News:

Willkommen im Notebookcheck.com Forum! Hier können sie über alle unsere Artikel und allgemein über Notebook relevante Dinge disuktieren. Viel Spass!

Main Menu

Post reply

The message has the following error or errors that must be corrected before continuing:
Warning: this topic has not been posted in for at least 120 days.
Unless you're sure you want to reply, please consider starting a new topic.
Other options
Verification:
Please leave this box empty:

Shortcuts: ALT+S post or ALT+P preview

Topic summary

Posted by uipogin
 - August 15, 2024, 18:31:00
Quote from: RobertJasiek on July 03, 2024, 08:37:26"AIDA64 / Memory Latency 7.8ns"
I continue to disbelieve these Snapdragon values that are about 4 times faster than the fastest desktop RAM latencies! Without evidence by application testing, such as database queries, confirming 4x speed, this must be a measurement error of the benchmarking tool.
Snapdragon phones had 7-8ns memory latency on LPDDR5 since like many years ago. Just measured mine, 8.1ns
Posted by RobertJasiek
 - July 03, 2024, 20:59:15
Quote from: RoJaBinks on July 03, 2024, 19:39:36"desktop RAM" is miles away from CPU physically

By your argument, every LPDDR5 should be faster than desktop RAM. Reality: LPDDR5 latency is ca. 3 to 5 times slower than the fastest desktop RAM.

The speed of electrons is ~3 * 10^8 m/s. The distance to desktop RAM is ~0.05m, in both directions together ~0.1m. So per 1s, electrons travel ~3 * 10^9 times from / to RAM for the travel duration 1/3ns. Insignificantly small relative to aforementioned RAM latency values from 7.4ns to 150ns?
Posted by RoJaBinks
 - July 03, 2024, 19:39:36
Quote from: RobertJasiek on July 03, 2024, 19:25:38Thanks for confirming that I am the only one who has discovered the issue
Man what is even wrong with you on so many levels.

Ignorant kid keeps on mumbling and doesn't realize "desktop RAM" is miles away from CPU physically, Qualcomm only uses 8 narrow 16 bit channels, and rants about his imaginary database after he got smacked into the fact he doesn't know anything about how databases work.

I don't really have any more time for you, Robert. Educate yourself. Or enjoy your state of mentality. No one cares.
Posted by RobertJasiek
 - July 03, 2024, 19:25:38
Before Snapdragon X, the fastest RAM latency has been a desktop RAM with, IIRC, 29.6ns. AIDA64 has measured for Snapdragon X the fastest RAM latency 7.4ns.

7.4ns * 4 = 29,6ns.

The alleged acceleration has the factor 4.

***

On my office PC, I do not care for RAM speeds because my applications on it are slow so it does not matter. On my PC with dGPU, the applications for the dGPU benefit from low latencies so it does matter.

Other people have other needs. For some it matters - for others it does not matter. That's why, usually, NBC makes every test for every device; in particular NBC tests RAM speeds for most notebooks (other than Apple's, for which testing them seems to be harder).

One might argue that first generation Snapdragon X notebooks are so restricted in their capabilities that RAM latency would not matter. However, many reviewers like to test demanding softwares on these notebooks and NBC provides AIDA64 numbers apparently thinking that it might interest some people - those with demanding softwares.

***

Thanks for confirming that I am the only one who has discovered the issue - affirmation by an anonymous is, wait for it, affirming;)

***

Databases (or any strongly latency-dependent software) work in the manner one tells them to work. If one programs and tells them to reside in RAM during program execution with automated database operations, they can test the impact of RAM latency on such a software.
Posted by JabeekRosirt
 - July 03, 2024, 18:56:16
Quote from: RobertJasiek on July 03, 2024, 17:41:31Qualcomm accelerated RAM 4x
It is completely ignorant to say it is "4x acceleration".
Quote from: RobertJasiek on July 03, 2024, 17:41:31Why would they not have advertised it?
Because it doesn't matter in real-world applications and no one cares.
Quote from: RobertJasiek on July 03, 2024, 17:41:31Do they all naively believe AIDA64 or are also doubts expressed?
You are the only one with doubts, OH MY GOD MAN YOU HAVE UNCOVERED A CONSPIRACY. For sure, aha.
Quote from: RobertJasiek on July 03, 2024, 17:41:31As if I didn't know the difference between loading from permanent storage, executing in RAM and saving to permanent storage.
Databases do work like that only in your imagination. Yeah, that's a yawn indeed.

Stop being an ignorant sissy reliant on external approval running around with these theories, getting constantly smacked, disappearing and then running around with other ideas.
Posted by RobertJasiek
 - July 03, 2024, 17:41:31
TechPowerUp has used the same AIDA64 and said "impressive". Unlike NBC, TechPowerUp has noticed the unusual value. Like, NBC and you, it has uncritically assumed that the benchmark tool would work correctly on the new Snapdragons.

Uh, no, I am not late but noticed on the day of NBC's first new Snapdragon notebook review.

Since you mention Redditt, what are people saying there? Do they all naively believe AIDA64 or are also doubts expressed?

Suppose Qualcomm accelerated RAM 4x. Why would they not have advertised it? Instead, we are flooded with ridiculous AI ads.

Databases in RAM: Yawn. As if I didn't know the difference between loading from permanent storage, executing in RAM and saving to permanent storage.

JFTR, I do not use acronyms on NBC (other than a missing blank before the surname).
Posted by JasirtRobeek
 - July 03, 2024, 16:38:46
Quote from: Memory Enthusiast on July 03, 2024, 13:07:43I agree with Robert.
Stop agreeing with yourself using other nicknames Robert. It's too clear it's the same clueless person when "Memory enthusiast" writes something like
Quote from: Memory Enthusiast on July 03, 2024, 13:07:43Measuring memory latency can involve using hardware performance counters.
Aha, hardware performance counters.

Quote from: RobertJasiek on July 03, 2024, 12:52:47It takes a "nobody" using his real name to even notice when the benchmarks suddenly indicate a 4x speed improvement.
You aren't even the one "noticing" it, it's like in the first line in Google in a TechPowerUp article. Reddit is all over it. Everyone kind of knows since day 1 and doesn't care, you are late to the party, self-proclaimed feature noticer.
Posted by JabiekRasert
 - July 03, 2024, 16:17:42
Quote from: RobertJasiek on July 03, 2024, 12:52:47if the database is in the RAM
Yeah, "database in RAM" to lose all data with a power outage. Guy keeps pretending he knows a thing or two. The whole point of database is to write your data down to persistent storage as fast as possible.

Someone has to educate the guy and explain him how can one lower RAM latency at the cost of slower read/write speeds. Or he will keep thinking he sees some kind of QUALCOMM MAGIC while looking at a kind of usual thing.
Posted by LL
 - July 03, 2024, 14:01:34
Comparing this with Xiaomi Redmi Book by half the price and better...
Posted by Memory Enthusiast
 - July 03, 2024, 13:16:32
Before anyone jumps on my statement that the Qualcomm laptops aren't doing anything that AMD or Intel aren't - I mean it solely in terms of the RAM used, with the LPDDR being soldered to the motherboard. Even if the soldered ram were integrated like Apple or like Intel's upcoming lunar lake, I don't think we'd see such enormous reductions in latency, LPDDR has fundamental limitations. While I don't know about Apple's access times, I know that HBM, which is similarly integrated has very high latencies (partially because of the way it is designed, favoring throughput at the cost of latency.) I was under the impression that LPDDR has lower power consumption and higher throughput (making it great for APUs and handheld gaming devices) at the cost of higher latencies as compared to equivalent DDR.
Posted by Memory Enthusiast
 - July 03, 2024, 13:07:43
I agree with Robert. 7-8ns RAM access latencies are unheard of; these Qualcomm laptops aren't doing anything that Intel or AMD aren't. Unfortunately, AIDA64 doesn't seem to be open source.

Can the reviewer confirm whether this is an x86 app/was there an ARM native app available? Measuring memory latency can involve using hardware performance counters. If this is an x86 app, it might be trying to access performance counters that don't exist in ARM devices; this should usually throw an exception, but maybe their code proceeds anyway - can't tell because it's closed source. 7ns seems more in line with what L3 cache access latency would be like, (it is usually 15-20 cycles in modern devices.)
Posted by RobertJasiek
 - July 03, 2024, 12:52:47
For a long time, RAM speeds have only increased slowly and steadily, and ca. 29.6ns has been the record for latency of desktop RAM while latency for notebook RAM has been much slower, such as 80 ~ 150ns. Now that suddenly 7.4 ~ 11ns is measured for Snapdragon means up to ca. 4 times faster than desktop and ca. 10x faster than pevious notebook RAM, and this suddenly, unexpectedly and unnoticed in its size of improvement by all tech media (incl. NBC, who has not noticed the relevance of its own measurement, still fails to comment, also publishes many immaterial news but cannot post about the greatest chip hardware news in years, the alleged 4x speed improvement) should mean only one thing: a measurement error that AIDA cannot report correct latency on new Snapdragon chips.

RAM latency is about fast execution of small data, as is database queries especially if the database is in the RAM, which therefore is a good example of a kind of application with which to verify or discard the AIDA benchmark.

It takes a "nobody" using his real name to even notice when the benchmarks suddenly indicate a 4x speed improvement.
Posted by chris@amd
 - July 03, 2024, 11:04:57
Funny, NB-Check also reviewed redmibook pro 14 with intel 155H at the same time as this HP. in every metric (display, build, performance, battery life) it is a head and shoulder above while it's 80 percent cheaper ! Real DOA scenario...
Posted by JebartRisoek
 - July 03, 2024, 10:51:54
$1700 for that ??

LOL
With their rather weak level of performance, and compatibility issues, Snapdragon laptop should not cost more than $1000/1200
Posted by JabertRosiek
 - July 03, 2024, 09:42:34
Quote from: RobertJasiek on July 03, 2024, 08:37:26I continue to disbelieve
Good thing you are a nobody so your disbelief doesn't matter much to reality.
"Database queries" in RAM, what a tool.