Nvidia gave us their internal tools… What could go wrong?

[Emily Young]

When Nvidia asked us to attend a presentation on something “not graphics related”, we had no idea how much we’d love what they were about to show us…

 

 

[eeeee1]

he didnt drop them did he?

[Abdullah Bhutta]

26 minutes ago, GabenJr said:

When Nvidia asked us to attend a presentation on something “not graphics related”, we had no idea how much we’d love what they were about to show us…

 

 

knowing Linus he drops something in every video accept for this one:-

https://www.youtube.com/watch?v=3t1PQJmM8P4

[Guest]

nice

[Results45]

Imagine if Nvidia sold these in 2-in-1 packs and they sold out before RTX 3090 pre-orders.

[HTMLmoron]

48 minutes ago, GabenJr said:

When Nvidia asked us to attend a presentation on something “not graphics related”, we had no idea how much we’d love what they were about to show us…

 

 

What an absolute gamer move by Nvidia. They must know you're going to be testing the 3000s as well as the AMD Navis. That's a ton of faith they have in their products. This is incredible marketing. 

[Senzelian]

Some smart person should backwards engineer these things and sell it.

[Velcade]

Why balance the GPU on a box when  you could turn the case on its side?  

 

[xnamkcor]

Power usage of the Power Cleaner?

[Nystemy]

As someone working with electronics.

The statement that "Nvidia claims it is within .25% with their oscilloscopes as far as accuracy goes." seems like you didn't really talk to an engineer, or talked to an engineer that oversimplified things or intentionally lead you astray, or they just didn't even mention anything about an oscilloscope at all.

What one would do for accuracy checks on a device like this is to first look at DC accuracy.
And then its flatness over its measurement range. (By sweeping a single tone with a known amplitude and impedance, though one can just design it with good parts and thereby know that it will be flat within X amount of bandwidth, unless one screwed up the PCB design by a lot. _{(Changing the current shunt for a 50 ohm resistor would honestly be sufficient down at the low frequencies we would be interested in here.)})

Non of that would use an oscilloscope.

Oscilloscopes are primarily designed to do 1 thing, and that is to measure a signal over time.

Most scopes tends to not be within even 3% accurate when it comes to voltage. While most cheep multimeters are 1% or better. (An an actually good multimeter should be better than 0.1%)

And measuring power with a multimeter would honestly work too. Downside is that most multimeters won't give you a nice graph _{(unless it's a Keithley with their 100+ K samples per second...)}. But even cheap multimeters tends to have rather flat signal response up to 10 KHz, some even goes to the "industry standard" of 20 MHz. _{(And there is also volt meters specifically made to measure larger bandwidths than this with exceptional flatness, going into the many GHz.)}

 

Though, a device like this power meter is frankly rather simple. And the latency meter seems like a nice weekend AVR (Aurdino) project. It's just a light sensor sampled by an ADC, and a counter getting triggered by a literal switch...

The power meter is also a simple weekend project, other than the "fancy" PCIe expansion board that one could just draw up in KiCad or the like and order from JLC or other PCB vendor for 10$. And from there, all one would mostly need is some current shunts, an op amp to amplify the singla a bit. Before likewise getting sampled by an ADC (that is within our microcontroller). Any inaccuracies as far as both amplification and current shunt values goes can be handled in software rather easily.

An AVR would be a nice fit with its multiple 12 bit ADCs (at up to 2 million samples per second), rather sizable internal memory, and also USB support would work rather nicely for both of these devices. _{(It's honestly something I wouldn't be remotely surprised if it is exactly what Nvidia uses... (And this isn't really even a new thing either, have seen plenty of projects like this over the years, not the latency one though. But tapping into a mouse button is rather trivial to be fair.))}

 

But, I should also point out that a device like this would still beat the accuracy of a mains power meter (like the kill a watt), mainly since we can eliminate most of the stuff that we aren't interested in. Thereby getting rid of a large portion of "unknown" power consumption within our system. _{(And yes, software readouts are iffy at best to be fair. Like even setting Vcore on a motherboard can vary by a noticeable amount from one board to the next, even from the same vendor and series...)}

 

But in the end, this device is nifty, but frankly old news, though nice to see someone make it into an off the shelf product, even though it honestly is really niche.

[soldier_ph]

Spoiler

Nvidia gave us their internal tools… What could go wrong?

Selling them on eBay ?

[SolarNova]

4 hours ago, GabenJr said:

When Nvidia asked us to attend a presentation on something “not graphics related”, we had no idea how much we’d love what they were about to show us…

 

 

I'd likely get both IF the price was reasonable, which to be blunt, Nvidia hasnt been doing well with lately.

As some one has already mentioned, its a simple tool, so it shouldnt be to expensive, but being Nvidia ...

 

Regardless, its a great tool for overclockers and those looking to tune their system for the lowest input latency.

 

[GionnyBanana]

If you are going to get this as a consumer, how are you going to cable manage that? XD

[Nystemy]

40 minutes ago, GionnyBanana said:

If you are going to get this as a consumer, how are you going to cable manage that? XD

To be fair.
The power measuring part is practically only useful for development of new cards. Though, when one develops a card one typically goes the more "bodge" way, since prototypes doesn't need to be practical. They just have to work sufficiently to test stuff of interest during development.

 

I can see LN2 overclockers maybe getting use of monitoring the PCIe power cables, but most overclockers kinda already do this with a current clamp that is accurate enough for this application. (After all, pouring LN2 is a mostly manual thing, _{unless one is King Pin and has a freaking machine for it...})
Not to mention that some overclockers just tap straight into the VRM controller itself, and get all the information they need... There is an actual connector for this on some motherboards and graphics cards.

 

In the end, a device like this isn't actually needed in the slightest for actual "normal" use.
A hardware reviewer like LTT or Gamers Nexus, or Bitwit, etc can have actual use for a device like this.
But for anyone just buying a card and putting it in a system, then this information is frankly nothing but a gimmick.

Latency testing on the other hand is a different story, that can have some actual use for the "normal" consumer. But for this, we do not need the power meter.

[YellowJersey]

Do vendors every send LTT something on the express condition that Linus not touch it?

 

 Maybe they should?

[WereCatf]

13 hours ago, Senzelian said:

Some smart person should backwards engineer these things and sell it.

What's there to reverse-engineer? Measuring voltage and current is easy: you just need a good ADC, a microcontroller and a current shunt resistor with a known value, measure the voltage-drop across the resistor and use that with the resistance-value to calculate current and thus you'll also know the power going through it. For the mouse-thingy, you basically need a microcontroller, a switch and a light-sensor.

 

The only thing that stands out here is the rather polished software, but even that's not hard to do.

[Senzelian]

4 minutes ago, WereCatf said:

What's there to reverse-engineer? Measuring voltage and current is easy: you just need a good ADC, a microcontroller and a current shunt resistor with a known value, measure the voltage-drop across the resistor and use that with the resistance-value to calculate current and thus you'll also know the power going through it. For the mouse-thingy, you basically need a microcontroller, a switch and a light-sensor.

 

The only thing that stands out here is the rather polished software, but even that's not hard to do.

Then do it!

[WereCatf]

2 minutes ago, Senzelian said:

Then do it!

I have no interest in measuring a GPU's power-draw.

[Senzelian]

3 minutes ago, WereCatf said:

I have no interest in measuring a GPU's power-draw.

But I assume you have an interest in money \^o^/

[Drama Lama]

Linus Drop Tips 

 

now it‘s official 

[WereCatf]

Just now, Senzelian said:

But I assume you have an interest in money (ω)

Actually, not much. I know I'm an odd duck, but money doesn't really work as a motivator for me for doing an uninspiring, lame project like that.

[Senzelian]

Just now, WereCatf said:

Actually, not much. I know I'm an odd duck, but money doesn't really work as a motivator for me for doing an uninspiring, lame project like that.

I'm disappointed. 

Hopefully someone is working on it. I bet that these would sell like hot cakes for $29 each or $50 in a bundle.

[X-System]

Nice tools but my Corsair HXi is enough for me to have the statistics by LINK software

[Senzelian]

4 hours ago, X-System said:

Nice tools but my Corsair HXi is enough for me to have the statistics by LINK software

Is that actually useable now?

I got the Corsair AX760i when it came out (the first PSU to feature Corsair link) and it was awful. The readings were all wrong, the software was very unresponsive and when it did work, you got basically no useful information out of it.

 

This was many years ago tho.

[X-System]

46 minutes ago, Senzelian said:

Is that actually useable now?

I got the Corsair AX760i when it came out (the first PSU to feature Corsair link) and it was awful. The readings were all wrong, the software was very unresponsive and when it did work, you got basically no useful information out of it.

 

This was many years ago tho.

I don't have problem with LINK software, it works perfect, stable and I got all informations. I bought my HX750i in January 2020.

 

On Corsair website, I see it's last version 4.9.9.3 for almost 2 years (01/28/2019). So I use this version (I downloaded it in 2020)

 

The current Corsair LINK looks like this (for RMi & HXi) :

 

 

And for AX1600i, it has more options like this :

 

 

For your AX760i, you would have like this :