The problem is finding any DDR5 ram for sale

Just curious, why aren’t we calling this Ghz yet? 6.4 GHz - we don’t even have cpus running that fast. No one would dare say I’ve got a 5600x running at 4200Mhz. But with memory it’s just like ehhhh… whatever.

Timings?

If it is doing so with a CAS latency of 28 or loewr then itd be interesting. Otherwise its very meh.

TheinsanegamerNIf it is doing so with a CAS latency of 28 or loewr then itd be interesting. Otherwise its very meh.

Not a chance at these frequencies, not at any safe daily voltages anyway.

ymbajaJust curious, why aren’t we calling this Ghz yet? 6.4 GHz - we don’t even have cpus running that fast. No one would dare say I’ve got a 5600x running at 4200Mhz. But with memory it’s just like ehhhh… whatever.

There is no ram running at 6.4 GHz. The bus speed is half the data rate. The reason we don't say GHz is because technically the DDR rate numbers are not referring to clock speed, but Mega Transfers per second.

DDR4-3200 for example runs at 1600 MHz (bus speed or "command rate", the rate at which commands are issued), but its data rate is 3200 (3200 Mega Transfers per second). Got that? Command rate is half the data rate. The data rate is the number you usually see quoted for marketing reasons. It is called "Double Data Rate" DDR for a reason.

It's confusing to think about "Hz" because it is the inverse of time, it is easier to just think of it as time directly instead. The problem with switching back and forth is that when something increases, the inverse decreases, so it is hard to think about clearly. Inverses, just like negatives, mess up the brain.

There are also a lot of definitions that people mess up, as there are many words that are basically talking about the same thing. So bus speed, clock speed, command rate, or cycle rate, those are all related.

Transfer speed, MHz rating, DDR speed, data rate, those are all the same, and are all double the others.

Speed and time are inverses, so the transfer rate is double the cycle rate, but the transfer TIME is half the cycle TIME. I'll show that here:

So CAS latency operates on the command rate, ie the clock speed of 1600 Mhz for DDR4-3200. Take the inverse of clock speed to get the minimum time required to issue a command: 1 / 1,600,000,000 hz = 0.625 x 10 to the power -9 is 0.625 nano seconds. CAS of 14 means it takes 14 cycles to finish the command, so 14 times 0.625 nanoseconds = 8.75 nanoseconds. That's the real speed of your ram. DDR4-3200 takes 8.75 nanoseconds for a first word read command.

Do the same thing for DDR5 6400. Actual bus speed is 3200. IE everything is twice as fast as the DDR4: instead of 0.625 nanoseconds, it takes half that, or 0.3125 nano seconds to cycle. But a command like CAS (the time to get data after a read command) could take 40 cycles. We don't know, Corsair didn't even announce it, and it isn't in this article. Let's say it is 40. Then 40 * 0.3125 nanoseconds is 12.5 nano seconds.

So DDR5-6400 CAS 40 takes 12.5 nanoseconds to start reading data (the first word, to start the read), DDR4-3200 CAS 14 takes 8.75 nanoseconds to read the first word of data, so DDR4 is actually much faster at this operation in latency terms, if you are reading one word at a time. That's why you don't necessarily get faster speeds with DDR5. However even though the read latency is worse with DDR5, the data transfer rate after a command can be faster as the number of Mega transfers per second possible is now higher with DDR5. Once the read starts, you can read more data in the same amount of time. 2nd word, 3rd word, etc. Think sequential speed is higher with DDR5, but random i/o is slower. Like an SSD.

Hope this makes sense. I'm a Mathematician, not an electrical engineer, but I think I got the basics right. It is hard to explain. I wrote this to try to make English out of my understanding. I've struggled to explain it to someone before. Enjoy :)

GarrusIt's confusing to think about "Hz" because it is the inverse of time, it is easier to just think of it as time directly instead. The problem with switching back and forth is that when something increases, the inverse decreases, so it is hard to think about clearly. Inverses, just like negatives, mess up the brain

Inverses and integers are fine, its when we start throwing in "i' or imaginary numbers that messes things up. :(

Whats DDR5 going to scale to, we going to see 64GB DIMMS?

I think I saw 128GB are possible/happening?

GarrusThere is no ram running at 6.4 GHz. The bus speed is half the data rate. The reason we don't say GHz is because technically the DDR rate numbers are not referring to clock speed, but Mega Transfers per second.

DDR4-3200 for example runs at 1600 MHz (bus speed or "command rate", the rate at which commands are issued), but its data rate is 3200 (3200 Mega Transfers per second). Got that? Command rate is half the data rate. The data rate is the number you usually see quoted for marketing reasons. It is called "Double Data Rate" DDR for a reason.

It's confusing to think about "Hz" because it is the inverse of time, it is easier to just think of it as time directly instead. The problem with switching back and forth is that when something increases, the inverse decreases, so it is hard to think about clearly. Inverses, just like negatives, mess up the brain.

There are also a lot of definitions that people mess up, as there are many words that are basically talking about the same thing. So bus speed, clock speed, command rate, or cycle rate, those are all related.

Transfer speed, MHz rating, DDR speed, data rate, those are all the same, and are all double the others.

Speed and time are inverses, so the transfer rate is double the cycle rate, but the transfer TIME is half the cycle TIME. I'll show that here:

So CAS latency operates on the command rate, ie the clock speed of 1600 Mhz for DDR4-3200. Take the inverse of clock speed to get the minimum time required to issue a command: 1 / 1,600,000,000 hz = 0.625 x 10 to the power -9 is 0.625 nano seconds. CAS of 14 means it takes 14 cycles to finish the command, so 14 times 0.625 nanoseconds = 8.75 nanoseconds. That's the real speed of your ram. DDR4-3200 takes 8.75 nanoseconds for a first word read command.

Do the same thing for DDR5 6400. Actual bus speed is 3200. IE everything is twice as fast as the DDR4: instead of 0.625 nanoseconds, it takes half that, or 0.3125 nano seconds to cycle. But a command like CAS (the time to get data after a read command) could take 40 cycles. We don't know, Corsair didn't even announce it, and it isn't in this article. Let's say it is 40. Then 40 * 0.3125 nanoseconds is 12.5 nano seconds.

So DDR5-6400 CAS 40 takes 12.5 nanoseconds to start reading data (the first word, to start the read), DDR4-3200 CAS 14 takes 8.75 nanoseconds to read the first word of data, so DDR4 is actually much faster at this operation in latency terms, if you are reading one word at a time. That's why you don't necessarily get faster speeds with DDR5. However even though the read latency is worse with DDR5, the data transfer rate after a command can be faster as the number of Mega transfers per second possible is now higher with DDR5. Once the read starts, you can read more data in the same amount of time. 2nd word, 3rd word, etc. Think sequential speed is higher with DDR5, but random i/o is slower. Like an SSD.

Hope this makes sense. I'm a Mathematician, not an electrical engineer, but I think I got the basics right. It is hard to explain. I wrote this to try to make English out of my understanding. I've struggled to explain it to someone before. Enjoy :)

That's all good and true (kudos for going in depth including latencies and all btw) but it's not really a reason to not move to the next metric prefix - giga.

Maybe it's because of the name similarity - mega hertz/mega transfers - but then again they could also say 3.2 giga hertz/giga transfers ¯\_(ツ)_/¯
Most likely it's a left over from the DDR1 (DDR :D) where max speeds were like 400 mhz or something

Anyone else dislike that they moved from diffused lighting, to squares?

Most of the lighting is only visible from the side, and i find that a really weird decision

Dr_b_Whats DDR5 going to scale to, we going to see 64GB DIMMS?

256GB is the maximum per stick on DDR5

I assume we'll see 32GB and 64GB dimms pretty fast, but not much beyond that outside the server world

GarrusHope this makes sense.

For what is worth, that was pretty well explained, IMO.

MusselsAnyone else dislike that they moved from diffused lighting, to squares?

Most of the lighting is only visible from the side, and i find that a really weird decision

... I dislike RGB so to me it just looks like shuffling poop from one place to another /joke

MusselsI assume we'll see 32GB and 64GB dimms pretty fast, but not much beyond that outside the server world

Kinda feels like a bit of a stretch to see 64 GB sticks in consumer/prosumer land. Though I can definitely see 32 GB getting a bit more common.

TheinsanegamerNIf it is doing so with a CAS latency of 28 or loewr then itd be interesting. Otherwise its very meh.

DDR5 does not operate like DDR4.
So cl 36 6000Mhz, stupid math would be like, ohh that's 3000 mhz cl 19, yet it produces far better latencies than that, it actually somewhat matches 4400cl18 ddr4, sooooooooooo.


6400 Cl38 would be pretty acceptable, 6400 cl36 would be amazing performance.

GarrusThere is no ram running at 6.4 GHz. The bus speed is half the data rate. The reason we don't say GHz is because technically the DDR rate numbers are not referring to clock speed, but Mega Transfers per second.

DDR4-3200 for example runs at 1600 MHz (bus speed or "command rate", the rate at which commands are issued), but its data rate is 3200 (3200 Mega Transfers per second). Got that? Command rate is half the data rate. The data rate is the number you usually see quoted for marketing reasons. It is called "Double Data Rate" DDR for a reason.

It's confusing to think about "Hz" because it is the inverse of time, it is easier to just think of it as time directly instead. The problem with switching back and forth is that when something increases, the inverse decreases, so it is hard to think about clearly. Inverses, just like negatives, mess up the brain.

There are also a lot of definitions that people mess up, as there are many words that are basically talking about the same thing. So bus speed, clock speed, command rate, or cycle rate, those are all related.

Transfer speed, MHz rating, DDR speed, data rate, those are all the same, and are all double the others.

Speed and time are inverses, so the transfer rate is double the cycle rate, but the transfer TIME is half the cycle TIME. I'll show that here:

So CAS latency operates on the command rate, ie the clock speed of 1600 Mhz for DDR4-3200. Take the inverse of clock speed to get the minimum time required to issue a command: 1 / 1,600,000,000 hz = 0.625 x 10 to the power -9 is 0.625 nano seconds. CAS of 14 means it takes 14 cycles to finish the command, so 14 times 0.625 nanoseconds = 8.75 nanoseconds. That's the real speed of your ram. DDR4-3200 takes 8.75 nanoseconds for a first word read command.

Do the same thing for DDR5 6400. Actual bus speed is 3200. IE everything is twice as fast as the DDR4: instead of 0.625 nanoseconds, it takes half that, or 0.3125 nano seconds to cycle. But a command like CAS (the time to get data after a read command) could take 40 cycles. We don't know, Corsair didn't even announce it, and it isn't in this article. Let's say it is 40. Then 40 * 0.3125 nanoseconds is 12.5 nano seconds.

So DDR5-6400 CAS 40 takes 12.5 nanoseconds to start reading data (the first word, to start the read), DDR4-3200 CAS 14 takes 8.75 nanoseconds to read the first word of data, so DDR4 is actually much faster at this operation in latency terms, if you are reading one word at a time. That's why you don't necessarily get faster speeds with DDR5. However even though the read latency is worse with DDR5, the data transfer rate after a command can be faster as the number of Mega transfers per second possible is now higher with DDR5. Once the read starts, you can read more data in the same amount of time. 2nd word, 3rd word, etc. Think sequential speed is higher with DDR5, but random i/o is slower. Like an SSD.

Hope this makes sense. I'm a Mathematician, not an electrical engineer, but I think I got the basics right. It is hard to explain. I wrote this to try to make English out of my understanding. I've struggled to explain it to someone before. Enjoy :)

Dude… You went the distance on that one! Thanks for the detailed breakdown. I learned a few things. But…:D While I get that the MHz speed is really more marketing, my point still stands. If you concede that 6400MHz is the same speed as 6.4GHz and Corsair can call it 6400MHz in the press release why not say 6.4GHz?

ymbajaDude… You went the distance on that one! Thanks for the detailed breakdown. I learned a few things. But…:D While I get that the MHz speed is really more marketing, my point still stands. If you concede that 6400MHz is the same speed as 6.4GHz and Corsair can call it 6400MHz in the press release why not say 6.4GHz?

0.64THz for the win

Dr_b_Whats DDR5 going to scale to, we going to see 64GB DIMMS?

May be not anytime soon.

www.anandtech.com/show/17117/sk-hynix-to-manufacture-48-gib-and-96-gib-ddr5-modules

Seems like you may find 48GB instead, if you are looking to step up from 32GB.

windwhirlFor what is worth, that was pretty well explained, IMO.


... I dislike RGB so to me it just looks like shuffling poop from one place to another /joke


Kinda feels like a bit of a stretch to see 64 GB sticks in consumer/prosumer land. Though I can definitely see 32 GB getting a bit more common.

32GB already exists in DDR4 (i know, i run it) - so we know that will happen from day one, and when they feel the need for an elitist product to showcase how awesome DDR5 is, they'll double that.
256GB is the official max stick size, but if it does come out it'd be at the JEDEC minimum speeds. I can imagine some threadripper build running 12x256GB for scientific work.

watzupkenMay be not anytime soon.

www.anandtech.com/show/17117/sk-hynix-to-manufacture-48-gib-and-96-gib-ddr5-modules

Seems like you may find 48GB instead, if you are looking to step up from 32GB.

The sith are going to be very upset that the rule of two is broken
(although i suppose anakin only counted for half at one point...)

GarrusCAS of 14 means it takes 14 cycles to finish the command, so 14 times 0.625 nanoseconds = 8.75 nanoseconds. That's the real speed of your ram. DDR4-3200 takes 8.75 nanoseconds for a first word read command.

This is just too oversimplified. Those 8.75ns are not the real speed of your RAM. I'm not an expert either but the fact that there are like 50 memory timings should give you pause in trying to simplify too much. As a mathematician you should know that if you try to simplify some subjects too much you'll just end up giving a bad explanation of the subject. Some subjects just require a lot of knowledge and can't be broken down in a handful of sentences.

Take DDR4-3200CL14 vs DDR4-3200CL16 for example. The CL16 memory could easily outperform the CL14 memory in every single test you throw at it, even read latency tests. In fact I could completely cripple the CL14's performance without touching the clock speed or CL. According to your explanation that shouldn't be possible.

Probably those will be crappy CL42 or CL44 even.....