*** The Official MSI GS65 Stealth Owners and Discussions Lounge ***

Nope, they are separate settings. Once you have lowered the AC / DC Loadline you'll know it very clear that they are completely different settings.

My initial goal was to get the lowest possible stable voltage, in the past I tried the same way you are doing now, moving both the core and cache together, but I noticed something, that whenever I have the CPU usage running at 100%, sometimes even moving the cursor could cause the voltage to bump for quite a few mV. And I discovered that, the undervolting on the core needs to undervolt more than the cache, a lot more, just to eliminate this voltage bumping situation.

 

Ha, noticed the same, just with the wattage output. So would you recommend the -160 -40 ish, as it says in the guide, or?

 

Core -160mV should do.

For Cache, the next level down is -37.1mV.

Always see the VID on the main screen of ThrottleStop when undervolting, that is the real number.

The voltage +- number (Only the voltage, the Speed Shift EPP at the bottom is still very useful to check if there is any EPP conflict between ThrottleStop and Windows Power Option) showed on the right side of the FIVR Windows is only to show the settings you have applied. And it doesn't mean it's working, it only tells you that you have set the value.

 

I have only run Ts Bench as a test, and i am getting around the 85C / 12 threads, 1024M / 3.9GHz (Stock repaste and replaced heatsink through RMA).
I am running around -180mV -43mV, and no errors or BSOD yet (i will experiment, to get it cranking furtgher)
I know this is probably not the best score, comparing it to the forums, but it has definetly improved my temps in gaming.
By clokcing the speeds down a bit, i get stable 70C in Forza Horizon 4, which stops it from sounding like a jet engine. I will have to try GTA V, since this game is a bit more requiring in CPU and GPU departments. No tweaking on graphics card yet, but probably could overclock it, since it almost never exceeds 75-78C (1060 6GB)

Thank you @hackness , it was a seemless process . BTW, do you guys think -200mV+ is achievable in battery mode?

 

Since you have already lowered the AC / DC Loadline,

Core -180mV is good enough, going further won't lower the voltage anymore.

Now what you need to do is focus on the cache.

Decrease in 5mV for the cache process,

After the AC / DC Loadline Optimization, the cache undervolting will become very sensitive.

But you'll see the error before BSOD kicks in, so you can react and revert back a few tick of the mV quickly.

As for battery, use this for your reference, Core -165mV Cache -100mV for AC/ DC Loadline set to 5.

Edit: If you need full control over the mutiplier in battery mode, check out chapter 2 and 3 here:

http://forum.notebookreview.com/thr...trol-of-the-i7-8750h-advanced-version.823065/

 

Hi guys, since 6th gen CPUs the CPU / Cache undervolt is a single setting to the CPU - it picks the smaller number of the 2 settings it finds and goes with that - so if you have a Cache of -180mV and a Core of -130mV (for example), the undervolt the cpu applies is -130mV, and it ignores the -180mV for the Core/Cache undervolt setting.

XTU was changed to combine the core / cache undervolt to a single number setting, and initially so was TS.

@unclewebb says he got so many complaints about this change that he changed it back, as some people 'see' some beneficial side effect from different settings, even if the undervolt setting is skewed to the lower setting and has no affect toward setting different core/cache undervolts.

http://forum.notebookreview.com/threads/the-throttlestop-guide.531329/page-776#post-10722214

http://forum.notebookreview.com/threads/the-throttlestop-guide.531329/page-776#post-10722181

So it's best to undervolt both core/cache at the same setting until you find the stable undervolt - then after you have a stable setting for undervolt you can play around with that TS "side effect" by increasing one of the two undervolt settings - usually people increase the core undervolt - although @unclewebb suggests increasing the cache undervolt to twiddle the side effect.

Just a heads up since this split undervolt core/cache setting with TS is a bit confusing to many.

 

Yes, that is pretty much what is happening there up to a certain degree.

I've come through that in the big ThrottleStop thread,

But when I saw the voltage sometimes bumps like a spike even when the CPU is at 100%.

The concept lost its value, I had to look for a new solution.

So far the only answer I came up with which solves this issue, is the Core voltage always needs to be undervolted more than the Cache.

 

Yeah, I really wish @unclewebb would remove that confusion by following spec and linking Core / Cache undervolt like XTU for Intel Gen 6 and newer CPU's.

It really messes with people's minds, and they start seeing things that aren't there - like mirage's and stuff like that. With uncertainty like that split yet not split undervolt between core / cache, all random environmental contributions get to play with people's minds and further confuse the issue.

Especially with helpful people with good intentions trying to help others, it just keeps the myth alive that there are 2 different settings for core / cache undervolt, when there isn't.

It's best to undervolt with both core / cache settings together at the same value, stable at 100% CPU usage, then test Idle stability - nothing running in the background - and reduce undervolt +5mV (cache and core) if you get a BSOD at lowest Idle, then leave the undervolt alone after that.

Following spec and combining core/cache undervolt into a single linked setting has worked for everyone with XTU for years, and still works today.

It would be nice if TS would set combined core/cache undervolt as the default, even with an option to "enable split core/cache side-effect" to appease those that insist the mirage is real.

 

I can't really comment much on this as I didn't try using the XTU on the GS65.

I went for the ThrottleStop because it allows profile switching between AC and Battery and also @unclewebb is receiving a lot of message from tweakers day by day. I think the outcome of ThrottleStop simply beat the XTU in almost every aspect.

But yes, it is time to do some verification, that if the voltage still bumps like spikes under 100% usage at AC / DC Loadline set to 5~20 by using XTU to undervolt.

 

Um I don't know what's going on but I just tested this.

Using static (override) voltages:

Core=1.177v, Cache=1.205v, CPU core uses the cache voltage (higher of the two).

Core=1.205v, Cache=1.177v, CPU core uses the core voltage (higher of the two).

this "voltage jump" you are talking about...is this a 30mv jump?
Or is it something else?

Can you two please set AC DC loadline to 1 (not 5) and see if this jump still happens?

Because this 30mv jump is supposed to be from the presence of AVX instructions being used. Windows does use AVX instructions for certain tasks.
Shouldn't happen just from moving the cursor.

 

I hope you only use my quote as reference and not for me to respond... Don't get half of what you're on about lol.

 

I'm not arguing that TS doesn't have benefits over XTU, although I haven't needed them myself, others report it's helpful for their usage.

I don't mind helping people using either XTU or TS. I suggest XTU for newbies where I've found the details of TS would be confusing. Setting a single setting in XTU - undervolt - is confusing enough for most non-technical people so it's a good place to start.

The undervolt picks the smaller of the 2 core / cache undervolt setting numbers - and that's the undervolt.

The side-effect some see by varying core / cache undervolt higher from the stable undervolt is more of a confusion between an unstable higher undervolt which is now not used because the lower number is used instead.

Once that unstable higher undervolt isn't being used - it appears more stable - because it is.

Using prime95 with AVX/FMA disabled to test for highest undervolt lets you see a more sensitive test - threads die with math errors before BSOD occurs, so it can find unstable 100% CPU load undervolt with more accuracy, which helps get the undervolt lower to a stable setting.

Then reducing the undervolt another +5mV to +10mV for Idle stable undervolt setting by letting the laptop sit idle with apps / background services stopped gets a good final highest stable undervolt.

When I've helped people tune using TS this is what we found, the original highest undervolt was just too high.

 

So based on the instructilns @hackness posted a couple pages back, what would be a good starting point for undervolt? It is running -160 core and -43 cache arm, after bios tweaking. Should I try and do -50 -50, or can I go to f.x. -130 core/cache, like I did before the bios tweaking?

 

Nevermind, I understand now. I'll test this out when I get the time and energy

 

I'd link the 2 settings - Core / Cache undervolt - to the same value to find the "real" undervolt, use prime95 with AVX / FMA disabled (see below) - use the "small fft's" test to find the best 100% CPU load undervolt.

Then test for idle stable undervolt setting - you might need to reduce the undervolt from best setting at 100% CPU load - generally another +5mV or +10mV is enough.

In your case try -120mV / -120mV if your -130mV / -130mV is already accurately the highest stable 100% CPU load undervolt.

How to disable prime95 AVX / FMA:

If you have prime95 running, exit prime95 and add these lines to the top of local.txt:
CPUSupportsAVX=0
CpuSupportsAVX2=0
CpuSupportsAVX512F = 0
CPUSupportsFMA3=0
CpuSupportsFMA4 = 0

Then reset the the current readings in hwinfo64 and restart prime95 and run "small FFT's".

Watch for threads exiting - the tiny windows stop scrolling and the message that an error was detected will show up in each tiny thread monitoring window - if the undervolt is too high - reduce the undervolt and restart prime95 again.

Also, run with 100% fans enabled, don't rely on automatic fan control - if you do use auto-fan control then click "reset" on hwinfo64 a few seconds after tests starts to clear initial "red" thermal throttle readings, giving the autofan curve time to spin up the fans.

Have fun

 

In most cases a simple first setting of -100mV/-100mV if stable is "good enough" to give a 10C drop in 100% CPU load temps, which stops any thermal throttling if present.

If you want to tune for more undervolt it may / may not make much difference in temps.

If you are ok at -120mV / -120mV (or even -130mV / -130mV) you are likely good for now, and can come back later when you feel so inspired to tune for maximum undervolt.

For most people a slightly lower undervolt setting choice (-120mV / -120mV) to assure stability is far more of use than that small improvement potential from a yet lower undervolt that may prove unstable in the long run.

 

Yes, it is the AVX instruction that is adding the voltage.

But what I'll show next is 3 different kinds of combination, just to demonstrate the differences between them, this is just for testing purpose so the multiplier is lowered:

Core -37.1 mV / Cache -44.8 mV Test

Core -37.1 mV / Cache -44.8 mV No AVX landed at 0.9348V



Core -37.1 mV / Cache -44.8 mV with AVX landed at 0.9799V ( That is +45.1mV with AVX instruction on)



Core -44.8 mV / Cache -37.1 mV Test

Core -44.8 mV / Cache -37.1 mV No AVX landed at 0.9248V



Core -44.8 mV / Cache -37.1 mV with AVX landed at 0.9698V ( That is +45 mV with AVX instruction on)



Core -165 mV / Cache -37.1 mV Test

Core -165 mV / Cache -37.1 mV No AVX landed at 0.9198V



Core -165 mV / Cache -37.1 mV with AVX it still remains at 0.9198V.

 

If you are not extensively using the AVX instruction like running the Prime95 with AVX enabled,

I'd suggest going for more undervolt on the core than the cache.

But if you need the AVX function with full support from the CPU, you can go with the 2 setting the same, would be something like -43 / -43.

Because with -160 / 43,

running Prime95 with AVX instruction,

you'll notice some cores are under performing by not utilizing all the CPU %, about 1-2 cores out of the 6 cores will run at 20% usage because of not enough power to supply to keep the AVX instruction running at full speed,

but in reality usually having the CPU run at 100% for AVX instruction is rare,

mostly happens when you insist to run the Prime95 with AVX instruction.

 

Hackness,
Can you do these tests with static voltage (the two core / cache voltages set manually and inverted?).

Thank you for the pictures, though.

 

From your testing...
Core -37.1 mV / Cache -44.8 mV Test (Core -37.1 mV / Cache -44.8 mV No AVX landed at 0.9348V)


Core -44.8 mV / Cache -37.1 mV Test (Core -44.8 mV / Cache -37.1 mV No AVX landed at 0.9248V)



Can you run same test but with Core and Cashe voltage (undervolt) equal? Put lower voltage on the Core and you will see power consumption will decrease. This is normal.

You could also test with a massive undervolt on Cashe vs. the Core.

Well known that it can be possible with put two different undervolt. But it's also possible that it does not work properly for everyone (We did the opposite voltage of what you do on older models - higher undervolt for the cashe). We / I used this for the Skylake chips, but for stock or very minor overclock. Useless if you wanted a nice overclock.

 

Oh shoot I've just noticed I put the number in the title wrong, supposed to be 48.8 for the 44.8 like in the picture.

OK, we need to test this out in phases, since there are many request:

Edit: the tests below will be tested again with static voltage.

All phases will be carried out with AC / DC Loadline setting to 1.

Phase 1:

Core -165.0 / Cache -43.0 @ 3200MHz
Core -43.0 / Cache -165.0 @ 3200MHz
Core -43.0 / Cache -43.0 @ 3200MHz

Purpose of this test is the clarify the result of setting the Core and Cache to the same level differs from the other 2 samples.

Phase 2:

Core -165.0 / Cache -43.0 @ 3200MHz
Core -43.0 / Cache -165.0 @ 3200MHz
Core -43.0 / Cache -43.0 @ 3200MHz

Purpose of this test is the clarify with the AVX instruction given by Windows, will there be any voltage bump for each sample.

@John Duncan I think the above 2 tests will give you the answer you are looking for.

Phase 3:

Core -165.0 / Cache -43.0 @ 3200MHz
Core -43.0 / Cache -165.0 @ 3200MHz
Core -43.0 / Cache -43.0 @ 3200MHz

Run using Prime95 with AVX instruction turned on.

Purpose of this test is the clarify with the Prime95 AVX instruction turn on, will the CPU be under performing for each sample. This test will produce massive heat. So I might decide to run with lowered Multiplier.

Phase 4:

XTU -43 Core @ 3200MHz

Purpose of this test is the clarify using the XTU vs ThrottleStop will there be any differences, test result from Phase 1 will be reused in this phase.

 

Why not put the max undervolt for core and cashe? This means you get the correct comparison between all 3 settings. If -165 is all too much for both core and cashe then change the others and lower their values so they all being equal.

 

There are probably a lot of reasons and factors to consider so to try and be as simple as possible, is it better to set AC/DC loadline to 5/5 or 10/10 when it comes to wanting more performance?

 

Because the test will be performed under the situation that AC / DC Loadline = 1

Under this situation, the undervolt slider cannot really move that much otherwise it'll just BSOD straight away.

Under AC / DC Loadline = 180, the default setting, it'll allow you to move the undervolt slider by a lot, down to maybe -125mV or more for both.

But the result of the temperature is going to be nearly the same, but setting to 1 should perform better, because when a spike happens under load, it will actually add temperature spike to it as well.

Usually with the Loadline setting to 1, under load, if there is no AVX instruction used, the VID will stay the same, where having the loadline set to 180, it goes all over the place.

Anyway, we want to see the clear result, it is better that the voltage doesn't spike so we know exactly what voltage it is running at right now.


As for the advantage of having low AC / DC loadline setting, maybe @Falkentyne can share more details about this.

Edit: Also I was trying clarify that is it really that the actually voltage is only affected by the slider that is lower than the other, in the other samples I have the lower side of the undervolt set to -43.0, so the one for the same level would be -43.0 for both.

I personally prefer 5~10 for normal usuage but for the testing purpose, and to have less factors affecting our test result, I think 1 is probably more suitable in this case.

 

I know all this hence I said you if needed could lower the voltage values. But Oh’well.

 

Please check your PM

 

Releasing Phase 1 test result:

All done with Adaptive voltage and AC / DC Loadline = 1.

Verdict:

Seems like the Core -43 Cache -43 and Core -43 Cache -165 are the same thing.

But Core -165 Cache -43 reacts differently, it can further lower the voltage.

Phase 1 Core -165 Cache -43 @3200MHz @0.8948 No AVX ( Sorry the voltage typed wrong in the text)



Phase 1 Core -43 Cache -165 @3200MHz @0.9098 No AVX



Phase 1 Core -43 Cache -43 @3200MHz @0.9098 No AVX

 

Releasing 2nd Phase test result:

Verdict:

Seems like with Core -43 Cache -165 and Core -43 Cache -43 are reacting the same,

they both have +45mV when AVX instruction kicks in.

And the Core -165 Cache -43 reacts differently,

It is further more undervolted,

Also,

Having +0mV when AVX instruction kicks in.

Phase 2 Core -165 Cache -43 @3200MHz @0.8948 -> 0.8948 with Windows AVX + 0mV



Phase 2 Core -43 Cache -165 @3200MHz @0.9098 -> 0.9548 with Windows AVX + 45mV



Phase 2 Core -43 Cache -43 @3200MHz @0.9098 -> 0.9548 with Windows AVX + 45mV

 

And another pm

So I tested this with static voltage and it seems as if AVX instructions use the CPU core voltage +30mv while non avx instructions use the higher of the cpu or cache voltages.

 

In the newest test I've found that sometimes the stable voltage for none AVX load may cause the CPU to not run at 100C% because the voltage was too low for the AVX instruction. But really, not many programs use AVX so I ignore the voltage needed for AVX and went for further lowered voltage

Releasing Phase 3 Test Result:

Verdict:

In order to demonstrate also that undervolt too much will sometimes cause the CPU to under perform,

I've bump the lower side to -60.5 mV.

It seems to be the same as Phase 2 that the Core -60.5 Cache -165 react the same as Core -60.5 Cache -60.5,

But both can run the AVX struction at 100C% no problem.

But with the Core -165 Cache -60.5, 2 cores are under performing when the prime95 AVX kicks in.

But the plus side of core higher than cache is that if you don't need to deal with AVX instruction,

this can get your a lowered voltage than the other 2 samples.


Phase 3 Core -165 Cache -60.5 @3200MHz @0.8798 -> 0.8798 with Prime95 AVX with C% Throttle on the 4th core and the 6th core



Phase 3 Core -60.5 Cache -165 @3200MHz @0.8948 -> 0.9397 with Prime95 AVX plus 44.9mV No C% Throttle



Phase 3 Core -60.5 Cache -60.5 @3200MHz @0.8948 -> 0.9397 with Prime95 AVX plus 44.9mV No C% Throttle

 

If you see a c0% drop like that check windows event viewer to see if you got any WHEA correctable errors there

 

I've checked but I didn't see anything like that though,

However the RING: Max VR Voltage, ICCmax, PL4 and GT: Fuses Limit in HWINFO64 were triggered.

My iccMax is already at 255.75A though.

But could it be it is just the voltage is too low for the AVX instruction to run all cores at full C%?

I'll check again with log enabled when I got a chance to run on AC power, currently running on battery, probably few hours later.

 

@hackness

So what you are saying is, via hours / days of many posts: to everyone not interested in endless hours screwing around, at default BIOS settings set your undervolt core / cache both to -100mV and enjoy your laptop.

There may be a measurable side effect of splitting core / cache undervolt to widely divergent values, with the smaller undervolt value actually being used for the undervolt, but it's a long and drawn out process with probably no user describable effect on FPS.

All of this wasted time, over and over, because the TS settings for undervolt are split between core / cache, which stopped being used by Intel CPU's starting with the 6th generation. @unclewebb

 

He was trying to find out why the voltage kept jumping up randomly when core and cache were undervolted to the same value.

What he found out, and I verified, that :

1) AVX instructions use the cpu core voltage + 30mv added to it. (Doesn't matter if the core voltage is lower or higher than the cache voltage).
2) NON AVX instructions use the HIGHER voltage of the CPU core and CPU cache, if both are at different values.

This explains a LOT Of things, @hmscott

it explains why some people had lower power consumption when they set the CPU core much lower than the CPU cache. Because they were using stress tests that tested AVX instructions.
While others saw it used the higher of the two voltages (non AVX).

 

We already knew all of this. Why do you pretend to say you didn't know?

That's why we run with AVX / FMA off, as it adds too much voltage to the CPU and increases heat output, giving a skewed reading compared to all other tests running without those extensions.

Your undervolt setting needs to account for your use case, if you run without AVX / FMA which is most of us most of the time, then disable them - if you are going to run software with AVX / FMA then tune for that usage too, and set a profile each for AVX/FMA and non-AVX/non-FMA usage.

We also already knew, me from listening to @unclewebb , that the smaller undervolt (-30mV is smaller than -100mV for example), is the one the CPU uses if the Core / Cache undervolt are split.

The side effect's for splitting the Core / Cache undervolt really aren't worth pursuing for most people, as it will have no effect on their perceived performance, it's just time wasted for the average user / gamers.

Most people just want to drop their 100% load CPU temps, and if -100mV / -100mV is stable, they can drop 10c off their peak temperature, which gets most CPU's under thermal throttle point.

All of the other tweaking is academic. It may be fun for you, and a few others, but forcing everyone - average users / gamers that could care less about endlessly tweaking for indiscernible performance differences - only want to get their undervolt set quickly for lower temps, and get back to gaming / work.

Again, I wish @unclewebb would make the default TS setting for the Core / Cache undervolt to be locked / the same, and provide an option for split undervolt, to avoid sending everyone using TS on a wide goose chase for performance that doesn't matter.

 

Actually, I think it is a good thing that the test was carried out.

Because now this test has told us that there are options to pick,

depending on how we want to approach when undervolting:

1. If you don't care about the AVX, setting Core higher than the Cache will give you a result with lowered temperature than setting both to the same value.

2. If you care about the AVX, setting Core and Cache equally can prevent the C% to drop from 100% usage.

Yes, I think the AVX only uses Core voltage, cache doesn't seem to have any effect to the AVX voltage.

The part with the Non-AVX is what I was thinking too.

For this I'm still not quite sure if the XTU affects the AVX voltage, I'll have to try it out when I can .

 

Like I said, this has all long been known. Look at @unclewebb comments I quoted earlier, and look for AVX voltage posts via google, here's one from 2013:

Adaptive Vcore and AVX temperatures - A revelation!
https://hardforum.com/threads/adaptive-vcore-and-avx-temperatures-a-revelation.1777509/

And, if you set the undervolt the same value for core / cache, -100mV / -100mV for example to get 10c drop in 100% CPU load temperature, it doesn't matter which value is used by what in the CPU - they are both the same.

Anyway, have fun.

 

What benefits will I get if I lower Cashe undervolt from -0.202 down to let’s say -0.050 and still keep Core at 0.202mV ? You mean I then can increase undervolt for the Core domain for better results? Eg -0.250 or even -0.300mV ? Nope!

None forcing people to tweak their computers a special way. Up to themselves.

 

Oh I mean setting it like this pattern Core -150 Cache -50.

 

Does anyone know of any MSI GS65-specific accessories? I've found decals and a keyboard cover on Amazon. Has anyone found anything else? Do you have any recommendations?

 

Hi, anyone encountered problem with steelseries engine not detecting the per key RGB keyboard? it started as just not connected status and now it doesnt even show me that. I was greeted by a screen that asks me to connect a steelseries enable device.

I am running Windows 10 1809 now (i know, that may be a problem) and the latest version of SSE software (3.12.13). I can still see the red lights when pressed on the Fn key so I am pretty sure it's not hardware level faulty.

Thx.

 

It's best to do a clean install since any issues can happen from an upgraded OS.

 

Hey everyone,
I've been trying the latest GPU drivers from nvidia for the GTX 1070 Max-Q and I noticed Optimus is not working. The last working driver was 399..24. After that, Optimus is broken.
I'm curious if anyone has had similar issues.

Setup:
Windows 10 64-Bit (1803 and 1709)
GTX 1070 Max-Q
Drivers tested -

• 411.63
• 411.70
• 416.16

Thing's I've tried:

• GPU is set to auto select
• Phys X is set to auto select
• No backround programs that utilize the dGPU
• All browsing programs set to use iGPU
• Tried setting ALL PROGRAMS to iGPU

Thanks.
-JRey

 

Exactly the same issue I am having. I posted on the MSI Reddit Boards, I also attempted DDU uninstalls and reinstalls in safe mode.

GTX 1070 Max-Q 399.24 is the latest one i could get with Optimus working.

Also, I ran Processor Explorer and GPU he is limited to dwe services and csrss.exe so it should switch to Optimus considering it is unsafe to cancel/kill these tasks.


Sent from my BLA-L09 using Tapatalk

 

Posting your problem in the Nvidia forum support thread for the problematic driver (latest) can help Nvidia find and track the bug, and hopefully fix it in the next release:

GeForce > Community > Forums > Support > GeForce Drivers > View Topic
Official 416.16 Game Ready WHQL Display Driver Feedback Thread (Released 10/4/18)
https://forums.geforce.com/default/...lay-driver-feedback-thread-released-10-4-18-/

 

My GS65 is the BestBuy version with a SATA SSD in the 1st Slot. I installed a NVME drive in the 2nd slot, installed Windows on the NVME, and converted the SATA into a storage drive. Would I notice a change in performance if I install games on the SATA drive?

 

Not much, 500MB/s vs 2000MB/s still give you 6 second load on the OS.

 

My keyboard is partially not lighting up. Is anyboady having the same issue? And is there a fix for this?

 

On the bottom side of your laptop find the manufacture date/year. Some earlier fabricated models have a rivet that compresses the keyboard flexcable. I would make sure you rule.out software errors before attempt to disassemble the laptop and check if the flex cable is being crushed by a rivet.

Sent from my BLA-L09 using Tapatalk

 

Regarding my post about thermal pads and the core temp differentials. 0.5mm pads near the CPU DO NOT make contact with the HS. I went back to 1.0mm pads on that area.

Sent from my BLA-L09 using Tapatalk