How Low Can You Go? Memory Latency Competition - AIDA64

[biffzinker]

Got it to boot at 3333MHz after adjusting timings. If I can't get it to boot at 3400MHz, I'll run memtest at 3333MHz to see if it's stable.

Going right to 3400MHz got me a black screen, and not adjusting timings at 3333MHz got me a BSOD on boot.

CPU: R5 2600X
Cooler: Cryorig R1 Ultimate
RAM 2x8GB Crucial Ballistix Sport LT DDR4-3000 @ 3333MHz
Mobo: MSI X470 Gaming Pro

Well that's a shame, you did reduce latency but lost bandwidth in Copy which is most representative of the real memory performance.

 

[storm-chaser]

default afaik.

Thanks for your sub... but to add you to the Points competition leaderboard I'm going to need the remaining data sets from AIDA64. Please install the full version if possible...

@Rock72
I need memory quality and make/model

if at all possible

FYI the leaderboards are updated but I cannot post them here yet. The issue is I can't seem to edit my initial post with updated leaderboards, so I have a request in with the mods to help resolve this.

 

[sam_86314]

Reading the bit on the first post, I vote for separating laptops from desktops because it'll put me in first place.

 

[riversde]

default settings...

 

[sam_86314]

Now at 3400MHz with same timings. Ran Memtest86+ for 25 minutes with no errors.

CPU: R5 2600X
Cooler: Cryorig R1 Ultimate
RAM 2x8GB Crucial Ballistix Sport LT DDR4-3000 @ 3400MHz
Mobo: MSI X470 Gaming Pro

Any ideas on how to improve copy speed?

Also synthetic benchmarks have stopped improving. Guess I've reached the point of diminishing returns for Zen+.

 

[biffzinker]

Any ideas on how to improve copy speed?

You would have to tighten up secondary timings. You could start with tRFC, and tRC.

 

[DeathtoGnomes]

Thanks for your sub... but to add you to the Points competition leaderboard I'm going to need the remaining data sets from AIDA64. Please install the full version if possible...

not gonna happen, i cant justify the cost vs how often I'll use it. maybe you should have thought this out better.

 

[Hugis]

**************************************************************
CPU..........................i7 4770k @ 4.3Ghz
COOLING................H55 AIO Push Pull
MEMORY.................16GB (8GB x 2) Crucial Ballistix BLT8G3D1608ET3LX0
MOTHERBOARD…..MSI Z97 Gaming 5

 

[storm-chaser]

not gonna happen, i cant justify the cost vs how often I'll use it. maybe you should have thought this out better.

Great investment, that's a shame. Thanks for playing!

@sam_86314

I see you have some updated runs. If you want me to use one of those in place of your current result on the leaderboard just let me know.

*Leaderboards Updated*

 

[sam_86314]

Great investment, that's a shame. Thanks for playing!

@sam_86314

I see you have some updated runs. If you want me to use one of those in place of your current result on the leaderboard just let me know.

*Leaderboards Updated*

My 3333Mhz run has the best latency so far. Could you use that one?

 

[birdie]

Nothing special except I'm rocking four 16GB modules. Ran Memtest86 8.3 Pro for three hours and it reported no errors (I'm quite sure many people in this thread cannot boast the same and then they get surprised when their system suddenly BSODs).

 

[phanbuey]

Nothing special except I'm rocking four 16GB modules. Ran Memtest86 8.3 Pro for three hours and it reported no errors (I'm quite sure many people in this thread cannot boast the same and then they get surprised when their system suddenly BSODs).

I dropped from my 4x8 to a 2x8 for the extra boost; four high density dimms is legit.

 

[Mad-Matt]

This Motherboard only allows memory voltage changes in 0.5 increments and to get this OC stable requires me setting 1.45v although it should require a little less according to the dram calculator. Its in a closed case, SilverStone ML09 which doesn't have much ventilation. If heat wasn't an issue maybe I could stabilize ddr3600 with 1.5v. CPU looks lower than it should be probably due to Spread Spectrum which is not an option to disable in this boards bios :/ I only enable PBO for benchmarks and not used for day to day use.

**************************************************************
CPU..........................AMD Ryzen 2600X @Stock+PBO
COOLING................Cryorig C7 lowprofile
MEMORY.................16GB (8GB x 2) teamgroup Xtreeme 4000 @3466 (16-14-14-28-1T)
MOTHERBOARD…..Asrock X470 ITX

 

[Rock72]

Thanks for your sub... but to add you to the Points competition leaderboard I'm going to need the remaining data sets from AIDA64. Please install the full version if possible...

@Rock72
I need memory quality and make/model

if at all possible

FYI the leaderboards are updated but I cannot post them here yet. The issue is I can't seem to edit my initial post with updated leaderboards, so I have a request in with the mods to help resolve this.

 

[R2DSF]

MoBo: Asus C8 Impact
CPU: Ryzen 7 2700
Cooler: NH-U12A
RAM: G.Skill Trident Z Black and White F4-3600C16-16GTZKW

 

[storm-chaser]

Nothing special except I'm rocking four 16GB modules. Ran Memtest86 8.3 Pro for three hours and it reported no errors (I'm quite sure many people in this thread cannot boast the same and then they get surprised when their system suddenly BSODs).

Thank you for your contribution. But... it seems you are missing the point.

First of all, this is a 1 minute benchmark competition not a 3 hour memtest86 marathon. I never said anything about reliability. Matter of fact, I set no rules, it's basically an unlimited class. Which means, any and all methods are valid to get that number lower (short of photoshop, lol). It's implied that at some point you will end up trading reliability for power. And that goes without saying for overclocking in general terms as well. Why on God's green earth would we sacrifice reliability for performance? See last paragraph.

My assessment is that if you can run the AIDA64 Cache and Memory Benchmark without the system crashing, you've passed the only rule in the book.

In drag racing, this is called "turning up the wick" where you push the engine beyond it's design limits to stay competitive. Sure, it's more at risk for catastrophic failure, But remember, the engine only needs to hold together for 8-10 seconds until the run is complete. If you have a good run and you don't blow it up, you just might get a warm, happy feeling knowing you've laid down some impressive #s and left the guy sitting next to you in a cloud of dust.

That's my take on it anyway. This is a performance oriented benchmark and it's all about pushing the limits. You can never have enough horse power

 

[Phenomenum]

CPU..........................AMD Ryzen 5 1600
COOLING................Stock (AMD Wraith Spire)
MEMORY.................16 Gb (8x 2) HyperX Fury Black 2400 HX424C15FB3
MOTHERBOARD…..MSI B450 PRO-VDH PLUS

 

[Arctucas]

9900K
Custom loop
F4-4500C19-8GTZKKE (2X8)
Z390 Dark

 

[storm-chaser]

I notice all the Intel systems in this comp are running 2T command rate. What is the deal? Shouldn't we be able to see 1T on the latest technology? AMD certainly has no problem with it, at least according to this benchmark.

9900K
Custom loop
F4-4500C19-8GTZKKE (2X8)
Z390 Dark

This is about as sick as it gets fellas! over 5000MHz on both NB and CPU!

Now you just need to go for the trifecta and you'll no doubt be godlike.

The trifecta being:
5.0GHz CPU (accomplished)
5.0GHz NB (accomplished)
5.0GHz Memory (good chance you can hit it)

I have my work cut out for me in achieving this goal, while you, on the other hand are almost there...

My 9600KF build will start next week. So I will be sure to get those results updated once I get it dialed in.

 

[phanbuey]

I notice all the Intel systems in this comp are running 2T command rate. What is the deal? Shouldn't we be able to see 1T on the latest technology? AMD certainly has no problem with it, at least according to this benchmark.



This is about as sick as it gets fellas! over 5000MHz on both NB and CPU!

Now you just need to go for the trifecta and you'll no doubt be godlike.

The trifecta being:
5.0GHz CPU (accomplished)
5.0GHz NB (accomplished)
5.0GHz Memory (good chance you can hit it)

I have my work cut out for me in achieving this goal, while you, on the other hand are almost there...

My 9600KF build will start next week. So I will be sure to get those results updated once I get it dialed in.

1T vs 2T on intel doesn't make a difference in most cases, and in some it actually hurts your performance. My ram latency increases from 39.3 ns to 42ns going from 2 to 1t - mostly because i think the Intel boards will loosen subtimings to get the memory to run a 1T rate. 2T just allows you to get more aggressive with the overall speed and main timings which ultimately makes a much bigger difference than 1T.

How are you testing the Uncore stability? The issue I run into past 4.8 is that it 'looks' stable - a few hours of p95 stable at 50x, but it isn't. I actually don't know anyone's Uncore 24/7 stable @ 5Ghz - maybe the new chips are better (not that it makes a huge difference, the diff between 4.7 and 5.0 is imperceptible).

 

[Anatom]

CPU..........................i7-5960X 4.7Ghz
COOLING................Alphacool Eisbaer 280 CPU AIO
MEMORY.................32GB (8GB x 4) G.Skill Ripjaws V F4-3200C16D-16GVKB(X2)
MOTHERBOARD…..Asrock X99M Killer/3.1

 

[phanbuey]

so this is interesting... i just bumped it to 5.1ghz for kicks, and it's actually passing a stability test (where before it wouldn't even start, this is a 3 year old rig) I wonder if the mitigations for spectre/meltdown changed the rules for cache.

 

[storm-chaser]

One interesting pattern to note when looking at our leaderboard is that Intel is apparently lightyears ahead of AMD in terms of memory latency. Matter of fact, the best Ryzen rig in the comp has roughly TWICE the latency of the best Intel system (32.8 ns vs 60.0 ns)

Not sure if this is going to have a noticeable impact on real world performance, but what exactly is the deal here?

It's odd that I can pull 49 ns on my dated Phenom II rig, yet it still has a -10 ns advantage on the best Ryzen chip in the comp. Granted, I do have the memory controller overclocked to 3.0GHz, but still, you'd think the latest Ryzen chips should be able to compete with Intel in latencies.

 

[phanbuey]

One interesting pattern to note when looking at our leaderboard is that Intel is apparently lightyears ahead of AMD in terms of memory latency. Matter of fact, the best Ryzen rig in the comp has roughly TWICE the latency of the best Intel system (32.8 ns)

Not sure if this is going to have a noticeable impact on real world performance, but what exactly is the deal here?

It's odd that I can pull 49 ns on my dated Phenom II rig, yet it still has a -10 ns advantage on the best Ryzen chip in the comp. Granted, I do have the memory controller overclocked to 3.0GHz, but still, you'd think the latest Ryzen chips should be able to compete with Intel in latencies.

It's a design that is very good for latency, but not so good for lots of cores - i think Zen 3 will beat them with mid 50's latency but then backed up by higher IF and core clocks as well as the customary crap-ton of cache that amd throws in these days.

Unless the whiskey-lake rumors are true, in which case keller and teams have worked their magic.

 

[ultrafx]

Spoiler: Pentium G4560

CPU..........................Pentium G4560 @ 3.5Ghz
COOLING................be quiet! pure rock slim
MEMORY.................8GB (4GB x 2) G.Skill Ripjaws V F4-3200C16D-8GVKB
MOTHERBOARD…..Asus Maximus VIII Hero

Spoiler: Ryzen 7 2700X

CPU..........................Ryzen 7 2700X @ 4.3Ghz
COOLING................be quiet! pure rock slim
MEMORY.................8GB (4GB x 2) G.Skill Ripjaws V F4-3200C16D-8GVKB
MOTHERBOARD…..Asus ROG Strix B350-F Gaming

Spoiler: Xeon E3-1220 v3

CPU..........................Xeon E3-1220 v3 @ 3.5Ghz
COOLING................be quiet! pure rock slim
MEMORY.................8GB (4GB x 2) Kingston HyperX Fury HX316C10FWK2/8
MOTHERBOARD…..ASRock H81 Pro BTC R2.0

There is one mistake on your leaderboard, there is Quad channel, not Dual channel.