If a Polaris can do 30+MHs at <200w I'd probably buy one.
I would hope for a good bit more than that, but we'll have to see what ends up happening. A 380 get about 20Mh/s, and a 480 will have ~30% more CUs, and about a 30% higher core clock.
That would be nice, but I am scared with the combo of the 256 memory bus and an unoptimized mining program and driver (I bet it won't be able to use the 15.12 in the folder trick) it might be sub 20. We will see, I still plan to get one in a month if only for actual gaming use in my Mini ITX gaming rig since my main gaming machine is a mini mining farm right now. I am so tired of the 750 ti, $200 Polaris 10 would be a great upgrade.
I finally got the reference 290 to behave. I put on some new paste, put the fan on 79%, and put the machine on its side with a box fan next to it. In that configuration it will do 1155 stable all day which basically makes it almost equal to the 390x. I like that the blower seems hardy even if it is loud, the Sapphire 7970 Ghz my friend was letting my borrow had a fan literally rip in half while mining. Luckily some zip ties and a 120mm fan fixed that for now.
I also decided to blow some Best Buy reward points I was sitting on and I picked up a 2GB 370 from there that will eventually go in my desktop hackintosh. Now that is in my "gaming" rig with the 390x and (hanging on the side via riser) the half broken 7970. Overall that setup works pretty well because it's cool enough that I can overclock the 390x and the 370, which is something I couldn't do when the 7970 was actually in the case. Now total between my three machines (ref 290, 280x, 7970, 370, 390x) I have broken the 100 hashrate barrier which is a good feeling.
Power here isn't exactly cheap here so thats a risk. Bigger risk is limited used market. ebay isn't really the thing here and with tolls and shipping cost not worth it. Most stuff is sold on "local ebay". Selling 6 GPUs of the same time will lead to lower prices because of high availability. Not kidding. And people are stupid here, AMD has much lower resale value.
But regardless, I'm not that into tinkering around that much to get the stuff up and running and build additional rigs. If the card pays for itself fine but for speculation I just buy ETH with cash (albeit price IMHO is too high now due to much of ETH being sucked into DAO).
I would think the RX 480 should end up somewhere around 35 MH/s, based on the early leaked benchmarks of it between the 390 and 390x.
The catch being it's going to be tough to tell what this will do to ether prices as a ton more cheap mining power pours in, on top of driving down used prices on 380s/380s/290s, etc. Ether prices are so volatile and tough to predict anyway, and this is just another major wrench thrown in. It will be a gamble for sure on whether to invest in more mining to try to cash in, knowing you can get burned if ether prices drop like a rock once truckloads of 480s start churning away.
That's odd. I've had zero issues like that running a headless Windows 10 machine with 4 GPUs that mines automatically on bootup from a batch file. Do you have the same problem if you use ethminer instead of claymore and explicitly state the number of threads?Anyone else have some weird issues with mining on Windows 10 with a headless box? AMD's drivers are pretty iffy on giving me access to the 2nd/3rd graphics card. If there's no monitor plugged in the driver often says the card is "Disabled" and launching claymore (or any OpenCL app) only gives me the primary card.
The mining software is crap really, if it took advantage of newer GPUs the Fury would kill in mining when in reality it's very close to a 390x. Just like console ports all the mining software we have prefer GCN 1.1.
That's odd. I've had zero issues like that running a headless Windows 10 machine with 4 GPUs that mines automatically on bootup from a batch file. Do you have the same problem if you use ethminer instead of claymore and explicitly state the number of threads?
"Crap" is a strong word. It's just not built for mining, that's all. Gaming requirements differ from mining requirements.
Anyway I would assume 20-25MH/s as well. I'm not sure about the exact relationship between memory width and SP with throughput. However, the 512-bit wide interface on the 290/290X looks to be what helps with its high hash rates.
Why is HBM different despite 4096-bit width? It's like the effect of dual channel on single-threaded applications. Single threaded applications are nowadays bound by latency, not bandwidth. It could just be that the way HBM is designed is very different from GDDR5. Perhaps a 1024-bit or 2048-bit GDDR5 would have brought us benefits. Or perhaps there's just a limit on how far of an impact memory makes for hash rates. Just like single threaded performance for CPUs.*
In fact, at certain rates, it behaves almost like single threaded application on a CPU. You can easily increase your hash rates by increasing clock speeds. Perhaps a best hash rate card would have been a 390X on 14nm @ Pascal-like 1.6-1.7GHz.
Anyway I would assume 20-25MH/s as well. I'm not sure about the exact relationship between memory width and SP with throughput. However, the 512-bit wide interface on the 290/290X looks to be what helps with its high hash rates.
Clock speed scaling is quite obvious. It scales everything except the memory.
Perhaps its only memory bottlenecked with some configurations. Also driver seems to play a part. The mininghardware list is really weird.
I'd like to play around with an APU to see how it fares. Man, now I am almost regretting going with Intel since I can't test it.
OK good to know. You went the extra mile so I would just make the most of what you have. I'm staying away from these Sapphire cards, seems to be tons of refurbs on Newegg but without proper voltage controls not really worth it. If the reference blower type come back in stock I'll pick a bunch of them up instead. Milk Crate Miner 3 and maybe 4 on the way. Switched from AMD FX99 platform to Z170 for power savings and should be easier to resell once done. Going to try my first board this week (Core i3-6100 + AsRock Z170 Pro4s). Only five PCIe slots, well technically six if you use the m.2 adapter but I usually only put 4 cards in each milk crate miner anyway. Purchased a new EVGA Superflower Leaded 1200W Platinum to replace my 1KW Platinum as I'm pulling 1030W from the wall now since switching mining to Claymore.
Speaking of claymore. I can confirm my hashing rates seem a little more stable so far but I'm sure if the extra power consumption is worth the hashes. Will run for a week to get a better idea across 6 miners.
Looking like Polaris is going to suck at mining if the rumours are true, so good idea to keep eyes out for cheap 390's or cheap Fury Nano's. Used 290's still the best deal if you can find ones with voltage controls.
Do any online stores except ethereum directly, or do you pretty much need to exchange before you can spend it?
The mining software is crap really,
I'd like to play around with an APU to see how it fares.
I wouldn't be so quick to condemn it. The way it works right now, it works well with old(er) GPU designs that almost anyone can afford new or used. If it were optimized, it might favor the wealthier buyers who can afford multiple Fury/FuryX cards. Or worse, it might present an open invitation to FPGAs and/or ASICs.
The way it is now, it's fairly democratic in how it works. The only objection I have is that it does poorly with newer builds of AMD's drivers.
Not terribly well. My 7700k will mine if I set the frame buffer to max (2Gb) and use the 2Gb card tweaks to let it load the entire DAG, but the most I've ever gotten out of my 384 shaders @ 1028 mhz is about 2.5 MH/s. It's really quite pathetic given that the iGPU burns about 40W while mining.
Since you're talking about motherboards I just bought one these for a new mining rig I'm putting together.
http://www.newegg.ca/Product/Product.aspx?Item=N82E16813157564
Added a cheap pentium to go with it.
http://www.amazon.com/Intel-Celeron...psc=1&redirect=true&ref_=od_aui_detailpages00
Hard to beat this combo for a base mining platform. I was going to go the Skylake route but it would cost at least double for a similar setup.