Share your AIDA 64 cache and memory benchmark here

[freeagent]

It's called not being stable, but stable enough to run a 2 minute bench.

That's what this would be


.

 

[PaulieG]

I'm liking these Micron Rev E. This is bench stable, but fails 25 minutes into OCCT memory stability test. 4 sticks.

 

[storm-chaser]

NF4 chipset was good for latency. I saw in the high 20s running cas 2. That’s why that pesky X2 Black has been at the top of the list for years lol.

LOL I dethroned the 6400+ with.... you guessed it, another 6400+!!

Here is something from the history books... This is going back a long way... when AIDA64 was known as EVEREST. I had a lot of fun with this chip and as you can see it's no slouch. Pretty much every enthusiast needs to have this chip on their resume if they want to be taken seriously. lol j/k only half serious. This is bone stock.

The old timers in the nursing home...

 

[brandon7171]

This AIDA was Done Yesterday

This Aida was done today, they bother good but slightly different

This is awesome im pretty impressed with it a BIG THANKS to @Zach_01

 

[storm-chaser]

Ryzen 5 3600 overclocked to 4.5 GHz

View attachment 189671

What is the deal with Ryzen and low write speeds? I noticed in flipping through the last few pages, some of the Ryzen chips "dip" really low in the write speed test. And then some are fine... Is this just a random coincidence or does Ryzen really have difficulties "performing" here? I mean we are talking about a difference of 15GB/s or more, which is not insignificant.

BTW coaxed a bit more out of my Patriot Viper Steel 16GB DDR4 kit....

 

[Zach_01]

What is the deal with Ryzen and low write speeds? I noticed in flipping through the last few pages, some of the Ryzen chips "dip" really low in the write speed test. And then some are fine... Is this just a random coincidence or does Ryzen really have difficulties "performing" here? I mean we are talking about a difference of 15GB/s or more, which is not insignificant.

BTW coaxed a bit more out of my Patriot Viper Steel 16GB DDR4 kit....

View attachment 191966

It’s “normal”. Single chiplet Ryzens are maxing at 30GB/s and dual chiplets double that. It is a deliberate cut per chiplet to give the other (copy, read) the resources they need for max performance. AMD engineers made the call as today’s workloads doesn’t have too many write requests.

Im not saying you do, but you (or any other) shouldn’t comparing results from completely different CPU architectures.

 

[storm-chaser]

It’s “normal”. Single chiplet Ryzens are maxing at 30GB/s and dual chiplets double that. It is a deliberate cut per chiplet to give the other (copy, read) the resources they need for max performance. AMD engineers made the call as today’s workloads doesn’t have too many write requests.

Im not saying you do, but you (or any other) shouldn’t comparing results from completely different CPU architectures.

That 15GB/s deviation I'm referring too was based only on the Ryzen architecture, when you take into account read and copy speeds, the write speeds sometimes have a deficit of 15GB/s or more, not comparing to Intel or anything like that.

Is it true that Ryzen uses an FX based memory controller? If true, why would they do this? Everyone knows how lackluster FX was.

And I'm not bringing these things up to bash AMD, just curious.

 

[Zach_01]

That 15GB/s deviation I'm referring too was based only on the Ryzen architecture, when you take into account read and copy speeds, the write speeds sometimes have a deficit of 15GB/s or more, not comparing to Intel or anything like that.

Is it true that Ryzen uses an FX based memory controller? If true, why would they do this? Everyone knows how lackluster FX was.

And I'm not bringing these things up to bash AMD, just curious.

Understood
And I answered what is the deviation about. It’s 1 chiplet vs 2. And apparently this deviation doesn’t have any significant play in performance, as we saw on benchmarks, other than pure AIDA64 scores which often do not tell the truth about a CPU’s performance. In a lot of cases the single chiplet CPUs outperform the dual ones in singe or reduced threaded workloads, like gaming.
I know this is not a debate, I’m just stating facts here.

About the UMC (UnifiedMemoryController) I have no info. Didn’t made any inquiry about it either, because it doesn’t matter to me anyway if it’s an FX carry over or some version of it, or a completely new design.

 

[tabascosauz]

That 15GB/s deviation I'm referring too was based only on the Ryzen architecture, when you take into account read and copy speeds, the write speeds sometimes have a deficit of 15GB/s or more, not comparing to Intel or anything like that.

Is it true that Ryzen uses an FX based memory controller? If true, why would they do this? Everyone knows how lackluster FX was.

And I'm not bringing these things up to bash AMD, just curious.

Memory controller =! Infinity Fabric. Most of Ryzen's problems come from Infinity Fabric. Halved write bandwidth for 1CCD SKUs, relative lower read/write/copy on chiplet SKUs, inability to sustain 1:1 past 1800-2300MHz, high DRAM latency, etc.

Aside from the obvious differences from DDR3 to DDR4, the UMC is most definitely not the the FX controller, because Ryzen 3000/5000 UMC by itself is competitive on achievable speeds with Comet Lake, and the Ryzen 4000 UMC is just about the best DDR4 controller on the market by far and it's unlikely Rocket Lake will change that. But unfortunately, Infinity Fabric is a thing.

 

[freeagent]

The memory.. honestly its not a big deal. When I ran this thing for the first time with Zen 2 the ram was at 2133 cl21 with a bunch of 20s lol.. I ran it like that for half a day before I noticed and it honestly felt just fine lol. I thought I had set DOCP but I didn't.. I wish I could compare it to a modern Intel but I cant, but I can compare it to older Intel and this thing is fast af out of the box. Its ridiculous.

 

[DR4G00N]

Here's one from a while ago, 4x4GB Quad-Channel FB-DIMM.

 

[nguyen]

24/7 settings on my ancient Gigabyte Z370 Gaming 5 motherboard

Pushing for higher Mem clocks than 3600mhz seems pointless.

 

[FilipM]

That's what this would be


.View attachment 191048

I get similar latency at 3933 & CL15

 

[jjnissanpatfan]

Is there a reason no one puts the vdimm voltage on there?

 

[freeagent]

View attachment 192544

I get similar latency at 3933 & CL15

You are running an all core OC, I was not.

 

[brandon7171]

Is there a reason no one puts the vdimm voltage on there?
View attachment 192556

My VDIMM IS N/A not sure how to enable it.

 

[FilipM]

You are running an all core OC, I was not.

Im on PBO, 150MHz offset

Curve negative 24

112 76 111 on the limits


PS. The pic is 1000% Memtest86 stable and several hours Aida 64 Memtest. VDIMM 1.45v

 

[tabascosauz]

Is there a reason no one puts the vdimm voltage on there?
View attachment 192556

Zentimings can only show what the SuperIO shows. Most Asus boards do not have a end user-visible VDIMM value.

Not to mention it's generally useless unless it's actually accurate. On Gigabyte boards ZT just displays the same VDIMM as Ryzen Master, ie. reading the value you set in BIOS. On most Gigabyte boards the actual VDIMM reading in HWInfo suggests that it's overvolted 0.03-0.06V over what you set in BIOS, so it's not like ZenTimings VDIMM would be useful to know.

 

[jjnissanpatfan]

Zentimings can only show what the SuperIO shows. Most Asus boards do not have a end user-visible VDIMM value.

Not to mention it's generally useless unless it's actually accurate. On Gigabyte boards ZT just displays the same VDIMM as Ryzen Master, ie. reading the value you set in BIOS. On most Gigabyte boards the actual VDIMM reading in HWInfo suggests that it's overvolted 0.03-0.06V over what you set in BIOS, so it's not like ZenTimings VDIMM would be useful to know.

New to this socket. And the MSI boards seemed to do okay with the vdimm voltage. Thank you for helping me understand.

 

[freeagent]

View attachment 192544

I get similar latency at 3933 & CL15

The whole point to that post was to show what running on the verge of instability was. It streams errors in memtest at that speed

Ok, I screwed up on my settings, this like the second time I tried to run C15. But I tried to match your latency..

 

[Det0x]

Like i wrote in a other thread some days ago, i have changed back to Agesa 1.1.8.0 because both boost clocks and latencies are better on older bios without training wheels.

After spending a few days finetuning i have now found my new 24/7 memory settings (can link better results suicide-run settings) which iam pretty happy with, considering i'm running 4x8GB sticks and my 5950x wont run WHEA-free above 1900/3800

No WHEA-errors on any of the runs (compared to others linking without proof/unstable settings)

Included some numbers for MLC for those interested.
(Also have to say its much easier to get low latency with single CCD CPUs like 5600x or 5800x)

Both CCD's enabled (5900x/5950x)

Stability testing in my bloaty windows install (lost 0.2ns compared to safemode)
Could have run 50 cycles or more with 1usmus cfg as these settings are 24/7 safe settings and dont fail.

Think this is also the fastest Zen3 L3 latency i have seen, all thanks to the much better singlecore boosting in agesa 1.1.8.0

 

[CandymanGR]

 

[FilipM]

The whole point to that post was to show what running on the verge of instability was. It streams errors in memtest at that speed

Ok, I screwed up on my settings, this like the second time I tried to run C15. But I tried to match your latency..

View attachment 192571

Now I need to re run mine

Some background stuff in Windows is eating my latency and I never found what it is. Might need a clean install

CL15 worked great for me so far. Soon I plan on upgrading to 4x8 of the same set.. I know its more difficult than 2x16 but hey, challenge accepted

 

[dgtech19]

Something wrong with L3 cache 29..9ns ??? anybody know why.
my spec in Zentimings.

thanks.

 

[freeagent]

Now I need to re run mine

Some background stuff in Windows is eating my latency and I never found what it is. Might need a clean install

CL15 worked great for me so far. Soon I plan on upgrading to 4x8 of the same set.. I know its more difficult than 2x16 but hey, challenge accepted

4x8 on my 3600XT is really easy. I can run 1900/1900 with no problems at all, really easy to do. But 4x8 on this CPU is pretty tough. I am not sure why. I would blame the IMC, but I can bench on it a 2100/2100 so idk.. This one here I can do 1833/1833 with four sticks. Sometimes I can get it up to 1900/1900 but it was only a couple of times and it wasn't even remotely stable. Not sure if it hardware or software, leaning towards software..