I'm sure you guys in boxes have been absolutely exploding.

Well, the AMG rise in and radio on step today.

We'll talk about something extremely important, as important as it could possibly get when it comes to Andy Risen.

And that is the fact that we actually changed out the extension cord with its unique radiator mounted pumped premium sleeve tubing, Infinity mirror, RGB block and cam software control the crack and M 22 all in one liquid cooler from NZX.

T is the perfect choice for small form factor liquid cooling.

To learn more, click the link in the description below.

All right, so we're gonna be doing some over clocking here today with the 3900 ex, um, interesting discussion to take place here because if you've looked at the reviews and we said in our review that we were gonna be doing a separate video regarding over clocking, but you probably already seen by now that there is not much over clocking headroom.

So we're gonna do today is we're gonna talk about some of the settings here.

They're pretty much the same as previous rise in we're gonna use our crosshair hero board is an example.

We're gonna go through some of settings what they mean.

Then we're gonna show you the max over clocks we can achieve on our 3900 X.

It should be no surprise where it's gonna land.

It's pretty consistent across the border where people are ending up.

We're talking about how to sort of prepare for over clocking and what to expect, whether or not it even is worth it.

This time around, it comes to Andy.

Now, obviously the 3900 X being the top tier at least until later this year, with a 39 50 x comes out more course, this is a small process seven centimeter with 12 course, 24 threads on the mainstream sized.

I guess package will call it because there's more than one dying there, so it's a lot of focused heat.

So we've ditched the prison cooler and obviously went with a course there h 100 I a cooler.

But what you're gonna see is that things still get pretty warm, so you obviously want to make sure that you're gonna be having cooling, taking into consideration with your over clocking because we're not in a chassis here.

Airflow over the V.

R.

Ems is also very important.

So I just said 120 millimeter fan here to blow directly on the power delivery.

If you've got a chassis, plenty of airflow.

You wouldn't have to do this, But obviously, for this demonstration, we're doing direct airflow.

Travis car doesn't really matter.

That's completely independent.

We just have a 2060 sitting on there.

We're really gonna be doing any gaming today.

Just a little bit.

We're looking at basically just max performance.

We can get out of CPU instructions, and then we have a 1200 watt power supply here.

This is overkill for this.

What you want to make sure you have a good quality power supply that you have nice, beefy 12 rails in there and that you have more, uh, obviously output.

Then you need.

So I would still recommend something like a 700 or even 100 watt power supply for this.

Because then you could stay in that nice efficiency curve.

The other thing you obviously need is a motherboard with over clocking chipset.

Usually the 70 skew is gonna have all the bells and whistles and step that you want with beefy V R.

EMS, beefy chipset coolers, all that sort of stuff so that you don't run out of power delivery because that more often than not when it comes to over, clocking risin ends up to being a point of failure because people are trying to over clock these high core count mainstream CPS from Andy with mother boards that are not exactly robust enough to handle that.

So, like I said, we're using the crosshair hero motherboard.

But here's some of the things you need to keep in mind, and this is gonna hold true for any mother would manufacturer.

I've never been a part of any Siri's lunch, where the BIOS didn't go through some sort of a maturing process where some of the stuff, I'm telling you, it's just completely caddy, Warm Piss and the White Caddy one.

It's a gray hair thing.

You don't have gray hair.

Don't use that term.

That's Archer.

But what I want to show here is, even though we're on the latest bios that we could basically get our hands on 7702 there.

There is a lot of weirdness that's still going out by us.

You early adopters.

You get to deal with this.

And guess what?

We're all bug testers.

Okay?

So keep that in mind if you bought rising 3000 in terms of the motherboards, there's some stuff that you need to be aware of.

What I'm gonna show you right now is even though we're set to defaults, this is the way it ships.

This is out of the box.

You plug in your CPU.

You turned it on, you touch nothing.

This is the way it is.

As of today, with this bios, RCP was sitting at 1.473 volts.

I can tell you right now, that's the reason why even with an H 100 I were sitting at 40.

See a title that's way too high, because that is a about ah 100 or 200 mill A volts or so actually higher than we needed to be for stock settings that there's There's zero excuse for that, Ace.

You should have never let this fly.

They'll probably come back and say all the reasons why this is fine.

And I'm telling you that my opinion?

It's not fine.

So the reason why it's not fine is we started hitting over temperature shut down on the CP because of it.

That's not fine.

And when that happens within a Iot, there's nothing you could say to excuse that same thing with the voltage on the memory, we're running 2133 based megahertz.

We did that for a testing.

For a reason will show you why.

In this video, 1.488 volts is higher than D.

O.

C.

P or Ex MP settings.

If you guys underwear.

DCP is the AMG version of Ex MP, which is basically where it tightens up the timings and speeds up the memory.

It's the actual printed numbers on the modules.

That's why that's different than what's printed on the module.

That's way too high.

That should be 1.35 or 1.2 for stock speed.

So again, we're over bolting the memory unnecessarily.

Um, the other thing Here, check this out.

Eso si voltage.

This is the voltage for the fabric.

The interconnect between the chip lets on the actual CPU substrate, so that should actually be more like 1.15 ish.

That's 1.32 Have no idea why it's so high.

In fact, even according to Der Bauer doing Ellen to over clocking, he said don't exceed, like, 1.28 to 1.3 of the most.

We're at 1.32 out of the box.

So that's a little this a little concerning.

So I'm gonna manually set the CPI right now to one point is go on 0.3.

Okay.

And then our associate old is we're gonna set that to again manual one point, too.

Okay, dear And voltage again.

One point too.

Well, I'm playing 35 We'll just go the deal, see people to to know we're gonna move it.

And then the c l deal the DDG voltage.

This is another piece of fabric.

So these two voltages, the S O.

C and the d d d g are tied together.

The associates, the input input voltage for the B d d g.

That is to be at least ah 100 mila volts below this stock, for that is 0.95 volts or one bolt.

So I'm sending it to one bolt.

I have no idea what it's sitting at now.

But based on what I'm seeing here in the PLL, one point a voltage sitting at 2.1 obviously is too high.

So I have no reason to believe that the V d d g wouldn't be over bolted, but, hey, they got the Southbridge, right?

So these there one for, like, 12.

So anyway, now that we've gone in and UNB or CTE are bolted settings that CPU or the motherboard did on its own, Now we're ready to actually over clock.

In fact, now we're ready to actually just show you the benchmarks and we'll talk about the over clocks, whether or not they're even worth it.

So why don't we just go ahead and show you this chart?

I guess that's why you're here.

Hope if you're just here to toe look at my pretty face Well, you know, his instagram twitter facebook kind of dead.

I've been uploaded there in, like, three years.

All right, so after spending the last couple of days because, like I said, we did some more work locking yesterday, actually a lot of other clucking yesterday and then went back through and redid all of our testing today for sanity checks and even added a couple of tests this that we weren't planning on.

So here's the deal.

The 3900 X and admittedly, this is only the 3900 X.

The behavior on 3700 X and the other excuse might be different.

You can look at all the reviews that have made it out in the market, both written and video format, and you'll find that 4 to 43 is a sweet spot.

Very few people hit four for if they did take a lot of votes to get there, Some people got complete duds and lost a silicon lottery where they couldnt even get for one.

And so it's one of the things that it's kind of interesting because the way that the performance boost that's built into the CPU is depending on the workload, it's gonna boost higher or lower.

You could see as high as 4.6 gigahertz on single core, and that's reflected when we do something like AR 15 or 20 or something, where we can physically tell it.

Onley use one court.

The problem is, with so much software now being at least multi threaded to the point where he could utilize four cores arm.

Or even though we've got 24 threads sitting here, what starts to happen is you don't see one or two cores bouncing at 4.6.

You see them dropping down to more like the 4.34 point two range.

What we saw in almost every single one of our tests, whether it be premier or it be like Blender.

We even did a far cry five test because we know that's a multi threaded game.

But we also know it's a bit of a CP bottleneck for a lot of different reasons.

But what we saw was we sat more so around the 4 to 4.2 to five range, and that's only about 75 megahertz ish because it would down slightly lower than our max manual O.

C.

Which is 4.3 on all course.

So the takeaway from that is that the 4.6 looks really good on paper, but depending what you're doing, you're not going to see that that often, even in games, even though there were threads sitting completely idle with far cry five at sitting at 0% utilized.

We still were seeing about 42242 to 5 on those cores bouncing around and sometimes Lois, 41415 It bounces all around.

It's doing what it's supposed to do.

But you would think, because most people go games are single threat optimization.

That's not entirely true these days, because most games that have been developed in the last four or five years are using multi threading now, and as soon as you go past one core being loaded, you're not gonna see that 46 You're not going to see a single core bouncing round 46 And the reason for that is because the CC exes have to act in a grouping.

Each CCX is independent of the other.

However, you cannot differentiate course speed between different course on the same CCX.

That's why you see them kind of moving around as a group.

So as soon as they're beyond that one core boost, you're not going to see the 46 But your behavior is only one side of the coin.

You have to take memory into account, too.

You can't directly compare intel A name D, though like frequency wise, the same as with G pews between named Ian and Video.

You can't just compare them for frequency speed either, because they're two completely different architectures.

What you're dealing with here in rise in and this is no different.

This has been the case all along, but what you're seeing is the maturity of their chip lit kind of infrastructure here, where you've got two different cc exes with half the amount of course and each plus s empties on both sides.

So six and 12 on either side, so one could argue it's +26 Court 12 thread Si, pues cut it.

I think intel use the joke of glued together Well, that's called infinity fabric, and you have control over that as well.

And the speed between the communication or the infinity fabric is known as the F clock on here, which we showed.

So some of the testing methodology included not only over clocking the memory and then tightening up the timings and playing around that or just let e d o c p.

Do it.

And then we over clock the memory as well as the fabric and what we saw, if you look at those charts, is when it comes to productivity, depending on the title defense determined whether or not we saw any improvement at all.

So that explains why, in premiere we saw a minimal difference between just over clocking the memory, leaving the the CPU speed the same.

And then we saw a big improvement by leaving the over clock memory in place.

But then increasing the fabric or the speed between the cc excess.

And then we saw again by over, clocking a 4.3 very minimal two second difference because again, it's only about a 75 megahertz over clock versus where it was losing two on its own.

But when it comes to games on the other hand, far cry five.

We chose this title specifically.

Now we were very aware this is only one title.

We want to do a whole another piece of content where we go through and test all of our games with our over clocks and all that sort of stuff to see if we see the type of improvements.

We saw him far cry five across the board, but we chose far cry five because we know for a fact it's very CP bound.

We know that the CPU is definitely holding back GP use even as far down as like the 2060.

And so that's indicative of the fact that it's only about 65 between 65 50% utilized on our 28 e t I.

That's over clock.

So for the testing, we did 10 80 p normal settings because we're trying to get the f ps hi to put more of a strain on the CPU.

Now, if we had gone 14 40 up the eye candy would have taken a lot of that load up the CPU, and it would have been indicative of whether or not the over clocking improved the gaming performance because we had to induce a bottleneck.