Elmor Labs AMPLE - First Look

[mllrkllr88]

Elmor recently started taking orders for the AMPLE 20A power card. It’s a simple little single-phase VRM that can be used to replace a broken VRM or simply overcome that pesky OCP/OVP. The most common potential usage would be for a GPU memory rail. However, with an output voltage from 1.0-3.4 V it could be used for many different applications. 

I obtained a test sample of this new device and I intend to put it to the ultimate test. I have an RX 580 GPU with a damaged memory voltage rail, but otherwise the GPU is in perfect condition. The plan is to solder on the AMPLE power card and observe the impact on the memory overclocking. I know what the card was capable of before the memory VRM died, so it’s the perfect test case for this little device. So follow along as I put this thing to the test and find out if it’s capable of delivering sufficient power to 8 GB of GDDR5.

The AMPLE Device
The device ships in an anti-static bag with the 6-pin connector added in as an accessory. This gives you the flexibility to power up the AMPLE from any qualifying voltage source you desire. 

Product Specification

• Output voltage adjustable 1.0-3.4 V using the on-board potentiometer (range can be increased by adding your own potentiometer)
• Nominal input voltage 12 V (functional from 4.5 V up to 16 V)
• Max output current 20 A (thermally limited). Active cooling may be required to achieve high output current.
• Efficiency at max output 91.5% (3.4 V, 20 A, 750 KHz)
• Switching frequency selectable between 300 KHz and 750 KHz
• OVP (+20%), OCP (24A) and OTP (Tj = 150*C)
• Output voltage can be monitored and adjusted using the EVC2 VMOD1-header
• Ships with a PCIE 6-pin power connector which can be soldered on by the user

Source Credit: elmorlabs.com

There is one switch on the device which acts as a mode selector. There are a  total of 4 different combinations you can set, which change the output delivery settings. 

• Switch 1 OFF = 750 KHz switching frequency 
• Switch 1 ON = 300 KHz switching frequency 
• Switch 2 OFF = PSM = Power Saving Mode, higher ripple but lower loss at low loads
• Switch 2 ON = CCM = Constant Conduction Mode. The PMIC is always regulating to the best of its ability, for the tightest possible voltage regulation. In some cases, for example with this project at 3.0V+, that will create a high loss in no-load or low-load situations.

For controlling the voltage there are multiple different methods you can choose from. To make things simple, there is a potentiometer on the power card for direct control. For more advanced control, you have the ability to integrate the device with an EVC2 module using the vmod connection.

Mounting the Power Card
Before you just hastily solder on the AMPLE device, you should take the time to power it on and make sure it works. I would suggest powering it up and setting the desired voltage before you attempt to solder it on a GPU. 



The underside of the AMPLE has power and ground planes. In my particular case I won't need these connections so I decided to insulate them with a few layers of Kapton tape. This is an unnecessary step, but in my case there were a few components on the card which were extremely close to shorting on the AMPLE device. 



I chose to mount mine with 0.8 mm copper plate. I kept the plane-to-plane distance as short as possible. Furthermore, I always prefer to make my ground connections on the front of the card after where the main connection is. 
In the picture below you can see both ground connections come after the memory plan connection. In my testing this provides the best realized MHz whether it be core or memory external VRM replacement.  




Test Setup
The overall goal is to compare stock PCB memory overclocking with AMPLE power memory overclocking. To make the test consistent, I will use the same load voltage for both test conditions. The test methodology is to overclock memory and record the passable frequency in Fire Strike GT1.  I will start at 2280 MHz, increase +10 MHz until it becomes unstable

• PowerColor RX580 8GB Golden Sample
• Z490 test platform
• AMPLE VRM on memory rail 
• Memory IC: Micron D9VVR
• 1625 MHz strap timings copied to 2000 MHz strap (Most efficient timings possible without diminishing returns on frequency)

In order to get a realistic look at the voltage response of the device, I have placed voltage test points (TP1-3) at various distances away from the main inductor. During the load tests, I will also be conducting thermal tests of the mosfet heatsink. Ambient temperature for all tests around 24c

The voltage read points are as follows:

• TP1 = AMPLE power voltage at C10 cap
• TP2 = GPU plane voltage at solder connection 
• TP3 = Memory IC voltage at MLCC cap on back of card, at least 100mm distance from AMPLE Vout solder connection 

Test Results 

Voltage droop is the enemy of external VRM mods. It’s important to know exactly where the voltage droop occurs because it can narrow down the cause. The voltage test point results indicate that the droop is not caused by the solder connection. The voltage droop is present at the C10 capacitor so therefore it’s coming directly from the AMPLE VRM. This makes me happy because the solder connection is relatively perfect, but it’s mildly concerning to see the AMPLE being stressed under load.

Regardless of any droop, the card was still able to pass FS GT1 test at the highest possible memory frequency. There is effectively no difference between the stock VRM and the AMPLE VRM for this test application. 
 
The temperature reached a peak at the end of GT2 and was still climbing. With the Fire Strike test, GT3 is CPU based so it had a cooldown period. The max temp of around 50c seems reasonable but further testing may be needed. Sustained loads of 1hr might show a different result. 

SW1 Test results proved inconclusive. Disregarding experimental error, there was effectively no difference in the realized MHz or voltage response.

Conclusion
Overall I am extremely pleased with this little device. It performed exceptionally well and the end result is that I was able to achieve the same overclocking performance as the stock PCB. The AMPLE single-phase power card has been validated for benchmarking purposes. 

In terms of the switch options, my expectation was that CCM mode with 750 KHz would be the optimal configuration. The switch test results showed that the various combinations didn't have any noticeable effect on realized Mhz or voltage response. The 10 mV voltage droop is a mild concern because it’s coming directly from the AMPLE, however, it did not affect the overall overclocking result. 

With a price tag of just $20,, it’s an absolute must-have tool for extreme overclockers. My RX580 is the perfect use case. The card was effectively dead and useless, but the mighty little AMPLE device brought it back to life and it still holds gold in TIme Spy. Even if the price was tripled, I would still recommend it because of its potential value. 

This was just my first test, stay tuned for more tests to come. 
 

[TerraRaptor]

Great review, also waiting one piece. Do you have a chance to measure ripple at high load (i.e. 1.8v 20A)?

[Splave]

@mllrkllr88 + @elmor = anything is possible 

[Pulse88]

Nice review. 

Lets hope elmor do one with 2-3 phases version also. Love the compact design.

[elmor]

Thank you for sharing your results The droop you see should be mostly due to the output sense location which is near C20. It could maybe be improved in a future version.

On 8/3/2020 at 11:20 PM, Pulse88 said:

Nice review. 

Lets hope elmor do one with 2-3 phases version also. Love the compact design.

What would you want to do with a 2-3 phase version?

[ground]

5 minutes ago, elmor said:

Thank you for sharing your results The droop you see should be mostly due to the output sense location which is near C20. It could maybe be improved in a future version.

What would you want to do with a 2-3 phase version?

I've never tested actual current draw, but I recon a 2-3 phase setup could be useful for smaller GPUs, maybe x48 chipsets and x58 VTT rail (never actually checked current draw of that rail, need to test at some point). There are probably more usecases that I can't think off right away.

[Pulse88]

4 minutes ago, ground1556 said:

I've never tested actual current draw, but I recon a 2-3 phase setup could be useful for smaller GPUs, maybe x48 chipsets and x58 VTT rail (never actually checked current draw of that rail, need to test at some point). There are probably more usecases that I can't think off right away.

Thats right.

21 minutes ago, elmor said:

Thank you for sharing your results The droop you see should be mostly due to the output sense location which is near C20. It could maybe be improved in a future version.

What would you want to do with a 2-3 phase version?

2-3 phase was just an idea. Think your design is a win when it comes to space taken. And would love it to scale upp in phases for more use areas.

Many old gpus might only nead 2-5 better phases to be maxed out on core. And evga powerboard is to big and clumsy and overkill etc. Not to mention the rarity of evbot today. 

Best regards

[TerraRaptor]

Tested Ample2 this weekend. The greatest thing about it is that it is really small pcb and fits everywhere - taking my below average soldering skills, it allows shorter connections etc. 

Was testing it with EP45T-UD3R (feeding VTT mosfet that originally feeds from dram 1.8v), no issues so far - no noticable vdroop (0.01v) @2.3v, not extremely hot. Will be great thing for legacy vga memory mods.

[ozzie]

more excellence dave ?

[Fragmeister]

Very Nice and a good review?

[Lucky_n00b]

Nice and informative test! (really hard to find these kind of quality content these days)

Do you think it can be used to mod 'small' gpus (1030-class gpus maybe?) or the application will be suited best for memory power and/or other minor/aux rails?

[mllrkllr88]

On ‎9‎/‎7‎/‎2020 at 7:10 AM, Lucky_n00b said:

Nice and informative test! (really hard to find these kind of quality content these days)

Do you think it can be used to mod 'small' gpus (1030-class gpus maybe?) or the application will be suited best for memory power and/or other minor/aux rails?

Thank you

That's a really good question you brought up and I was wondering the same thing myself. Some of the really old stuff like 6600 GT/GTS have very weak VRM's, which can be problematic when trying to bench them. In those cases I think the AMPLE would work really well. A for the 1030 I'm not sure, it might be a close call, but I would guess that it's possible.

More testing is needed for sure, but it's an interesting use case you brought up!!