Overclock.net banner
1 - 8 of 8 Posts

mike7877

· Registered
Joined
·
1,627 Posts
Discussion starter · #1 ·
So recently I got a laptop with LPDDR5, and now I'm wondering...

For the longest time I've been under the impression that laptop memory is the same as desktop, just on a smaller PCBs and lower voltage. For example, LPDDR3, I thought, was essentially DDR3L (so DDR3 that runs at 1.2V), soldered to a laptop-sized PCB. But after reading the articles I've found, I've learned that LPDDR5 has higher latency than "normal" DDR5, and also, it may not be good for constant throughput?

So is LPDDR5 a crappier version of RAM that will throttle? Can RAM somehow do that to itself now?

I hope the performance difference between them isn't too severe - that LPDDR5 is as close to undervolted + binned DDR5 as possible...

I've done the AIDA memory bandwidth test:

58,466 MB/s (Read)
69,558 MB/s (Write)
56,104 MB/s (Copy)
114.3 ns (Latency)

I think the RAM is 4800. It's said 4800 on the manufacturer's description sheet, but AIDA reports its speed is 6400. HWiNFO then reports the stuff actually runs at 4800MHz (9600!!). And when I look up the RAM, it appears to be 7500... What a pile!

Going by the results, it does appear to be running at 4800
Dual channel 4800 max speed (sans timings) is 76,480MB/s

Timings reported are 52-44-44 at 2400MHz...

It kind of adds up, but I don't know - I've never had DDR5 (still on 5800X3D and 9600K lol), is 58GB/s read acceptable? I think I get more with my DDR4-3900 lol.
 
LPDDR5 is the same technology. They solder it for better signal integrity (closer to the CPU) for higher frequency at lower voltages but they have to bump up latency timings as well. Hypothetically, you can increase voltages and overclock and get better timings.
 
  • Rep+
Reactions: mike7877
RAM latency has massive implications with CPU performance and LPDDR5 have significantly worse latency over DDR5.
Most AMD mobile chips also only have half the L3 cache over their desktop counter parts so they have to rely more on RAM latency performance.
 
Discussion starter · #4 ·
LPDDR5 is the same technology. They solder it for better signal integrity (closer to the CPU) for higher frequency at lower voltages but they have to bump up latency timings as well. Hypothetically, you can increase voltages and overclock and get better timings.
It would have been really nice in the 3-4 articles I was fed by search had explained the performance difference between the two is essentially entirely the product of clock speed and timings...

The articles I found were really bad... So bad they should be taken down...The way they were talking, you'd think LP-DDR they'd done some real work : like Tualatin to Pentium-M. Or better yet, Core 2 mobile to Sandy Bridge mobile

Anyway... It makes sense it's not some super involved system. Thanks
 
It would have been really nice in the 3-4 articles I was fed by search had explained the performance difference between the two is essentially entirely the product of clock speed and timings...

The articles I found were really bad... So bad they should be taken down...The way they were talking, you'd think LP-DDR they'd done some real work : like Tualatin to Pentium-M. Or better yet, Core 2 mobile to Sandy Bridge mobile

Anyway... It makes sense it's not some super involved system. Thanks
Manufacturers are highly incentivize to use LPDDR5 over DDR5 or even CAMM2 LPDDR5x

LPDDR5 is soldered which means the consumer is forced to spend more money to get higher capacity and the manufacturers are more than happy to charge the consumer exorbitant prices.
LPDDR5 being soldered also forces the consumer to buy a new device once their old device can no longer meet their growing demand.
LPDDR5 being soldered also means that the average user cannot service their device in the event the RAM had failure and the manufacturers can justify the value proposition of their warranty services.
 
RAM latency has massive implications with CPU performance and LPDDR5 have significantly worse latency over DDR5.
Most AMD mobile chips also only have half the L3 cache over their desktop counter parts so they have to rely more on RAM latency performance.
Its also use case specific. Gaming prefers latency, but bandwidth is also important for other tasks. However, LPDDR5 isn't designed for performance, it designed for low power.

It would have been really nice in the 3-4 articles I was fed by search had explained the performance difference between the two is essentially entirely the product of clock speed and timings...

The articles I found were really bad... So bad they should be taken down...The way they were talking, you'd think LP-DDR they'd done some real work : like Tualatin to Pentium-M. Or better yet, Core 2 mobile to Sandy Bridge mobile

Anyway... It makes sense it's not some super involved system. Thanks
Articles these days are click bait for views and traffic. Google search has sold out and I don't use it anymore. Unfortunately, I haven't found a great alternative but surprisingly I've found Bing to be decent at time.

Manufacturers are highly incentivize to use LPDDR5 over DDR5 or even CAMM2 LPDDR5x

LPDDR5 is soldered which means the consumer is forced to spend more money to get higher capacity and the manufacturers are more than happy to charge the consumer exorbitant prices.
LPDDR5 being soldered also forces the consumer to buy a new device once their old device can no longer meet their growing demand.
LPDDR5 being soldered also means that the average user cannot service their device in the event the RAM had failure and the manufacturers can justify the value proposition of their warranty services.
While true, LPDDR5 has some benefits to include low power/heat and low profile for thin and light laptops, tablets, and phones. Theoretically, soldered ram should always have better signal integrity than any other alternative.

I've always wanted to see CPU's or even motherboards come with soldered high performance ram (that can be tweaked by the user). Maybe keep two ram slots for optional expansion and treat the faster soldered ram as a big last level cache?
 
  • Rep+
Reactions: mike7877
Its also use case specific. Gaming prefers latency, but bandwidth is also important for other tasks. However, LPDDR5 isn't designed for performance, it designed for low power.


Articles these days are click bait for views and traffic. Google search has sold out and I don't use it anymore. Unfortunately, I haven't found a great alternative but surprisingly I've found Bing to be decent at time.



While true, LPDDR5 has some benefits to include low power/heat and low profile for thin and light laptops, tablets, and phones. Theoretically, soldered ram should always have better signal integrity than any other alternative.

I've always wanted to see CPU's or even motherboards come with soldered high performance ram (that can be tweaked by the user). Maybe keep two ram slots for optional expansion and treat the faster soldered ram as a big last level cache?
LPCAMM2 is fully modular and has similar power consumption over soldered LPDDR5.

L1 has less than 1 ns latency
L2 has 2~4 ns latency
L3 has 10~15 ns latency
LPDDR5x 7500 has 110+ ns latency
SODIMM DDR5 5600 has 90+ ns latency
a highly tuned DDR5 6000 1:1 on AM5 with virtualization off has 60 ns latency
Optane DIMM has ~300 ns latency
Optane SSD has ~10,000-20,000 ns (10-20 µs) latency
High-End PCIe Gen4 Nvme SSD has ~25,000-150,000 ns (25-150 µs) latency
SATA SSD has ~100,000-500,000 ns (100-500 µs) latency

RAM is way too slow to be used as a cache.
You also do not want to mix and match RAM because it will only run as fast as the slowest link.
 
LPCAMM2 is fully modular and has similar power consumption over soldered LPDDR5.

L1 has less than 1 ns latency
L2 has 2~4 ns latency
L3 has 10~15 ns latency
LPDDR5x 7500 has 110+ ns latency
SODIMM DDR5 5600 has 90+ ns latency
a highly tuned DDR5 6000 1:1 on AM5 with virtualization off has 60 ns latency
Optane DIMM has ~300 ns latency
Optane SSD has ~10,000-20,000 ns (10-20 µs) latency
High-End PCIe Gen4 Nvme SSD has ~25,000-150,000 ns (25-150 µs) latency
SATA SSD has ~100,000-500,000 ns (100-500 µs) latency

RAM is way too slow to be used as a cache.
You also do not want to mix and match RAM because it will only run as fast as the slowest link.
I don't think you're picking up what I'm putting down, but ok...
 
1 - 8 of 8 Posts