Why does Buildzoid say CAS Latency doesn't matter?

I haven't watched this video, but what I have heard elsewhere is that in DDR5 the headline CAS value doesn't make as much of a difference in performance anymore.

More benefit can be seen by optimizing other subtimings.

I don't know if there is any truth to it (I don't have any DDR5 systems, so I haven't tested it myself) but this is what I have heard and read.
 
It is a bit of a click-bait YT title but what he is essentially saying is Low CAS with crappy sub-timings will probably perform worse then slightly higher CAS sticks that are super dialed in. All things equal lower CAS is better but the black majik is in the subs.
 
He shows 6400 28-37-37-28-1T auto subs performing ~58-59ns in AIDA64 and ~ scoring about 10k lower in write\copy vs 6400 40-40-40-84-2T tuned subs which managed ~54-58ns
Also PYPrime scored 9.862 vs 8.894
 
Not going to lie. I haven't bothered with RAM overclocking in a really long time.

These days I usually just shop for sticks that support the highest rated non-overclock speed in the spec of the motherboard, look for the lowest CAS timings available at that speed and go with that.

Sounds like I may need to adjust my method to look more closely at the tRAS/tRCD, but these used to not even be published in many specs. I'm glad there is more data today.

My current RAM in the Threadripper is running at the default XMP timings of DDR4-3600, 16-19-19-39 and honestly, I don't even know if that is any good :p

The system seems to perform OK :p

I feel like RAM timings are such a rabbit hole to go down that you might get lost and never return :p
 
Zarathustra[H] said:
Sounds like I may need to adjust my method to look more closely at the tRAS/tRCD, but these used to not even be published in many specs. I'm glad there is more data today.
Click to expand...
No, that's not really what you want to look at. It's the manufacturer and die type of the chips in question. For DDR5, I think the king is Hynix A die for Intel, and Hynix M die for AMD (well either Hynix kind of works for AMD if you're lazy). If you want to find out what type is in what (sort of), motherboard RAM support lists might give you some idea because sometimes they list manufacturer and die type.

Either Hynix die will work for these low effort timings from Buildzoid:
https://www.patreon.com/posts/low-effort-rank-77403831

I've even used them on my dual rank (2x32 rather than 2x16) Hynix M die kit and havne't had any issues so far. Got the kit for $110 open box, so I'm pretty happy with that lol.

If you want to see the performance difference (in 1080p at least), Hardware Unboxed did a video:

View: https://youtu.be/MOatIQuQo3s?si=eMo8x775XhMQFxTr

It won't really answer your question because a lot of the benchmarked kits that have the Cas Latency as low values also have other timings at lower values, so it's hard to compare with just the claim in his video, but either way you can easily see the tuned kits beating the crap out of the not tuned kits. How well this scales into higher resolutions isn't really answered, but I assume he's using 1080p to test CPU limited scenarios since that's where the RAM matters more.
 
StoleMyOwnCar said:
How is it click bait? I've watched that video and he literally empirically shows exactly what he says in the title.
Click to expand...
I said A BIT of a CB title to draw the watcher in (nothing wrong with that) and he then describes the concept in much more detail, which he does well. If it was all as simple as his title there would have been no need for a video now would there?
 
funkydmunky said:
I said A BIT of a CB title to draw the watcher in (nothing wrong with that) and he then describes the concept in much more detail, which he does well. If it was all as simple as his title there would have been no need for a video now would there?
Click to expand...
"RANT: THE CAS LATENCY TIMING DOESN'T MATTER AS MUCH AS YOU THINK IT DOES"

This is the full title of the video. He made the video because he was actually getting annoyed with people acting like CAS Latency of their RAM kit was a big deal. He then goes on to show this empirically. Like, exactly this. Nothing else. Traditionally, clickbait involves misleading or empty titles that don't actually necessarily address their point or just don't contain any content of value. I'm not going to sit here and continue arguing semantics over this, but there is not a single thing about this video's title that I see as being clickbait except that it's in all caps (which does attract attention, because he's tired of getting the question or seeing the assertion). If that barebones minimum (title is in all caps) is your definition of clickbait, then fine I guess.
 
Last edited:
Dopamin3 said:
This is a click-bait thread. Posting a question and then linking a video that exactly answers that question. Is this The Twilight Zone?
Click to expand...
I am legitimately trying to understand RAM speeds and timings etc. RAM tuning is complicated imo so I was looking for more answers in a discussion.
 
Zarathustra[H] said:
This is all good and well, but I'd like to see real world performance, not some synthetic benchmark.
Click to expand...
If it is slower in synthetics for both latency and bandwidth then it will be slower in the real world but depending on the program it just may not be a noticeable or even measurable difference.
If latency or bandwidth was better for one setting vs the other then in the real world programs would sway one way or the other depending on there preference.
 
StoleMyOwnCar said:
It won't really answer your question because a lot of the benchmarked kits that have the Cas Latency as low values also have other timings at lower values, so it's hard to compare with just the claim in his video
Click to expand...
For all of the timings buildzoid provided----they show the stock subtimings of the kits which they applied buildzoid's timings to. Same thing in the Intel version of the video, where they use buildzoid timings on an Intel system. I use those same timings on my Intel system, daily driving.

View: https://youtu.be/RTmbYak_8gE?si=CQZo7PGSZiH7_o4e&t=289
 
StoleMyOwnCar said:
"RANT: THE CAS LATENCY TIMING DOESN'T MATTER AS MUCH AS YOU THINK IT DOES"

This is the full title of the video. He made the video because he was actually getting annoyed with people acting like CAS Latency of their RAM kit was a big deal. He then goes on to show this empirically. Like, exactly this. Nothing else. Traditionally, clickbait involves misleading or empty titles that don't actually necessarily address their point or just don't contain any content of value. I'm not going to sit here and continue arguing semantics over this, but there is not a single thing about this video's title that I see as being clickbait except that it's in all caps (which does attract attention, because he's tired of getting the question or seeing the assertion). If that barebones minimum (title is in all caps) is your definition of clickbait, then fine I guess.
Click to expand...
Okay we now know your trigger words!
3600CL14 is faster then 3600CL16. Ya? Buildzoid made an INTRIGING (better?) title to draw viewer in so he could explain further, ya?
 
funkydmunky said:
Okay we now know your trigger words!
3600CL14 is faster then 3600CL16. Ya? Buildzoid made an INTRIGING (better?) title to draw viewer in so he could explain further, ya?
Click to expand...
Depends on the sub timings, so no, not necessarily. Intriguing title doesn't mean clickbait.
 
funkydmunky said:
Okay we now know your trigger words!
3600CL14 is faster then 3600CL16. Ya? Buildzoid made an INTRIGING (better?) title to draw viewer in so he could explain further, ya?
Click to expand...
I think we've seen plenty of examples of me actually getting triggered and it's a far cry from this. I'm just confused since the video title states exactly what he's going to discuss and then proceeds to discuss it lol. I guess people don't like caps...
 
Zarathustra[H] said:
Not going to lie. I haven't bothered with RAM overclocking in a really long time.

These days I usually just shop for sticks that support the highest rated non-overclock speed in the spec of the motherboard, look for the lowest CAS timings available at that speed and go with that.

Sounds like I may need to adjust my method to look more closely at the tRAS/tRCD, but these used to not even be published in many specs. I'm glad there is more data today.

My current RAM in the Threadripper is running at the default XMP timings of DDR4-3600, 16-19-19-39 and honestly, I don't even know if that is any good :p

The system seems to perform OK :p

I feel like RAM timings are such a rabbit hole to go down that you might get lost and never return :p
Click to expand...
I run 4 sticks of 8Gb each of DDR4- 3600. Cl 14 made by Muskin Redline,
Threadripper approved
Click to expand...
 
xDiVolatilX said:
I am legitimately trying to understand RAM speeds and timings etc. RAM tuning is complicated imo so I was looking for more answers in a discussion.
Click to expand...

I briefly entered the rabbit hole but it is fucking complicated me. Or maybe not so complicated, but at the very least very complex.

You have to decode things from RAM serial numbers that the manufacturers don't want you to know, in order to figure out which ram dies are on your ram sticks, and only then you have to use that information to set all of these crazy timings.

I'm still not convinced that it is actually worth the time. I'd try it, but I don't feel like looking for the serial numbers on the stickers of my RAM...

If you want to go down this rabbit hole, for DDR4 this appears to be quite the reference:

https://www.reddit.com/r/overclocking/wiki/ram/ddr4/

As is the information captured here:

https://github.com/integralfx/MemTestHelper/tree/oc-guide

But it is kind of scatter-shot, and it doesn't seem to help that the people writing many of these guides seem to have a terrible time when it comes to organizing their thoughts into written content.
 
Zarathustra[H] said:
This is all good and well, but I'd like to see real world performance, not some synthetic benchmark.
Click to expand...
I've been doing these tests for years reviewing motherboards and occasionally memory modules. The thing is, CAS latencies have never meant all that much outside of synthetic testing and benchmarks. If you put two rigs side by side with the same clock speed memory, dividers, etc. you won't be able to tell a difference between them without some type of benchmark to quantify the difference. It's virtually undetectable to human perception.
 
Zarathustra[H] said:
I briefly entered the rabbit hole but it is fucking complicated me. Or maybe not so complicated, but at the very least very complex.

You have to decode things from RAM serial numbers that the manufacturers don't want you to know, in order to figure out which ram dies are on your ram sticks, and only then you have to use that information to set all of these crazy timings.

I'm still not convinced that it is actually worth the time. I'd try it, but I don't feel like looking for the serial numbers on the stickers of my RAM...

If you want to go down this rabbit hole, for DDR4 this appears to be quite the reference:

https://www.reddit.com/r/overclocking/wiki/ram/ddr4/

As is the information captured here:

https://github.com/integralfx/MemTestHelper/tree/oc-guide

But it is kind of scatter-shot, and it doesn't seem to help that the people writing many of these guides seem to have a terrible time when it comes to organizing their thoughts into written content.
Click to expand...


Just for shits and giggles, I did some AIDA64 tests on my Threadripper:

1.) Threadripper 3960x (24C/48T) SMP disabled. Otherwise All stock CPU settings in bios.
2.) 4 sticks of Gskill Ripjaws 5, Gskill F4-3600C16-16GVKC, 64GB total. 3600Mhz 16-19-19-39
All at XMP settings.

Results:
Read: 95,821
Write 97,230
Copy: 93,963
Latency: 81.4

All of these (except latency) seem pretty phenomenal for DDR4 until you realize that I am dealing with quad channel RAM on the Threadripper. Halve these numbers to be comparable to a dual channel system, and its nowhere near as impressive.

I'd try to fine tune them, but I suspect the fact that I am deal with quad channel means it's probably fast enough for my needs anyway, so I'd likely be wasting my time.
 
You must log in or register to reply here.
Back
Top