Binned DDR5 supposedly hitting 10,000 MT/s

nine.gif
 
MHz and MT/s are two different things, and I really wish they'd stick to just one in their marketing material. :/
The fact they bounce between them throughout the articles and the reference articles doesn’t help.
But either way those are fast numbers.
 
Not for nothing but DDR5 being just around the corner is what makes me not feel so bad about being unable to put together any sort of modern gaming PC at the moment.

Because even if the parts were available, it's kind of a bad time to upgrade. Sure, it's the best of the last of this generation, and the first of the next gen is never all that great, but we're talking about the modern PC market, when a rig can and will last you several years without needing much if any upgrading.

We're at the end of the line for 4c/8th, with 8 gigs of system memory. But man, talk about a long haul for anyone who got in on that back in ... 2012? 2013?
 
Just upgraded our oldest system 4c(8t) for my daughter whose going in to digital art. She’ll have a nice AMD 3900x 12c(24t) system with 32Gigs of RAM. Dad gets the 5900x. :)
 
Not for nothing but DDR5 being just around the corner is what makes me not feel so bad about being unable to put together any sort of modern gaming PC at the moment.

Because even if the parts were available, it's kind of a bad time to upgrade. Sure, it's the best of the last of this generation, and the first of the next gen is never all that great, but we're talking about the modern PC market, when a rig can and will last you several years without needing much if any upgrading.

We're at the end of the line for 4c/8th, with 8 gigs of system memory. But man, talk about a long haul for anyone who got in on that back in ... 2012? 2013?


3570K 16GB DDR3 1333 built in Feb 2013

5950x arrives Sunday, 64GB DDR4 3600

I ride or die with the last of the last gen
 
We're at the end of the line for 4c/8th, with 8 gigs of system memory. But man, talk about a long haul for anyone who got in on that back in ... 2012? 2013?
IIRC 2011 was when I built my 2600k rig, I've never had so long out of a rig (okay it got fried so I replaced MB/CPU with same platform) - a groundbreaking dual core Opteron 939 was the only thing close.
 
Not for nothing but DDR5 being just around the corner is what makes me not feel so bad about being unable to put together any sort of modern gaming PC at the moment.

Because even if the parts were available, it's kind of a bad time to upgrade. Sure, it's the best of the last of this generation, and the first of the next gen is never all that great, but we're talking about the modern PC market, when a rig can and will last you several years without needing much if any upgrading.

We're at the end of the line for 4c/8th, with 8 gigs of system memory. But man, talk about a long haul for anyone who got in on that back in ... 2012? 2013?
This My i7-4790K I bought the parts used in 2014 or so just as DDR 4 was being pushed out I only bought last gen as I wanted to be able to use my existing 2 x 8GB DDR 3 from the i7-2600 PC that a friend gave me minus the HDD (it failed) and power supply! This was in 2012 or so he thought the MB was bad!

My current Ryzen 3700X was built in March of 2020 right before PC parts became unobtanium at MSRP so I got this at a good time!
So far this is my only desktop that has DDR 4 most are DDR 3 and one (My Retro build) has DDR 333!
 
  • Like
Reactions: N4CR
like this
MHz and MT/s are two different things, and I really wish they'd stick to just one in their marketing material. :/

No, they're not two different things. They're two different ways of saying the same thing. Intel has used MHz and MT/s interchangeably for years, I think since the Pentium 4 era.

MT/s was originally designed as a moniker for the effective speed of double or quad data rate in DDR or the Pentium 4's QDR FSB, but MHz kept being used for the same purpose and MT/s lost the reason for it existing in the first place.
 
This is gonna be great for APUs. Even the 6400Mhz rated speeds are a big boost over DDR4 and enough to bring dual-channel configs to 100GB/s (in theory)
 
Last edited:
No, they're not two different things. They're two different ways of saying the same thing. Intel has used MHz and MT/s interchangeably for years, I think since the Pentium 4 era.
Exactly, this site has the best explanation for it:

What Does 2400 MT/s Mean?​

A RAM with 2400 MT/s speed signifies that there are two processes, each operating at 1200 Mhz. The device does not operate at 2400 Mhz, but it offers a data transfer rate of 2400 Mhz. In other words, the user does not need to know what goes on inside the component. Hence, manufacturers mention 2400 MT/s instead of 2400 Mhz.

This is also helpful, and note the difference between MHz with MT/s between SDR SDRAM and DDR SDRAM:
DDR.jpg
 
Exactly, this site has the best explanation for it:



This is also helpful, and note the difference between MHz with MT/s between SDR SDRAM and DDR SDRAM:
View attachment 348451
They're two different things. MHz has been used incorrectly for years, that doesn't make them the same. Hz is a measure of cycles per second. In double data rate RAM 133MHz is still 133MHz, but you have 266MT in that same time frame.

From that link:

What Does 2400 MT/s Mean?​


A RAM with 2400 MT/s speed signifies that there are two processes, each operating at 1200 Mhz. The device does not operate at 2400 Mhz, but it offers a data transfer rate of 2400 Mhz. In other words, the user does not need to know what goes on inside the component. Hence, manufacturers mention 2400 MT/s instead of 2400 Mhz.
And they even manage to mess up their description of MT/s. 2400MT/s means there are 2400 MT/s, regardless of whether you are talking about SDRAM or DDR RAM. A 133MHz SDRAM module has 133MT/s, a DDR2 module would have 266MT/s at the same freq (as your table shows). They mean different things.
 
Last edited:
Not for nothing but DDR5 being just around the corner is what makes me not feel so bad about being unable to put together any sort of modern gaming PC at the moment.

Because even if the parts were available, it's kind of a bad time to upgrade. Sure, it's the best of the last of this generation, and the first of the next gen is never all that great, but we're talking about the modern PC market, when a rig can and will last you several years without needing much if any upgrading.

We're at the end of the line for 4c/8th, with 8 gigs of system memory. But man, talk about a long haul for anyone who got in on that back in ... 2012? 2013?
I was able to grab a 5800X on launch day and I plan to use it in my main system for quite a while. I can wait out the first generation or 2 of DDR5/PCIe5 to let bugs get worked out and let performance develop.
 
Not really. Latency has stayed roughly the same since ddr3 and ddr5 will come in about the same.
Didn't see the initial tests? Latency is roughly double against DDR4, even though bandwidth is much improved. Going to take properly latency adjusted premium memory to get on the same levels. That's years down the road.
 
I'll wait for Buildzoid's inevitable "3 hour rambling about DDR5" video before I make any assumptions about DDR5. Anandtech, Toms & friends may as well not even exist.

Buildzoid / ActuallyHardcoreOverclocking is the guy that will somehow get a mediocre DDR4 kit up to 6000MHz on a $90 motherboard and old $80 zen1 CPU just for fun.
 
This is gonna be great for APUs. Even the 6400Mhz rated speeds are a big boost over DDR4 and enough to bring dual-channel configs to 100GB/s (in theory)
This is probably why AMD hasn't bothered to try slap a faster/better IGPU on... wait for DDR5 and also make some reason to upgrade from current APUs lol.
 
I've been hearing the latency thing for like 18 years since DDR2 launch and yanno, it's never actually been an issue afaik. Yeah the lowest speed grades of the new type are a wash against the highest grades of the old standard but nobody except the earliest adopters and OEMs use the lowest bins anyways. 6400MT/s DDR5 with even a terrible-sounding 32 cycle latency would have the same latency in nanoseconds as 16 cycle 3200MT/s DDR4 which is considered perfectly fine for gaming now.
 
it's never actually been an issue afaik.

And even when it was an issue, it was typically a synthetic issue if not just something obscure. In general use and gaming in particular, it was always better to get the higher bandwidth over lower latency.

I mean, that's usually true even when comparing like memory standards. Sure, there can be sweet spots with certain configurations where the differences can mean more or less, but yeah, we've got a good two decades where as a rule of thumb...
 
Didn't see the initial tests? Latency is roughly double against DDR4, even though bandwidth is much improved. Going to take properly latency adjusted premium memory to get on the same levels. That's years down the road.
At those high mhz that sound very unlikely.

According to anandtech DDR5 3200 latency will be the same has ddr4 and about 4-5% higher at double the bandwidth

DDR5-6400 at CL 46 will have lower latency than DDR3 800 CAS 6
 
This is probably why AMD hasn't bothered to try slap a faster/better IGPU on... wait for DDR5 and also make some reason to upgrade from current APUs lol.

My guess would be that the lack of better igpu has more to do with amd developing a more chiplet-like design for (i)gpu's. The advantage to yields (also in a production constrained environment) makes such a move nearly inevitable, imo.
 
Not really. Latency has stayed roughly the same since ddr3 and ddr5 will come in about the same.
Unless my math is off single access latency of the proposed DDR5-6400 with CL of 40 would be 12.5ns. My DDR4-4000 with CL of 15 is 7.5ns. That's a 67% increase. The Longsys test above is even worse, showing almost a 100% increase in latency.
 
Unless my math is off single access latency of the proposed DDR5-6400 with CL of 40 would be 12.5ns. My DDR4-4000 with CL of 15 is 7.5ns. That's a 67% increase. The Longsys test above is even worse, showing almost a 100% increase in latency.
Is that test latency reported in ns or clock cycles?
 
Unless my math is off single access latency of the proposed DDR5-6400 with CL of 40 would be 12.5ns. My DDR4-4000 with CL of 15 is 7.5ns. That's a 67% increase. The Longsys test above is even worse, showing almost a 100% increase in latency.
Cherry pick your super low latency ddr4 at the end of the ddr4 cycle and compare it to brand new ddr5 on JDEC standards and call it a realistic comparison. LoL
 
Cherry pick your super low latency ddr4 at the end of the ddr4 cycle and compare it to brand new ddr5 on JDEC standards and call it a realistic comparison. LoL
That could be said at the end of any RAM cycle and the beginning of a new one.
If only we could go back to the days of low-latency and ultra-tight timings of DDR1... 2-3-2-5, those were the days!
 
Cherry pick your super low latency ddr4 at the end of the ddr4 cycle and compare it to brand new ddr5 on JDEC standards and call it a realistic comparison. LoL
Okay.

1618852353118.png


DDR4 JEDEC Specifications
1618852379571.png
 
It’s about, what, a 15 percent increase in latency in exchange for a 100 percent increase in bandwidth?

That seems acceptable...
Or 100 percent increase in latency in exchange for 40 percent increase in bandwidth. Maybe the truth will be somewhere in the middleground, but meh still.
 
Or 100 percent increase in latency in exchange for 40 percent increase in bandwidth. Maybe the truth will be somewhere in the middleground, but meh still.
The truth of the matter is most users are not going to see a difference. Consumers are going to see the new shiny big number and swallow their placebo without protest.
 
I don't understand RAM, timings, and Mega-super Hertz. I just find out which one works the best without any massaging from me. Install > Power on > good to go.
 
Back
Top