Ethereum GPU mining?

[SK10H]

ETH seems to be recovering, reaching $11.10 as I type.

Good outcome for those of us who traded their ETC at $3.30 for ETH at $7.50. BTC seems to be picking up a little as well. Only question is how long ETC will hold on for.

Nice trade.

Got 2 Sapphire Nitro RX470 8GB as it has 8Gbps memory. Very overpriced as it's 2 separate shipping.

$240->$330 =1.37 exchange rate in Canada, double gouging by Newegg, 1st on non-ref price, then on exchange rate, finally on separate shipping.

May look for the cheap Powercolor 4GB when it show up.

 

[DrMrLordX]

I thought the 470 MSRP was meant to be $149?

 

[thilanliyan]

I thought the 470 MSRP was meant to be $149?

That's what I thought too but apparently we were misinformed. I think in the Toms article it said AMD set the MSRP at $179?

 

[Newbian]

I thought the 470 MSRP was meant to be $149?

That's what people wanted and hoped for.

 

[n0x1ous]

That's what people wanted and hoped for.

Thats what would make logical sense, so for AMD to not put it at that makes sense, if that makes sense

 

[Madpacket]

Miners ruining everything! ;D

Look at it this way, just mine for a few weeks and that extra $$$ difference disappears.

Oh AMD...

 

[IEC]

It looks like there's quite the variation in RX470 card performance too, depending on the memory speed. The models with 8GHz, 8GB perform within 8-10% of the RX480 no sweat. Models with 6.6GHz memory are quite a bit below that.

 

[SK10H]

Just got 2 more RX480 8GB, I'm done buying, no more slots and power consumption headroom, will retire the old cards.

RX470 8GB is too close to 480 price, got way more ref card than expected.

 

[IEC]

I only have reference cards. And they are all 8GB (even the 4GB ones)

The blower fan is perfect for the card when undervolted, and in open air cases (like crates) they are quiet, running at ~1700 RPM.

 

[SK10H]

I only have reference cards. And they are all 8GB (even the 4GB ones)

The blower fan is perfect for the card when undervolted, and in open air cases (like crates) they are quiet, running at ~1700 RPM.

I am kind of hesitated to put 3 ref card on 1 board with the slot power issue, got 2 boards with 3 pci-e x16 although each card is only about 120-130W. I am not aware of the review that point out how much power it's going to draw from the slot after you undervolt with the driver fix.

I bought lot of power riser though.

 

[IEC]

I am kind of hesitated to put 3 ref card on 1 board with the slot power issue, got 2 boards with 3 pci-e x16 although each card is only about 120-130W. I am not aware of the review that point out how much power it's going to draw from the slot after you undervolt with the driver fix.

I bought lot of power riser though.

Powered risers (separate SATA power for the riser card) avoid any potential overloading of the motherboard +12V.

I am running 6 cards on a Asrock H97 motherboard with no issue.

 

[Madpacket]

Powered risers (separate SATA power for the riser card) avoid any potential overloading of the motherboard +12V.

I am running 6 cards on a Asrock H97 motherboard with no issue.

IEC,

I built a 6 x reference Rx40's on a H97 Anniversary rig this week. Noticed weird throttling happening where Claymore will start out mining at 27Mh and then after a few minutes the cards drop to 20-24Mh...

Have you noticed any issues like this?

Basically the same settings on one card will hold stable at 27Mh but if applied to 6x cards the cards (even the ones that work perfect individually) they start to throttle with the same settings applied.

I was able to get them mostly stable at 26Mh by upping the GPU voltage considerably (900mv to 975mv) and lowering the clock speeds from 1040/2220 to 980 core / 2200 memory but this is undesirable.

I'm sure with more tweaking (ran out of time last night) I can get them to a better Mh/W balance but curious if you ran into similar issues. I'm using powered risers and a 1KW Platinum EVGA G2 for reference. Plenty of cooling as well.

I still have two more milk crate H97 Anniversary's to fill up with 480's so I need to figure this out

 

[Madpacket]

It looks like there's quite the variation in RX470 card performance too, depending on the memory speed. The models with 8GHz, 8GB perform within 8-10% of the RX480 no sweat. Models with 6.6GHz memory are quite a bit below that.

Do you know if there's any 4GB models with 8Ghz memory? Or is it just the 8GB models.

 

[IEC]

IEC,

I built a 6 x reference Rx40's on a H97 Anniversary rig this week. Noticed weird throttling happening where Claymore will start out mining at 27Mh and then after a few minutes the cards drop to 20-24Mh...

Have you noticed any issues like this?

I haven't personally. I'm using the latest beta Claymore, if that helps. Along with 16.7.2 drivers on Windows 10 Pro. Power supply is a eVGA 1300W G2 (I know, overkill).

Do you know if there's any 4GB models with 8Ghz memory? Or is it just the 8GB models.

I'm not aware of any with faster 4GB memory. They all seem to be 6.6GHz-7.0GHz from what I can see.

 

[Madpacket]

I haven't personally. I'm using the latest beta Claymore, if that helps. Along with 16.7.2 drivers on Windows 10 Pro. Power supply is a eVGA 1300W G2 (I know, overkill).



I'm not aware of any with faster 4GB memory. They all seem to be 6.6GHz-7.0GHz from what I can see.

Thanks for the replies. That's too bad about the 4GB cards, oh well. If you get a chance please let me know what version of BIOS your motherboard is using.


Here's more detailed analysis. A friend of mine with an identical rig (using a 1300 G2) is having the same problem. Posting this here mainly for others who may have similar issues:

---


Analysis of the problem:

Using Claymore 5.3 beta with 1 card, a setting of 900mv (core/mem) @ 1060Mhz GPU / 2200 Mem gives you approx 27Mh. No throttling experienced.

When you build a milk crate miner or just adding additional cards you start to see throttling after a few minutes of mining with the exact same speed and voltage settings applied. Throttling can easily be shown with GPU-Z running, just need to monitor the GPU core speeds and you'll see it drop to roughly 800-950mhz (depends on card).

Things I tried which didn't help:

- Added more cooling (this appeared to help my 5x rig but it just prolonged the throttling)
- Reduced GPU Core to 1040Mh (no change in throttling)
- Increased available power (bottom setting under fan settings) to +50 on all cards (no change)
- Reduced power to -15 on all cards (no change)

I finally mitigated most of the throttling via the following changes:

- Set every card back to default settings (and validated all six cards mined at 24.xxMhz without throttling).
- Changed GPU Core to 980Mhz, upped memory at 2200Mhz
- Changed GPU Core voltage to 975mv, changed memory voltage to 975mv (if you're still on 16.7.2 driver just change the memory voltage). I'm using 16.7.3 and yes I can change memory speeds up to 2200 or higher.

These changes stopped throttling in 5 of the 6 cards (one of the cards still drops to 24Mh).

975mv is likely overly aggressive (either on core or mem) but I was very time limited. I will retest tonight and dial in the voltages as each card is a little different but the above settings should get you over 26Mh per card. My rig has been running all day at 155Mh. I'm pulling around 925W though so obviously I'll need to spend more time tuning. My 5x rig only pulls 680w at 130Mh with the same 1KW EVGA P2 power supply.

Be careful if testing at default voltages. 6x cards will pull 1150W from the wall (which is safe for short testing but could be detrimental on a 1KW power for prolonged periods).

I'm still unsure as to why the cards are throttling when combined but I suspect it may have something to do with inadequate power delivery to the risers supplied via the SATA/Molex connectors. The Radeon 480's have a split rail system where the power draw on the PCIe port is higher than the more traditional mining cards. They pull about half their power from the PCIe slot and other half from the 6x PCIe power connector.

Time permitted I will split up the PCIe riser power delivery more evenly and see if this makes a difference. Alternatively I may take two power supplies and split up the load.

I haven't messed with anything in the BIOS yet but there could be a setting in the H97 Anniversary board that needs attention.

 

[IEC]

I don't recall changing anything in the BIOS, or even bothering to check the BIOS version. I do however have each card and riser with its own separate power cable running to the PSU. It could also be luck of the draw with GPUs and getting the voltage dialed in right (I had to raise voltage on a few compared to the 898mV of my better cards).

 

[DrMrLordX]

Anyone know if 470s have the same split power problem as the 480s? Also, Madpacket, what OS are you running on your 6x 480 system?

 

[Hitman928]

Anyone know if 470s have the same split power problem as the 480s? Also, Madpacket, what OS are you running on your 6x 480 system?

The 470s are all custom designs so it depends on the particular card. The one review I saw at toms showed less than spec draw from the MB but over spec on pci-e. That's not really a problem but those were the results.

 

[DrMrLordX]

That's a shame. These cards need to draw more heavily over the PCI-e power connector. AMD had a chance to fix that with the 470 . . . pity they didn't.

 

[Madpacket]

I don't recall changing anything in the BIOS, or even bothering to check the BIOS version. I do however have each card and riser with its own separate power cable running to the PSU. It could also be luck of the draw with GPUs and getting the voltage dialed in right (I had to raise voltage on a few compared to the 898mV of my better cards).

OK thanks. I'll rewire the risers so they are spread out better.

 

[Madpacket]

Anyone know if 470s have the same split power problem as the 480s? Also, Madpacket, what OS are you running on your 6x 480 system?

Windows 10.

All my miners (9) use Windows 10. I haven't had any issues with 390's or 290's. Something weird with the 480's when running multiple cards.

 

[n0x1ous]

madpacket, I hope you get the throttling resolved.

 

[tracerit]

I just bought a RX 470 primarily for gaming but figured I'd try some mining while I'm at it since I leave my computer on 24/7 for my home security cameras.

How much stress does mining put on the card (and overall system)? I've read that mining ruined XFX's lifetime warranty on their cards so I'm curious now.

 

[Madpacket]

Thanks n0x.

 

[Madpacket]

I just bought a RX 470 primarily for gaming but figured I'd try some mining while I'm at it since I leave my computer on 24/7 for my home security cameras.

How much stress does mining put on the card (and overall system)? I've read that mining ruined XFX's lifetime warranty on their cards so I'm curious now.

Short answer,

Very little stress, and likely much less than actual gaming.


Long answer,

It depends on a number of factors (quality of the card, heat, voltage, dust, power delivery, type of algorithm mined, etc).

If you're conservative you should be undervolting / underclocking your 470 GPU and undervolting but overclocking GPU memory to get to the ideal Mh/w balance.

Mining ETH/ETC this will draw less power and generate less heat waste on your 470 than gaming.

Now with mining your card will be running for prolonged periods of time so just be mindful of heat and dust and the card will last you many years.

Most horror stories from mining are from clueless or careless people who run their cards overclocked and overvolted to get every last Mh out of them. Heat be dammned. Do this for prolonged periods and yes you can shorten the life of your GPU.

However compared to the Hawaii 28nm 290/390's, the 470/480's sip power and generate a lot less heat therefore I suspect prologoned mining, even in the worst conditions, will be less of an issue for the Polaris cards.


tldr;

Keep the card running cool (under 80C) and blow out any dust every 4 - 6 weeks and the card will outlive the warranty.