Question How fast is DDR4 & DDR5 compared to DDR1?

[GunsMadeAmericaFree]

In effective real world tests, how fast are DDR4 and DDR5 memory dimms compared to DDR1?

I found myself thinking about this today, wondering if DDR4 is actually 4x as fast, etc.......

 

[Tech Junky]

Well, each generation doubles the speed not 1>4 = 4X it's....

https://en.wikipedia.org/wiki/DDR_SDRAM

 

[GunsMadeAmericaFree]

Ah, I see, bandwidth has increased a lot. Also, I didn't realize that voltages had decreased that much. At this point, is there really much difference in building a system with DDR4 vs DDR5?

 

[Tech Junky]

Cost.

DDR5 is still a premium option when you factor in the MOBO and CPU costs in addition to the 50% premium of DDR5. The performance boost of DDR5 in its current release might get you a 10% increase.

Going from a single stick / single controller to a single stick / dual controller / split bus only makes it more complicated to do an apples to apples based on the speed listed on the sticker.

Dropping the 6000mhz to 3000mhz when you factor in the DDR5 controllers splitting each stick into a dual channel single stick. Most of the boost you see in things that matter for a game would come from the GPU being used rather than the RAM of the system.

Now, if you find something that uses a ton of RAM and is sensitive to speed going DDR5 might be better. For most though DDR4 is a better value. Thing have changed though a lot since it released last year with ADL and even more with AMD releasing DDR5 only CPU/MOBO options. It's till quite the ask though for the paltry performance increase.

 

[SPBHM]

DDR1 was mostly 266 to 400MHz
typical DDR4 is 2133 to 3200 both 64bit wide as base.
that's a lot more than 4 times.

memory latency gains are not as impressive I suppose.

 

[GunsMadeAmericaFree]

I've found it interesting how older RAM often goes up in price for a few years, while the price of the newer generation comes down....

 

[Tech Junky]

Supply and demand.

Legacy supply dwindles and new rises.

 

[IntelUser2000]

Most of the boost you see in things that matter for a game would come from the GPU being used rather than the RAM of the system.

Now, if you find something that uses a ton of RAM and is sensitive to speed going DDR5 might be better.

This is like wondering how much total time in your day you save by using the highway that goes at double the speed of normal roads.

Yea you save significant amount of time spent on driving but compared to the entire 24 hours of the day? A much smaller portion.

If you are memory bound that you get anywhere near the bandwidth gains of new memory, then your system is badly, badly optimized. It's like a 12900K system having DDR2-800 speeds. Sure if you double that to DDR3-1600 you'll get a significant improvement. But at modern DDR4 speeds the gains are lot less because it's balanced.

Ever wondered why CPU doesn't scale 100% with clocks? Well, you will if you increased memory and drive performance equal to the amount you increased clock speed. So you are still bound somewhat by memory performance.

In effective real world tests, how fast are DDR4 and DDR5 memory dimms compared to DDR1?

You could say CPUs and memory are pretty much designed to work with a certain generation. Putting DDR4 on a Pentium 4 system would be a total waste. DDR1 on a 13900K system would heavily bottleneck the CPU. But DDR1 on Pentium 4 and DDR4 on 13900K isn't, because the whole system is designed with components that were available at the time.

 

[igor_kavinski]

Putting DDR4 on a Pentium 4 system would be a total waste.

Don't see the point of this statement. How can you be sure it would be a total waste? Maybe the P4 really needed DDR4 at that time, since it performed best with Rambus. It was a very forward looking design at a time when the rest of the industry just wasn't ready for it. They further destroyed it with Prescott. I remember the first time I installed Microsoft Office on a Northwood with DDR. Damn. It whizzed by unlike anything I had seen before. That CPU was a data crunching monster of its time. I also used an Athlon XP 1400+ or something like that at the time. It felt just like a Pentium III. The feeling of speed wasn't there.

Also, this late in DDR4's life cycle, I would have appreciated it if Intel had taken the time to make 13900K work with DDR4-4800 at stock speeds. But no. They just HAVE TO flaunt their dumbness and push DDR5, since that's the future. What about now, you fools? Don't you care about the present, Intellidiots?

 

[Insert_Nickname]

This is like wondering how much total time in your day you save by using the highway that goes at double the speed of normal roads.

Yea you save significant amount of time spent on driving but compared to the entire 24 hours of the day? A much smaller portion.

Just to complete the analogy; when there is heavy traffic, having multiple lanes helps everybody. Otherwise things come quickly to a standstill. The Germans have such a useful short word for that: stau.

Same with dual channel memory.

Don't see the point of this statement. How can you be sure it would be a total waste? Maybe the P4 really needed DDR4 at that time, since it performed best with Rambus. It was a very forward looking design at a time when the rest of the industry just wasn't ready for it. They further destroyed it with Prescott. I remember the first time I installed Microsoft Office on a Northwood with DDR. Damn. It whizzed by unlike anything I had seen before. That CPU was a data crunching monster of its time. I also used an Athlon XP 1400+ or something like that at the time. It felt just like a Pentium III. The feeling of speed wasn't there.

Could also have been the disk system used. Remember, HDDs weren't real fast at random access back then. 4200RPM drives were the worst. Fortunately, they were mostly confined to laptops.

Having SSE2 optimised code also helped the P4 a lot.

 

[IntelUser2000]

How can you be sure it would be a total waste?

Very simple. Memory scaling tests were done and if it were memory bound as to benefit from DDR4, then those same tests would have shown tremendous gains. Alternately what you are suggesting is P4 was entirely bound by memory and could have been competitive with chips like Athlon - no it was just a terrible design.

It was a 1-issue design with Trace Cache(ancestor to a uop cache) in addition to severely lacking in execution resources, L1 cache and super long pipeline. They wanted high clocks for marketing purposes but knew that performance would be atrocious with that long of a pipeline so they had to add resources, but at the same time they cut the amount of decoders to pre-Pentium level but still end up having a 200mm2 die.

Instead, it was normal as any other CPU paired with "generation appropriate" memory. How can I be sure? Well go and find reviews of it. I read dozens of reviews for every P4 release back then.

Also, scaling levels off as bottlenecks get lifted.