Regarding DDR5 Specs

[HuLkY]

Hi All,
I got a small curious question... since the release of DDR5 we are getting silly CL timings, like 40 40 40 190 and weird numbers compared to the more mature DDR4 since it has been there in the market for a longer while.
Looking at the market now, We get DDR5 Kits running at 6000MHz - this is what AMD recommended for their new platform AM5 as a sweet spot - and with CAS like 30 and I didn't even dig it's T1 or T2.

I am just wondering, why manufacturers are blasting the roof with 7500MHz and 8500Mhz with weird so slow CAS latencies! let me re-phrase my question, why can't we get a 6000MHz kit that runs cool 16-16-16-32 CAS for example? what's holding back the process? or this is a milking process like Intel and nVidia doing differently with their products? I am kinda new to that so apologies if my question looks naive.

I have seen OC3D doing an amazing experience for DDR4 vs current DDR5 with different speeds, I would say from my findings that moving to the new DDR5 platform, you don't need a whooping thing, a kit that runs 4800MHz will do you justice compared to the prices of higher speed ones, the differences are so minor and the gains in games doesn't justify the increase in price, what do you think all?

Again apoligies if my thoughts are naive or so but I know that RAM Gurus know a lot about these details, thank you all in advance.

 

[thewan]

Because RAM makers know that gamers is not where the money is coming from nowadays with everything going up, including the games themselves. They have shifted markets, and maybe you should shift your point of view too instead being stuck in the mindset that computer gaming is what drives its innovation. (hint: it hasn't for a quite a long time)

 

[Calenhad]

Latencies are just part of the story. For one, they are directly related to the MHz of the ram. Similarly to how DDR4 first released with higher CL than you can get now., you will see lower latency DDR5 modules in the future.

As a more extreme example than yours, we used to have CL2 memory back when we used DDR. Then that jumped to CL4 for DDR2 (we had different quality ram back then to. So there was a spread, but you get the idea). And I can guarantee you that they were not faster than the current DDR5 offerings.

 

[HuLkY]

Because RAM makers know that gamers is not where the money is coming from nowadays with everything going up, including the games themselves. They have shifted markets, and maybe you should shift your point of view too instead being stuck in the mindset that computer gaming is what drives its innovation. (hint: it hasn't for a quite a long time)

I totally see your point here!! and agree with it.... so they can release us now 6000MHz at CL14 if they wanted to not the weak half cooked CL30 they are putting now...

Is it true that not all motherboards running DDR5 can work with any CL below 20 also? like is someone released a ram at 6000MHz with CL19, not all motherboards can work on that XMP?

Latencies are just part of the story. For one, they are directly related to the MHz of the ram. Similarly to how DDR4 first released with higher CL than you can get now., you will see lower latency DDR5 modules in the future.

As a more extreme example than yours, we used to have CL2 memory back when we used DDR. Then that jumped to CL4 for DDR2 (we had different quality ram back then to. So there was a spread, but you get the idea). And I can guarantee you that they were not faster than the current DDR5 offerings.

they are milking us Badly!! I was aiming to update my desktop and I am not buying any fancy ram with cooling on it!!! I don't mind losing 4 frames in favor for saving more than 200USD!

 

[Calenhad]

I totally see your point here!! and agree with it.... so they can release us now 6000MHz at CL14 if they wanted to not the weak half cooked CL30 they are putting now...

Is it true that not all motherboards running DDR5 can work with any CL below 20 also? like is someone released a ram at 6000MHz with CL19, not all motherboards can work on that XMP?

Another important note here. You should not compared latencies between generations, since there are so many other differences in the technology. If you dig into the improvements from DDR4 to 5 you'd be amazed (at least I hope so). For one, DDR5 is always multi-rank. So there is no single rank vs dual rank performance difference anymore.

I am sure we will see that CL30 drop to something in the ~CL26 range within a few years, maybe even lower. But you will never see a 6000MHz DDR5 kit at CL14, that is just not physically possible. Look at DDR4 as a guide, you can have CL22 down to CL16 for the same speed rating depending on the ram modules used. So DDR5 might be something like CL26-CL34 for the same speeds. (hypothetical numbers here)

Now regarding the bling, I agree. I don't need it either. Save me some bucks by removing all that nonsense. I prefer a non-windowed case anyways.

 

[HuLkY]

Another important note here. You should not compared latencies between generations, since there are so many other differences in the technology. If you dig into the improvements from DDR4 to 5 you'd be amazed (at least I hope so). For one, DDR5 is always multi-rank. So there is no single rank vs dual rank performance difference anymore.

I am sure we will see that CL30 drop to something in the ~CL26 range within a few years, maybe even lower. But you will never see a 6000MHz DDR5 kit at CL14, that is just not physically possible. Look at DDR4 as a guide, you can have CL22 down to CL16 for the same speed rating depending on the ram modules used. So DDR5 might be something like CL26-CL34 for the same speeds. (hypothetical numbers here)

Now regarding the bling, I agree. I don't need it either. Save me some bucks by removing all that nonsense. I prefer a non-windowed case anyways.

you hyped me to dig in the improvements from DDR4 to DDR5, if you have any recommended materials displaying it please feel free to post it or PM me.
I am not comparing them but like, the idea itself, if we have the same speed let's say DDR5 at 4800MHz with a CL 15 is definitely better than the same speed at CL40 for example.. the tighter these times the better I believe, right?
I am just curious, why can't we get a 6000MHz kit running at CL14, apologies if missed the answer to that in your last message.

this is the kit I am aiming for if you have any better suggestion, please do!

Patriot Memory Signature Line Series DDR5 16GB (2 X 8GB) 4800MHz Udimm Kit - PSD516G4800K 16GB (2X8GB) 4800MHz Kit: Buy Online at Best Price in UAE - Amazon.ae

Buy Patriot Memory Signature Line Series DDR5 16GB (2 X 8GB) 4800MHz Udimm Kit - PSD516G4800K 16GB (2X8GB) 4800MHz Kit online on Amazon.ae at best prices. ✓ Fast and free shipping ✓ free returns ✓ cash on delivery available on eligible purchase.

www.amazon.ae

 

[Calenhad]

Yes, they are called latencies for a reason. Lower latencies are always better, given identical specs otherwise.

VERY oversimplified: Latencies are how many clock cycles you have to wait between reading from, or writing to the memory. Half the speed rating is how many clock cycles are available per second. Basically as speed goes up, you need higher latencies to handle the extra speed. A performance increase is gained by having a better speed:latency ratio within the same DDR generation. From generation to generation you see larger increases due to, for example, being able to do more operations in parallel. Again, this is the very very short version. I could probably write a thousand lines and still only cover the basics.

Anyways... Go watch some of Buildzoids memory timing videos on Youtube (Actually Hardcore Overclocking) if you want an overload of technical information. Or just start at Wikipedia and look at the technical sources linked from there.

I would not recommend diving to deep into this unless you have a technical need to. The most important detail is that DDR4 and DDR5 latencies/speed rating is not comparable apples-to-apples. So, buy the best speed/latency memory you can get for your money.

Why can't we get a 6000MHz kit running at CL14? Shortest answer: Physics

 

[ir_cow]

Even the best binned kits with a ton of voltage can't go below CL26. Like DDR5-4800 with 1.65V can't go below 24.

This is just the cycle. It's like asking why DDR4 can't do CL1 like DDR-400 could.

I think a lot of people confused about how memory works. system memory is just dumb and operates at a static frequency. So it's up to the memory controller to send commands and that can only send commands so fast and receive data without errors. The higher you go in frequency the looser the timings have to be. Some of that can be countered with more voltage but generally it's a fairly linear process.

 

[Wirko]

If you want to make any meaningful comparison between generations, you need to calculate latencies in nanoseconds. For any DDR memory, it's CL times 2000 divided by speed. You'll see that it has remained basically the same since ... ever. The underlying technology of memory cells is as slow as it has ever been. DDR5 chips do far, far more internal operations in parallel than the original DDR, actually the number doubles with each generation, and that's why speed went up and keeps going up.

Now to be fair, there was some regression with DDR5, which remains to this day. I bought a fairly low end and cheap DDR4-2400 CL16 kit, one without heatsinks, in July 2016, about a year after DDR4 was introduced to consumer PCs. The CL is 13.3 nanoseconds. DDR5 is now a year old, and everything that can be described as fairly low end and cheap (and some may take it as an insult) is still DDR5-5200 CL40, which amounts to 15.4 nanoseconds.

 

[ir_cow]

The primary timings are less important to the frequency and sub-timimgs. The "latency" calculation isn't that helpful in comparison. This assumes all memory has identical timings and the frequency is the only changing value.

I can gain 5 fps just by changing the sub-timimgs. Which is why when I get a 6000 CL40 kit, it can still preform decent to a CL30 kit. Just depends on what you're doing with the subtimings and of course games matter as well. Some games are more memory sensitive than others.

 

[80-watt Hamster]

Latency is limited in part by how fast the individual transistor gates can switch. This is a physical limit regardless of how many MHz you're running. "Good" memory runs CAS at 10ns or less, which works out to the DDR rating divided by 200. So back in the DDR1 days, decent sticks were DDR-400 CAS 2. Then DDR2-800 CAS 4, 1600/8 and 3200/16. DDR5 works slightly differently than previous generations, so we may not see 200:1 CL ratios for DDR5.

 

[ir_cow]

Latency is limited in part by how fast the individual transistor gates can switch. This is a physical limit regardless of how many MHz you're running. "Good" memory runs CAS at 10ns or less, which works out to the DDR rating divided by 200. So back in the DDR1 days, decent sticks were DDR-400 CAS 2. Then DDR2-800 CAS 4, 1600/8 and 3200/16. DDR5 works slightly differently than previous generations, so we may not see 200:1 CL ratios for DDR5.

That's a good description of what the best CAS value should be for each generation. But it does not factory in any of the subtimings that play a role in the performance of the memory kit. Adding things like CPU frequency, being GPU bound for it given resolution, makes it very very hard to say this is better than that. Unless you have set parameters to work with.

I can setup a system with identical hardware and beat out 6000 CL32 memory with a CL40 kit. This is because the CAS value isn't the only number that matters.

 

[80-watt Hamster]

That's a good description of what the best CAS value should be for each generation. But it does not factory in any of the subtimings that play a role in the performance of the memory kit. Adding things like CPU frequency, being GPU bound for it given resolution, makes it very very hard to say this is better than that. Unless you have set parameters to work with.

I can setup a system with identical hardware and beat out 6000 CL32 memory with a CL40 kit. This is because the CAS value isn't the only number that matters.

Right, but I don't understand secondary and beyond timings, so focused on CAS because that's relatively straightforward and kind of acts as standard bearer for a module's rating.

 

[OkieDan]

Latency is limited in part by how fast the individual transistor gates can switch. This is a physical limit regardless of how many MHz you're running. "Good" memory runs CAS at 10ns or less, which works out to the DDR rating divided by 200. So back in the DDR1 days, decent sticks were DDR-400 CAS 2. Then DDR2-800 CAS 4, 1600/8 and 3200/16. DDR5 works slightly differently than previous generations, so we may not see 200:1 CL ratios for DDR5.

We already have sub 10ns DDR5

 

[80-watt Hamster]

We already have sub 10ns DDR5
View attachment 269664

Did not know that.

 

[Count von Schwalbe]

IMHO the first word latency is a better measure, because it can be compared across various frequencies.

I would love to see a comparison of generations with identical frequency and timings and see if any of the other changes (such as subdividing the bus) actually affects performance.

 

[Wirko]

We already have sub 10ns DDR5

Good news for those who are planning to build an Epyc server with ECC DDR5: you can get twice as many gigabytes for the same amount of money.

 

[ir_cow]

IMHO the first word latency is a better measure, because it can be compared across various frequencies.

I would love to see a comparison of generations with identical frequency and timings and see if any of the other changes (such as subdividing the bus) actually affects performance.

Only one you can do that with is Intel DDR4 vs DDR5. Also you would have to raise the DDR4 timings since DDR5 cannot go as low.

 

[Count von Schwalbe]

Only one you can do that with is Intel DDR4 vs DDR5. Also you would have to raise the DDR4 timings since DDR5 cannot go as low.

Ya, I was more thinking DDR3 vs 4.

 

[P4-630]

We already have sub 10ns DDR5

And again so cheap in your country.....
You are lucky you live in the US..

 

[ir_cow]

Ya, I was more thinking DDR3 vs 4.

You need a CPU that supports both DDR3 and DDR4. Do any exist?

 

[Count von Schwalbe]

Didn't some Broadwell models?

 

[ir_cow]

Didn't some Broadwell models?

Yes, but not the same socket, so you can't use the same CPU which defeats the whole test.

Edit: Take that back - BGA 1667.

 

[Shrek]

Don't forget that the faster the RAM runs the CAS latency (in clock cycles) increases for the same delay...
for example, for my DDR3 RAM

Frequency	CAS# Latency
381.0 MHz	5
457.1 MHz	6
533.3 MHz	7
609.5 MHz	8
685.7 MHz	9
761.9 MHz	10
800.0 MHz	11

the latency is basically staying the same

So, if one gets faster RAM (DDR5) the CAS may look like it is worse

You want really good CAS number, run the RAM really, really slow...

 

[ir_cow]

@Shrek Okay, but if CL24 is the lowest DDR5 goes even at DDR5-1600, that kinda breaks that idea of slower memory being better. Also I want to point out once again, while the CAS value will determine the the "theoretical" performance, it accounts for nothing else that plays a equally important role.