News DDR5-4800 Clashes With DDR4-3200 In Alder Lake Benchmarks

[Admin]

New UserBenchmark submission reveals the impact of DDR4 memory with Intel's Alder Lake hybrid processors.

DDR5-4800 Clashes With DDR4-3200 In Alder Lake Benchmarks : Read more

 

[mamasan2000]

Both are CPU throttled. And the DDR4 systems background CPU is at 57%, that is far from normal. They must be gaming or something at the same time.
These are my results: https://www.userbenchmark.com/UserRun/45402779
Ryzen 5600x, static 4.5 Ghz OC, HyperX 3600 Mhz CL18 OC'ed to 3800 Mhz CL17.
Plus it's UserBenchmark...hard to draw any conclusions.

 

[Howardohyea]

oh userbenchmark... the April Fools joke that never ends - YouTube
I'm sure even with unreleased hardware UBM will find a way to bias it towards Intel

 

[Drazen]

Real test must ensure both RAMs have identical latencies, only different frequency could be allowed but better to keep same.
Also what matters is number of blocks (pages) in DDR chips.

Also when specifying benchmark results please specify is higher number better or worse!

 

[Makaveli]

I couldn't find much info on these kingston modules but they look to be the standard Cas 16 stuff.

And as noted above this is userbenchmark....

 

[mihen]

400mhz vs 300mhz clock. Surprise - the 400mhz clock reads and writes faster.

 

[ottonis]

If these early benchmarks are remotely indicative of performance deltas we are going see between DDR4 vs DDR5 modules on AL, then this situation reminds me slightly of the mid 1990s with Intel Pentium 60 MHz vs 486 DX-100 - two generations of CPUs clashing, with the new architecture having much higher potential but its very first iterations still having a very hard time in competing with the latest and fastest of its predecessors.

So, back in the day, for someone owning a 486 DX-100 it was wise not to get the P60 but rather wait a couple of iterations, e.g. the Pentium 120 or 166 MHz in order to fully embrace the benefits of gen-on-Gen performance leap.

There is some possibility that we might see a similar performance overlap between early DDR5 and latest and fastest DDR4 modules on the AL platform, with DDR5 probably needing 1-2 years until it's mature enough to substantially outperform the previous gen.

 

[roblittler77]

Real test must ensure both RAMs have identical latencies, only different frequency could be allowed but better to keep same.
Also what matters is number of blocks (pages) in DDR chips.

Also when specifying benchmark results please specify is higher number better or worse!

All things being equal higher frequency RAM is always going to have higher latency timings than lower frequency RAM because the latency is absolute and the timings are relative to the frequency.

 

[escksu]

Real test must ensure both RAMs have identical latencies, only different frequency could be allowed but better to keep same.
Also what matters is number of blocks (pages) in DDR chips.

Also when specifying benchmark results please specify is higher number better or worse!

Ddr5 and ddr4 have different lactencies and frequency.

 

[JerryC]

oh userbenchmark... the April Fools joke that never ends - YouTube
I'm sure even with unreleased hardware UBM will find a way to bias it towards Intel

You think it is bad now. Go hunt down the articles of AMD and Apple complaining because Intel has almost every drop of fab space locked up for the next 4+ years

 

[mdd1963]

So it will take DDR4-4800 MHz to roughly equal the same performance when equipped with DDR4-3200 MHz?

Not exactly the performance 'awe- inspiring' performance jumps from RAM-sensitive apps was looking for...thus far....; but, I vote we wait for some real applications that matter... (Like BF1 and BF5 frame rates at 1080P, Cinebench, etc...!)

 

[Specter0420]

Why all the hate for UserBenchmark? I know AMD fanboys had an issue with how the overall CPU ranking was calculated, which was warranted in many cases, but it is a benchmark... A series of math and physics problems that are timed\scored. Why would it not be useful to compare these individual scores between configs, especially CPUs of the same brand?

 

[Conahl]

Why all the hate for UserBenchmark?

why ? go to google and type " is userbenchmark biased towards intel " and you will probably realize why.

dunno how reliable this is but it sure doesnt paint a usable picture for userbenchmark.

there are a few other sites that say the same, and GN and HUB also dont seem to care much for the site and benchmark as well.

 

[Alex/AT]

Totally unsurprising.
Given the specs, a 16-bit 4-channel pair of DDR5 "6400 MT" should be roughly equal to 32-bit 2-channel pair of DDR4 "3200 MT", with a bit lower random access latency (due to split channels).
This may actually prove the theory that if you don't have an APU, you can safely skip moving to DDR5. But well, these are just early tests so only the future will show if it's true or not.

 

[Specter0420]

why ? go to google and type " is userbenchmark biased towards intel " and you will probably realize why.

dunno how reliable this is but it sure doesnt paint a usable picture for userbenchmark.

there are a few other sites that say the same, and GN and HUB also dont seem to care much for the site and benchmark as well.

I guess you need to be above a brown-belt it Google-foo to understand what I am saying. You clearly didn't read the rest of my comment and\or the article you linked because we say the same thing. The "overall" rating is controversial, where they rank AMD's 12-core below a four-core Intel because the Intel has better single-core performance, especially after overclocking. They put too much weight on single-core performance for their "overall ranking", everyone knows that. But...

Nothing you said applies here. We're comparing the benchmark results, the raw numbers... Even if you pretend the controversial "overall ranking" is in play (it's not), we're comparing Intel to Intel here. It's simple, you can still use the results to compare the raw numbers for each stage of the benchmark, ESPECIALLY within the same brand.

 

[bigpinkdragon286]

You think it is bad now. Go hunt down the articles of AMD and Apple complaining because Intel has almost every drop of fab space locked up for the next 4+ years

This is certainly one way to deal with competition...

 

[Conahl]

I guess you need to be above a brown-belt it Google-foo to understand what I am saying. You clearly didn't read the rest of my comment and\or the article you linked because we say the same thing. The "overall" rating is controversial, where they rank AMD's 12-core below a four-core Intel because the Intel has better single-core performance, especially after overclocking. They put too much weight on single-core performance for their "overall ranking", everyone knows that. But...

Nothing you said applies here. We're comparing the benchmark results, the raw numbers... Even if you pretend the controversial "overall ranking" is in play (it's not), we're comparing Intel to Intel here. It's simple, you can still use the results to compare the raw numbers for each stage of the benchmark, ESPECIALLY within the same brand.

actually i did read what you typed, and a good chunk of the article i linked. you asked, " Why all the hate for UserBenchmark? " and i showed you, considering userbencghmark was biased before, i wouldnt trust them now, many others on here dont as well, toms did an article on how biased the site and benchmark is towards intel.

either way, userbenchmark, is just fluff, and garbage, ill go to a more reputable site like here, AT, GN etc for cpu reviews and comparisons.

 

[Alex/AT]

Userbenchmark is utter garbage when we attempt to compare different vendors using their 'results'.
But here's the comparison inside the favored vendor of theirs, so we can still draw some conclusions.

 

[Howardohyea]

Userbenchmark is utter garbage when we attempt to compare different vendors using their 'results'.
But here's the comparison inside the favored vendor of theirs, so we can still draw some conclusions.

okay I have another piece of suggestion as to why userbenchmark is bad and we can't draw any conclusions:

their graphics "tests" are ran at 800x600 res for one second, not to mention DX 9 and 10... ancient by today's standards and DX12 GPUs won't be accurately compared. Just look at 3DMark, a full minute of 1440P DX12 rendering, making it extremely less prone to interference from the system OS as well.
I can only assume their CPU benchmark is the same reliability.

 

[Alex/AT]

I can only assume their CPU benchmark is the same reliability.

Well, I surely can't disagree it's not reliable per se. But here we have 'raw' memory bandwidth/latency numbers which are not any 'points', so we still can take them at face value, even if with a good grain of salt.
Not for 'absolute' comparisons of course, just for relative weighing inside a single kind of platform. I also assume single core part of these numbers is core performance bound mostly, so we get close results for both RAM kits.