TechPowerUp RAM Latency Calculator Feedback

[Chrispy_]

(copied from your link to TPU's scaling article)

Proof that there's barely even margin of error between 6000 and 8000, so why waste money on faster RAM when it doesn't achieve anything?
That's what sweet spot means - the last point at which spending more money nets you any real gains.

 

[ShrimpBrime]

Proof that there's barely even margin of error between 6000 and 8000, so why waste money on faster RAM when it doesn't achieve anything?
That's what sweet spot means - the last point at which spending more money nets you any real gains.

I love your system specs. cooling: No it's a 2.2w atom. "braggin rights" XD, thats pimp.

 

[Chrispy_]

I love your system specs. cooling: No it's a 2.2w atom. "braggin rights" XD, thats pimp.

Ah yeah, I'm basically an SI - I can have almost any hardware I want, at any time, provided it's available from distributors/resellers I have accounts with and it's relevant to my industry (CAD, rendering, realtime-3D, VR) so listing my current rigs would be a never-ending flex that would paint me as some boastful tw*t and I'm all about making do with what is sensible rather than throwing all of your income at Nvidia, AMD, or Intel just to stroke one's ego.

"Bragging Rights" is currently inactive, poor little bugger got put in a storage box when I dismantled my mining rigs and sold off all my GPUs. I should probably fish it out and see if it runs W11 24H2 for shits and giggles, or donate it to one of the schools we send old workstations and server hardware to.

 

[ShrimpBrime]

Ah yeah, I'm basically an SI - I can have almost any hardware I want, at any time, provided it's available from distributors/resellers I have accounts with and it's relevant to my industry (CAD, rendering, realtime-3D, VR) so listing my current rigs would be a never-ending flex that would paint me as some boastful tw*t and I'm all about making do with what is sensible rather than throwing all of your income at Nvidia, AMD, or Intel just to stroke one's ego.

Little bit of everything is nice. I support a 5 famo and pretty much can have w/e hardware I'd like as well. It's not necessary. So I stick with 1 rig. I bench it and just use it simultaneously. I really should invest in an SS or cascade unit, but really the returns aren't worth all the energy to keep cold for as much as the rig runs.

 

[Metroid]

Proof that there's barely even margin of error between 6000 and 8000, so why waste money on faster RAM when it doesn't achieve anything?
That's what sweet spot means - the last point at which spending more money nets you any real gains.

My post said that the sweet spot for Zen 5 is 6400, I did not say to buy anything higher than 6400.

 

[Chrispy_]

My post said that the sweet spot for Zen 5 is 6400, I did not say to buy anything higher than 6400.

W1zzard posted official confirmation from AMD that 6000 is the sweet spot, not 6400.
The link you posted (also W1zzard's work) proves that 6000 CL28 is faster than any 6400 configuration.

Why, bro? Why are you trying to defend what is presumably an honest mistake? There's no shame in owning it!

Don't get me wrong, if 6400 was worth buying I'd be all over it, but the reality is that 6400 is more expensive, less likely to be stable, and technically slower, even if we're only talking half a percent which is margin-of-error or run-to-run variance. The only advantage 6400 has over 6000 is "oMg LoOk, NuMbAr Is BiGgAr LoL", and that's not worth paying money or time for... :\

 

[Ruru]

"you should always use MT/s when talking about computer memory" - even if this calculator is of no use for LPDDR and GDDR, it would be better to say "computer memory on DIMMs" or something, because MT/s only really applies to those.

I always say "effective MHz", no matter what others say.

 

[Metroid]

W1zzard posted official confirmation from AMD that 6000 is the sweet spot, not 6400.
The link you posted (also W1zzard's work) proves that 6000 CL28 is faster than any 6400 configuration.

Why, bro? Why are you trying to defend what is presumably an honest mistake? There's no shame in owning it!

Don't get me wrong, if 6400 was worth buying I'd be all over it, but the reality is that 6400 is more expensive, less likely to be stable, and technically slower, even if we're only talking half a percent which is margin-of-error or run-to-run variance. The only advantage 6400 has over 6000 is "oMg LoOk, NuMbAr Is BiGgAr LoL", and that's not worth paying money or time for... :\

Better you take a look on that review posted by wizzard, in 90% of tests it shows 6400 cl32 is better than 6000 cl28 on zen 5 and it clearly says that 6400 is 1:1, so that means is the sweet spot.

 

[Chrispy_]

Better you take a look on that review posted by wizzard, in 90% of tests it shows 6400 cl32 is better than 6000 cl28 on zen 5 and it clearly says that 6400 is 1:1, so that means is the sweet spot.

Yeah, Wizzard's article you linked where it's a wash? Gaming wins at lower resolutions where the GPU isn't the bottleneck being measured.

720p gaming is faster on 6000

1080p gaming is faster on 6000

Application performance is less than 2% slower overall, less than 1% slower if you boost SoC voltage to the same 1.45V in the interest of a fair test against the overclocked IMC for 6400 1:1. Speaking of that 1:1 label, that's not there to indicate the sweet spot, it's there to indicate that result isn't using the default/auto dividers - because that would be 1:2. It doesn't need stating for any of the other results because they'll be using their default dividers of 1:1 at 6000 and below, 1:2 at 6000 and above. Let's make this very clear, 6400 is overclocking the IMC and your mileage/stability/success may vary. The reason AMD set the default divider to drop down to 1:2 above 6000 is because they aren't confident all their memory controllers can actually manage it.

If there's a specific application you need that has a massive bandwidth requirement - then sure, it might be worth spending the extra and hoping that you got dealt a lucky sample in the silicon lottery that can actually run the IMC at 3200MHz - but for most people running a range of tasks there's basically no significant performance on tap for your extra outlay, and it might even be slower in your favourite apps and/or games.

 

[Count von Schwalbe]

Remain on topic please folks.

 

[W1zzard]

Switched to v2 as the default version, added links to switch between v1 and v2

Who likes to have the table in v2 and why?

 

[Count von Schwalbe]

V2 formatting has an issue on mobile; the input boxes are off the side of the page. V1 is ok.

 

[W1zzard]

V2 formatting has an issue on mobile; the input boxes are off the side of the page. V1 is ok.

Yeah, with the four fields I wasn't sure if/how to wrap or turn them into some kind of side-by-side column format somehow

https://imgur.com/Wb9Fscw

wrapping groups of two looks a bit confusing

 

[Chrispy_]

Yeah, with the four fields I wasn't sure if/how to wrap or turn them into some kind of side-by-side column format somehow

https://imgur.com/Wb9Fscw

wrapping groups of two looks a bit confusing

All looks good to me.
Can you choose number of columns based on window size, to sort mobile out (but also allow 4 column in landscape, perhaps?)

 

[W1zzard]

Can you choose number of columns based on window size

Right now that is essentially what's happening. It wraps to fit. Technically I could use 4 columns when available, and force 1-column when not enough space. Not sure if better?

 

[Wirko]

How important is mobile access, really? We don't calculate latencies while our only desktop PC is in pieces, or do we?

 

[W1zzard]

How important is mobile access, really? We don't calculate latencies while our only desktop PC is in pieces, or do we?

Mobile has to be supported nowadays. I'd still lean toward making desktop the targeted AAA experience, but mobile shouldn't be miserable

 

[Count von Schwalbe]

Perhaps you could make latency and nanoseconds into columns instead of rows? (Only on mobile)

That way it would still be grouped by twos but sort out the aspect ratio.

 

[Chrispy_]

Right now that is essentially what's happening. It wraps to fit. Technically I could use 4 columns when available, and force 1-column when not enough space. Not sure if better?

Scroll bar? It's not exactly the prettiest option but functionality comes before aesthetics.

If you absolutely don't want a scroll bar, can you put each overall table in a second table and wrap only the outer table for columns that overspill so that you get something like this?

 

[A Computer Guy]

Maybe like this?

 

[W1zzard]

Maybe like this?

View attachment 370884

This seems to be ideal, but will be pretty hard to achieve with HTML/CSS

 

[W1zzard]

Improved mobile support is now file, it looks like @A Computer Guy suggested

 

[Wirko]

The step size is 100 MT/s (speed), 1 (latency cycles), 0.1 ns (latency in ns). That's probably how it was meant to be.

Here's the result if I click on +Compare a few times to get several groups of fields:

The step size in the last group is correct.
One group above, the step is 2x that, in all fields.
One more group above, the step is 3x that, in all fields. And so on. Browser is Firefox 137.

 

[W1zzard]

The step size is 100 MT/s (speed), 1 (latency cycles), 0.1 ns (latency in ns). That's probably how it was meant to be.

Here's the result if I click on +Compare a few times to get several groups of fields:

The step size in the last group is correct.
One group above, the step is 2x that, in all fields.
One more group above, the step is 3x that, in all fields. And so on. Browser is Firefox 137.

Nice find. Fixed