMMs that are starting to pop up.

T-topology for 4 dims is excellent on z390 (gigabyte) -- buildzoid covered it even (it will out perform 2 dims though at greater power draw).
 
I'm new to this topic, so I have what may be a noob question. My new system has 2 x 16 GB RAM sticks. I have 3600 speed RAM, and since this is a new rig, I want to run at stock settings for a few weeks before I try any overclocking. But when I do, let's say with some work i can get the RAM to run at 4000. How much will that speed up Windows or major software like MS Word? Or a memory hog like Photoshop?

If I add in two more 16 GB sticks, for a total of 64 GB, won't Windows run faster, even if I don't overclock the RAM?

Please don't misunderstand. I'm all for overclocking and I'm all for fast RAM. That's why I paid extra for 3600 RAM instead of 3200 or something slower.

Faster memory means more memory bandwidth. Unfortunately, most desktop applications cannot take advantage of it. This is why moving to quad-channel HEDT systems rarely yields major increases in application performance. Most desktops simply aren't capable of using that many threads or that much memory bandwidth. That's not to say its totally useless. Games can take advantage of faster memory to some point. You do reach a point of diminishing returns pretty quickly. Going to DDR4 3733MHz for example is good for X570 and Ryzen 3000 series processors. DDR4 3800MHz is counter productive as it forces a divider to be used where the Infinity Fabric and memory clocks are no longer synchronous. In other words, DDR4 3733MHz is faster than DDR4 3800MHz which is ordinarily counter intuitive. Ryzen also benefits from lower memory latencies. Often, your better off with slower clocked RAM if the latencies are better. This too is a balancing act as going super low on the clocks would negate any advantage gained by the lower latencies. Of course, that depends on individual applications as well.

Intel systems are different. Intel systems are capable of clocking RAM much higher than AMD systems are as a general rule. They can also do so with greater ease. Unfortunately, they benefit far less from doing so. It's the same with latencies. Intel systems don't really benefit from it all that much. If your at DDR4 3200MHz or better, your golden. Sure, some increases can be seen in synthetic tests and benchmarks. However, this rarely translates to the real world. I've run DDR4 3600MHz, 3800MHz and 4,000MHz modules on the same system and found virtually no difference in performance across the spectrum. Mostly you see the difference in synthetic tests like Sandra or AIDA64.

Lastly, people usually make the mistake of thinking that increasing system RAM increases the speed of the system. This simply isn't true. If you have 32GB of RAM and install 64GB of RAM, absolutely no speed increase will be seen in the OS'es performance or application performance unless those applications can actually make use of the extra physical RAM. There are a few desktop applications that can do this, but outside of content creation with things like After Effects or Premiere Pro, you aren't going to see that. If your system is primarily for gaming, increasing your memory to 64GB is basically useless. Games simply don't use that much memory. Going from 2x DIMMs to 4x DIMMs also increases the likelyhood that you will have to reduce your memory clocks or loosen your RAM timings. This is especially true on AMD systems. Depending on the memory topology of the motherboard, this will become more or less likely on either AMD or Intel systems.

The only reason I have 4x16GB DIMMs is because I was running an HEDT setup before and I got the RAM at a good price. I'm running four modules now for a total of 64GB because it's more aesthetically pleasing than two modules are. If I wasn't able to reach the same clocks the RAM was rated for (DDR4 3600MHz), I'd deal with it and pull the extra modules out. Four modules at DDR4 3200MHz or less would be slower than 2x DDR4 modules at 3600MHz.

Getting back to the topic of RAM increasing system speed, this is something that comes from the late 80's and early 90's where RAM was extremely expensive. Software better kept apace of hardware advances and systems built by OEMs were often light on RAM. Hard drives were much slower and when applications ran out of physical RAM and hit the pagefile, performance suffered. When you would install enough RAM to minimize pagefile usage and run the application entirely out of memory, the system would perform faster. But RAM by itself has nothing to do with the actual speed of the system by itself. It's all about making sure your programs have sufficient resources. Today, RAM is cheap and often times systems are built with more memory than they need. Gaming systems today only need about 16GB of RAM. 32GB gives you plenty of room to grow, but going beyond that is simply a waste of money. RAM is cheap, so if you want to, go right ahead. Just keep in mind there are caveats to running 4x DIMMs on some motherboards. IMC's in CPU's vary too, so sometimes being able to run such a setup is simple luck of the draw.

Going back to your Photoshop example, I took several images for a motherboard review via my phone. I'm using a Pixel 2, which has a 12 Megapixel camera if I am not mistaken. These files are all 4032x3024 pixels and I'm watching stuff on Youtube, running Photoshop with 42 images loaded up and so on. Photoshop is using 3GB of RAM according to task manager. I'm at 22% memory usage with 64GB of RAM.

1588104612788.png


The extra 32GB of RAM is absolutely useless in this case. Sometimes I have games running in the background that I minimize to work and vice versa. Even then, I wouldn't be near 50% memory usage. I can make After Effects or Premiere Pro use that kind of RAM. It does in my benchmark testing. 32GB systems smoke 16GB systems all other things being equal. But for Photoshop, you don't need that much unless your doing far crazier things in it than I am. I know Photoshop can use more RAM, but all you need to do is open task manager and see how much RAM you have left with it doing whatever it is you normally do. I've never seen it use more than about 4GB of RAM.

Hope this helps.
 
Faster memory means more memory bandwidth. Unfortunately, most desktop applications cannot take advantage of it. This is why moving to quad-channel HEDT systems rarely yields major increases in application performance. Most desktops simply aren't capable of using that many threads or that much memory bandwidth. That's not to say its totally useless. Games can take advantage of faster memory to some point. You do reach a point of diminishing returns pretty quickly. Going to DDR4 3733MHz for example is good for X570 and Ryzen 3000 series processors. DDR4 3800MHz is counter productive as it forces a divider to be used where the Infinity Fabric and memory clocks are no longer synchronous. In other words, DDR4 3733MHz is faster than DDR4 3800MHz which is ordinarily counter intuitive. Ryzen also benefits from lower memory latencies. Often, your better off with slower clocked RAM if the latencies are better. This too is a balancing act as going super low on the clocks would negate any advantage gained by the lower latencies. Of course, that depends on individual applications as well.

Intel systems are different. Intel systems are capable of clocking RAM much higher than AMD systems are as a general rule. They can also do so with greater ease. Unfortunately, they benefit far less from doing so. It's the same with latencies. Intel systems don't really benefit from it all that much. If your at DDR4 3200MHz or better, your golden. Sure, some increases can be seen in synthetic tests and benchmarks. However, this rarely translates to the real world. I've run DDR4 3600MHz, 3800MHz and 4,000MHz modules on the same system and found virtually no difference in performance across the spectrum. Mostly you see the difference in synthetic tests like Sandra or AIDA64.

Lastly, people usually make the mistake of thinking that increasing system RAM increases the speed of the system. This simply isn't true. If you have 32GB of RAM and install 64GB of RAM, absolutely no speed increase will be seen in the OS'es performance or application performance unless those applications can actually make use of the extra physical RAM. There are a few desktop applications that can do this, but outside of content creation with things like After Effects or Premiere Pro, you aren't going to see that. If your system is primarily for gaming, increasing your memory to 64GB is basically useless. Games simply don't use that much memory. Going from 2x DIMMs to 4x DIMMs also increases the likelyhood that you will have to reduce your memory clocks or loosen your RAM timings. This is especially true on AMD systems. Depending on the memory topology of the motherboard, this will become more or less likely on either AMD or Intel systems.

The only reason I have 4x16GB DIMMs is because I was running an HEDT setup before and I got the RAM at a good price. I'm running four modules now for a total of 64GB because it's more aesthetically pleasing than two modules are. If I wasn't able to reach the same clocks the RAM was rated for (DDR4 3600MHz), I'd deal with it and pull the extra modules out. Four modules at DDR4 3200MHz or less would be slower than 2x DDR4 modules at 3600MHz.

Getting back to the topic of RAM increasing system speed, this is something that comes from the late 80's and early 90's where RAM was extremely expensive. Software better kept apace of hardware advances and systems built by OEMs were often light on RAM. Hard drives were much slower and when applications ran out of physical RAM and hit the pagefile, performance suffered. When you would install enough RAM to minimize pagefile usage and run the application entirely out of memory, the system would perform faster. But RAM by itself has nothing to do with the actual speed of the system by itself. It's all about making sure your programs have sufficient resources. Today, RAM is cheap and often times systems are built with more memory than they need. Gaming systems today only need about 16GB of RAM. 32GB gives you plenty of room to grow, but going beyond that is simply a waste of money. RAM is cheap, so if you want to, go right ahead. Just keep in mind there are caveats to running 4x DIMMs on some motherboards. IMC's in CPU's vary too, so sometimes being able to run such a setup is simple luck of the draw.

Going back to your Photoshop example, I took several images for a motherboard review via my phone. I'm using a Pixel 2, which has a 12 Megapixel camera if I am not mistaken. These files are all 4032x3024 pixels and I'm watching stuff on Youtube, running Photoshop with 42 images loaded up and so on. Photoshop is using 3GB of RAM according to task manager. I'm at 22% memory usage with 64GB of RAM.

View attachment 241120

The extra 32GB of RAM is absolutely useless in this case. Sometimes I have games running in the background that I minimize to work and vice versa. Even then, I wouldn't be near 50% memory usage. I can make After Effects or Premiere Pro use that kind of RAM. It does in my benchmark testing. 32GB systems smoke 16GB systems all other things being equal. But for Photoshop, you don't need that much unless your doing far crazier things in it than I am. I know Photoshop can use more RAM, but all you need to do is open task manager and see how much RAM you have left with it doing whatever it is you normally do. I've never seen it use more than about 4GB of RAM.

Hope this helps.
Yes, that was a very helpful post. Glad I got 2 x 16 GB instead of 4 x 8 GB and I'm really glad I didn't shell out for a quad-channel motherboard. :) The part about 3733 memory being faster than 3800 is really unintuitive, especially since I have read in some other threads that you should try for 3800 but not higher.
 
Yes, that was a very helpful post. Glad I got 2 x 16 GB instead of 4 x 8 GB and I'm really glad I didn't shell out for a quad-channel motherboard. :) The part about 3733 memory being faster than 3800 is really unintuitive, especially since I have read in some other threads that you should try for 3800 but not higher.

I could be off on the number. AMD's documentation actually says DDR4 3733MHz not 3800MHz. But I think you can make 3800MHz work. But going to 4000MHz would certainly be slower because of the divider. There is no doubt about it, especially since clock speed increases usually mean more performance.

Also, there is no quad-channel option unless you step up to X299, X399 or TR40x.
 
I'm running 4x16gig memory sticks in my 9900k setup, it's not the fastest memory, but I love having oodles of it an not having to think about it running out - ever.
 
I could be off on the number. AMD's documentation actually says DDR4 3733MHz not 3800MHz. But I think you can make 3800MHz work. But going to 4000MHz would certainly be slower because of the divider. There is no doubt about it, especially since clock speed increases usually mean more performance.

Also, there is no quad-channel option unless you step up to X299, X399 or TR40x.

3800 will work on AMD systems but not higher. It seems that Infinity Fabric really can't be pushed past 1900Mhz. Some systems will hit it and be stable where others, not so much. I have had mine running at 1900/3800 for months and it is fine, with 3600 Ram modules, 4 of them. So it can be done but if you want higher speed you need to deal with both lower CAS and lower IF.
 
3800 will work on AMD systems but not higher. It seems that Infinity Fabric really can't be pushed past 1900Mhz. Some systems will hit it and be stable where others, not so much. I have had mine running at 1900/3800 for months and it is fine, with 3600 Ram modules, 4 of them. So it can be done but if you want higher speed you need to deal with both lower CAS and lower IF.

According to the documentation AMD provided us prior to the launch of the CPU, it was 3733MHz.

1588182404117.png


I also recall some motherboards setting the divider to 2:1 when you select 3800MHz speeds. I have to check on that, I haven't messed with that in awhile.
 
Seems like 3800 is right on the line between 1:1 and 2:1, and AMD didn't mention it since it's not a multiple of 133.33Mhz like all DDR memory is. So motherboard makers are left to use their own judgement on how to handle speeds between 3733 and 3866 if they allow them.
 
  • Like
Reactions: x509
like this
i have all 8dimms on my x299 populated and running at 4133cl19. all 8gb bdie sticks. board didnt complain at all and just does its job without a fuss.

some games and benchmarks will give better results with all dimms populated and others wont really care if you only fill 2 of 4dimms.
if i were you i would buy two of the patriot viper 4400c19 kits and use those in all 4 dimms
they are 129$ each in the US from newegg. its BY FAR the best deal in fast low latency bdie.

im buying a 2dimm board becuase those are even better for memory overclocking usually. fingers crossed the ROG MAX 12 APEX is available in the US tomorrow..
and hopefully EVGA launches another DARK board. it will probably be two dimms total also for better overclocking.


in regards to zen2 and the line between 1:1 and 2:1
seems since most CPUs have trouble running the IF at or past 1900 so the line is a little under that for most users. some of the really good samples can go further to 1933 but that really is the line.
different story for threadripper tho. seems that frequency is more important that keeping the 1:1 ratio in some cases. but im not well versed on that. just recall Buildzoid (from the video above) talking about that when he was testing his 3960x a while ago
 
According to the documentation AMD provided us prior to the launch of the CPU, it was 3733MHz.

View attachment 241322

I also recall some motherboards setting the divider to 2:1 when you select 3800MHz speeds. I have to check on that, I haven't messed with that in awhile.
Probably a bios option, for me I just set the Infinity fabric to 1900mhz run ram at 3800mhz for 1:1, ran that for over 6 months.

Now that being said, I backed down to 3733mhz, couple weeks ago I had to double boot periodically (new thing), went to 3733mhz and have not seen that. I may bump it up again and start tweaking but no hurry.

Pretty much 3600mhz- 3800mhz is golden and one would not notice much of any difference in real things.
 
Ill be populating all 4 dimm slots on (whatever) Z490 board I choose... which layout will provide me with the best chance to hit 4133+ Speeds, T-Topology, or Daisy Chain? Appreciate any feedback.


What board did you go with in the end? MSI Z490 Unify, I've built 2 (10900K) of these now which supported gskill 8 MB sticks of 4133 MHz for total 32 MB Ram -- So far I've not seen any T-Topology boards with Z490 Chipset (which is preferred with filling all 4 dimm slots).
 
Back
Top