Bitcoin Forum

.

the fury with HBM memory could probably be very interesting with memory hard algo

yes the nano seems like an interesting card, will probably get one for mining.  :)

Does this 2X performance/watt applies only to Nano or all Fury series?

Let wish those enhancement is not limited to games.

The R9 290(x) does not have per core mining performance increase over 79xx.

I mean per core performance. The performance of 290x is about 2816/2048= 1.4 times, not higher than 7970. I hope Fury XT performance is 4 times of 7970 with double core count (4096/2048=2).

Thanks for the post. I will be following the cards release. I have been thinking about busting out some of my old GPU rigs. :)

is there a set date for when we will be seeing the fury cards in stores? Read somewhere that they will be in short supply at the release.

Note that the stolen wolf0 binaries for x11,x13,x15 binaries will most likely not work on the new hardware, so you are stuck with the opensourceminer. I doubt that wolf0 will be kind enough to compile new binaries and hand them out for free, but he will probobly sell them.

The current speed of the opensourceminer on the 280x is around:

x11: 4,4mhash/s
x13: 3Mhash/s
x15: 2.5mhash/s
quark: 2mhash/s
qubit: 3.5mhash/s

wolf0's private AMD kernal speeds on the 280x (in MHASH/s):

x11: 6,5 mhash/s (1.47x faster)
x13: 5,0 mhash/s (1.66x faster)
x15: 4.5 mhash/s (1.50x faster)
quark: 11 mhash/s (5.50x faster)
qubit: 10 mhash/s (2.80x faster

My private AMD kernal(more speed and kernals are coming):

quark: 4,5mhash/s (2.25x faster)

NVIDIA sp-mod(open source):

quark: 980ti (23MHASH)
          980 (19MHASH)
          970 (15.5MHASH)
          960 (10MHASH)
          750ti (6MHASH)

A Fury X mining quark with the opensource kernal will be slower than a 7950 with the wolf0 kernal.. and 3x slower than the NVIDIA 980ti

I doubt that they will be slow, you are talking about something that is not seen or tested come here when then can claim that they are slower

Without optimized code and a bether compiler the new cards will mine slow. this is simple math.

If the new card is 4x faster than the 280x, it will still mine slower than the 280x without a optimized miner.
Wolf0's quark implementation is 5.5.x faster than the opensource miner in the quark algorithm.

yes, they use more power.. The problem is that the optimal quark kernal needs a modified sgminer in order to work. The modified bin files is not enough. My slow 1.25x faster kernal will work on the old sgminer, and I can generate hawaii bins..
Wolf0 hasn't leaked his quark work yet I think, but there is a Russian? kernel/sgminer windows exefile out there, but without hawaii bins. only tahiti.

The current global hashrate on the quarkbased coins is around 150GHASH/s. This is around 75 000 Radeon 280x gpu's using the opensource miner. But the sharks use private kernals that are 5-6 times faster... The rental sites are currently paying around 0.5 BTC per GHASH..
This meens that around 75BTC ($18,750) is mined per day in the quark algo alone..

This is $6 843 750 per year in mining revenue.

Enough to power one of these:

http://www.barnorama.com/wp-content/images/2013/02/bitcoin_farms/02-bitcoin_farms.jpg

The performance between 980ti and 970 is 23/15.5=1.48. The processor difference is 2816/1664=1.69
The performance between 980 and 970 is 19/15.5=1.22. The processor difference is 2048/1664=1.23

Why 980Ti is less efficient per core? Is that limited by power?

These numbers are abit inaccurate because the highend cards are running on lower default clockrates.

An overclocked 1mb 750 can be overclocked stable  run at almost 6mhash/s quark wiith only 512 shaders.

Here is lyra2re on a 1gb 750 with 512 shaders using ccminer sp-mod  release 52:

The clock is 1510/1620(*2)


https://ip.bitcointalk.org/?u=https%3A%2F%2Flh3.googleusercontent.com%2F-S5J8reDvZdo%2FVXlJ3O531oI%2FAAAAAAAAAzQ%2FlCN7kASIXD4%2Fs800%2Flyrarec.png&t=553&c=kaTZvW4XJ4tY-g

I don't know from where you are getting those number tbh, my card never did anything like with your version... it was at best 770kh/s overclocked
(anyway still far away from my new version with my "bad card"  ;D 1140kh/s)

core 1500MH/s 3200MH/s, well don't even think my card can do that...

It's not a 750ti card. It's a 750 with 1gb of memory , 512 shaders overvolted and heavily overclocked. The user Rednow posted it in my thread some time back.

the price difference is (or was) about 20-30euros so not much, but 1gb is a no go for mem hard algos (which is a bit surprising here with lyra) and the overclocked is mind blowing (probably gpu blowing too...) it is something like 200MHz higher (both core and mem) than what I use...  (the bios might have been moded too)

So...?
Any news?

Cant seem to find anything related to R9 fury and mining.

I got an R9 390x this weekend.

My first impression.

I ran it next to a Sapphire R9 290X 8Gb Mem, hash rate was almost exactly the same. If you didn't know they were different, you wouldn't have noticed.

What it "did" very well, and yes, it "did" it.. haha, was OC, Sapphire is known to be able to OC very good.
This was no exception.

+ that it stayed almost 10degrees cooler then the 290X
Maybe that's a coincidence, maybe not.

10 min after i did some tests, i rebooted and it never came alive again.

So for those interested in buying a 390x. Think Twice... Buy a 290x, which does the same thing but cheaper!


As soon as The Fury X arrives (next week it seems) i'll be posting my findings.


Greetings

if you don't measure power usage, you might not find big difference, also if you just tried an algo might not be significant either...

This is a bit thin, what you say here (and I am not defending amd or else... as I am rather an nvidia guy...).

Yes you're right, tested it on 3 algo's, quark, X11 and Sia

Difference was close to nothing on all 3 algo's

I should get me a new meter to see how much electricity it was pulling.

I'll order one tomorrow :-)

wow  :o
I have VGA more than that but the size is really big

try also a memory hard algo (with the fury X), those algo doesn't use any memory so basically it is just the performance at doing calculation. Not sure it should really change a lot, they just depend on the number of core/shader/thread

What hashrate do you get in  the quark algo on the 390x?

AMD Radeon Fury X : Fiji XT   
AMD Radeon Fury: Fiji PRO   
AMD Radeon R9 390X: Hawaii XT
AMD Radeon R9 290X: Hawaii XT

It looks like wolf0 can make some more Bitcoins. the furyx and fury chips are probobly not compatible with the hawaii bin files..
Without wolf's bin's the furyx will run x11 slower than the 7970


The 7970 is a 4 year old design (the first cards came around Jan 9, 2012)

A single 980ti is aleady doing 23MHASH on the default clocks. (quark)

http://cryptomining-blog.com/wp-content/uploads/2015/05/ccminer-1-5-51-sp-mod-580x293.jpg

Fury X will beat that :p

Got a new 390x today, it's running fine now

Will order a meter and post my findings later this week.


Greetings

I have no idea how you guys insist on top-end cards. They have a well known premium.
It looks to me the card to buy is 380. It seems very affordable around here and Tonga has byte-lookup instructions. With recompiled kernels there could be a nice speedup (or not...).

As a side note: 7750 can do well in excess of 2 qubit MHS with publicly available kernels. Users able to run it on big cards (I don't remember which card) told me they go over 10M. I'm inclined to believe another 50% is possible.

I have working hawaii bins that will work on the new cards for qubit and quark.

Around 15MHASH +++ on the 390x.

Please donate 0.1 BTC to my BTC adress and I will tell you how to do it.

Who cares. The shelf price is at least 30% lower here.

So, maybe it's the same thing to you. It isn't to me.

http://cryptomining-blog.com/5201-trying-out-the-new-amd-radeon-r9-fury-x-for-crypto-currency-mining/

The Quark and qubit performance was pretty good, the rest needs some more work... Lyra2RE and neoscrypt is slooow...

AMD Radeon R9 Fury X Results:
– X11 default: 6.778 MHS
– X11 Wolf0 Mod: 8.123 MHS (closed source)
– X13 default: 5.614 MHS
– X13 Wolf0 Mod: 7.176 MHS (closed source)
– X15 default: 4.69 MHS
– X15 Wolf0 Mod: 6.335 MHS (closed source)
– Quark modified: 22.37 MHS (windows only, closed sgminer and kernal sources)
– Qubit modified: 21.15 MHS (windows only, closed sgminer and kernal sources)
– Neoscrypt default: 147 KHS
– Lyra2RE default: 287 KHS
– Lyra2RE Pallas Mod: 450 KHS

Without the private kernals (only using  the opensource kernals), the furyx perfoms a bit slower than an overclocked 280x.
Lyra and neoscrypt is dead slow...

AMD Radeon R9 Fury X Results:
– X11 default: 6.778 MHS
– X13 default: 5.614 MHS
– X15 default: 4.69 MHS
– Neoscrypt default: 147 KHS
– Lyra2RE default: 287 KHS
– Lyra2RE Pallas Mod: 450 KHS

quark: ?
qubit: ?

there are most likely doing something wrong  ;D
they need to play a bit with parameter tuning... technically the high memory bandwidth should allow a lot faster than the 290x and most likely anything around

lyra2re depends much on latency, so the added bandwidth may not be of help, especially if they traded bandwidth for latency (did they?) ;-)
Results from algos which do not use global memory access are more interesting than those, IMHO: we could see if the "same wattage but more shaders" equation is true.

The miners need some more free DJM34 hardcore coding. Can you do it? I will donate 0.02BTC ;)

sure if someone is willing to donate a fury X first  ;D

I would expect smaller latencies (assuming it is related...) as the memory is pretty close  to the gpu

According to this page:

http://support.amd.com/en-us/download

The drivers for radeon 300 and fury series are available for windows 8.1/7 64bit only.
I'm wondering if they work correctly with the standard driver on linux and what about opencl.
Let's hope they upload the linux version soon!

lmao you won't be convincing pallas to switch to Windows anytime soon

those linux geeks are so close-minded! :-D

I have an esternal drive booting to win 7. Good for compiling miners or kernels ;-)

Back to the topic: any new fury x mining hashrates available?

Nice lipstick.

Regarding the new 390X:

It's now pulling 267,5 Watts at full load, Clocks: 1080/1500, Fans @ 100%, 35C outside, card is at 65C

Haven't had time yet to do more tests but will definitely do soon.

Will update this post when i do.

Greetings


Edit:

Here's a picture of the wattage when running at full load with 5 cards. (Mixed Rig)

https://drive.google.com/file/d/0B_cBBoZ_PTwOYVd4Sk92ZnJFdDd6b2FlbjFubm5vSWtXSGY4

A 390X, 2x 290(regular) and 2x 280x


@Pallas

I might just do a test for you :)

Give me a few minutes!


Edit 2:

So the wattage i said earlier was at full load when mining Sia.

Here's the wattage when mining Quark.. serious difference i might say
https://drive.google.com/file/d/0B_cBBoZ_PTwOZHEwWXNaa3NRZVFQQUI4STFfU2luQUExeFhv
So let's say 75 Watts for Mobo, proc, mem, etc.. (including Fans connected to mobo) , so around 250 Watts for the card at full load. Give or take..

Also here's a screenshot with some more info
http://s15.postimg.org/hi31rrky3/quark390x.png

Will get some testresults of one of my 290x's soon. They're in another rig and don't want to get them out now. :)

thanks for the info.
is it less or equal to a 290x with the same configuration?

It should be the exact same.  It's the same card essentially overclocked.  Doesn't even rename the bin to Grenada it still works on under the Hawaii name.

I sold a couple of r9 280x and I picked up a cheap r9 nano. I wonder if the 4096 shaders@175w can be used for mining. :)

cheers

I did some tests and found it behaves pretty much like Hawaii, maybe a bit less power usage.
But a new driver just came out so the tests need to be done again, as the new driver may include an updated compiler (which is not uncommon when a new chip is out).
Maybe someone tested the new driver already?

I got it new for $500 with free shipping(black friday sale). Will probobly never ROI, but I need some more challenges. Furyx users renamed wolf0's hawaii bins and upgraded to the latest driver to get a boost. (x11)

ccminer Sp-mod release 74 is doing

26.5 MHASH on the standard clocked gigabyte windforce G1 980ti (8pin+8pin power)
20 MHASH on the standard clocked zotac amph 980 (8pin+6pin power)
17.5 MHASH on the standard clocked gigabyte windforce G1 970  (8pin+6pin power)
10.7MHASH on the  standard clocked asus swix 960 (6pin power)
8.5MHASH on the standard clockes EVGA SCC 950  (6pin power)
6.25MHASH on the standard clocked asus swix 750ti (no external power)
5.1MHASH on the standard clocked 1gb gigbyte 750  (no external power)

That's very interesting sp, i own a few fury's and i'm very interested in what you can achieve.

I've done some tests on few algo's a few months ago. You can see the results here. (you probably remember those as you were active in that thread as well..)

Quark: https://bitcointalk.org/index.php?topic=1116027.msg11993624#msg11993624
X11: https://bitcointalk.org/index.php?topic=1116027.msg11880191#msg11880191
X13: https://bitcointalk.org/index.php?topic=1116027.msg11953566#msg11953566

I'm curious!

Greetings

Tried it now. Fiji is indeed slightly faster, but the new driver doesn't let me change clocks on the 290x any longer! :-(

Interesting.

Going to try that out later today.

What OS is that Pallas?

Ubuntu 15.04 64bit.
I'll add slightly faster and slightly smaller fiji binary, not sure if it changes anything without recompiling the kernel.
It looks like it doesn't detect the 290x properly, as it's the only card which "hasn't got a name":

Adapter 0 - AMD Radeon (TM) R9 Fury Series
Adapter 1 - Supported device 67B0          <=============
Adapter 2 - AMD Radeon R9 200 Series

That's probably the reason why you're not able to change clocks on that 290x.

I'm going to do a test run on windows to see what it does over there.

Will update as soon as possible :), i'll throw in some power usage as well.

Fury cards may be good for X11 or Quark. It is not good for Ethereum mining. The memory speed is not high even though the bandwidth is high.

The kernels and/or compiler may improve for Fiji chips in the future, meanwhile the non-ram based algos are performing just fine, very similar to hawaii.

Most GPUs are mining Ethereum at the moment. The total hash rate is 500G. The 280X speed is about 25MH/s. So there are about 20,000 280X equivalent cards there. It is much more than any other coins added up.

Since HBM should have lower latency as well, leading to potentially better random access to memory as needed by eth mining, I see no reason why an optimised opencl compiler (and maybe opencl code) couldn't bring nice performance to the fury as well.