Yes. Anybody got an estimate for how many MHash/s this can pull while OCed ?
ArtForz was a pro at that kind of stuff ...
The current trend is to clock for max efficiency instead of max hash rate. Make up for the hash rate with moar GPUs  I care about MH/$ and density. Efficiency : I am not very concerned about that. |
|
|
|
|
Yes. Anybody got an estimate for how many MHash/s this can pull while OCed ?
ArtForz was a pro at that kind of stuff ...
|
|
|
|
THE OCTOBAMT CHALLENGE
Spent some time trying to help LittleDuke get 8 GPUs working in bamt. Also reviewed this thread and a found everyone else who has tried 8 got the same results we did: It doesn't work.
7 GPUs = totally fine
8 GPUs = the ati driver crashes, machine locks up, game over
Pretty sure more than 8 is impossible on any linux box, but it would be nice to get that 8th slot working for people who cram 4 5970s into a box etc.
I don't have any machines anything like that, so testing things is quite difficult here. If you do have a rig that can support 8 GPUs, I hearby invite you to enter the first annual OCTOBAMT challenge!
The rules are simple: Using any tool or technique, get BAMT working with 8 GPUs. The solution must be something we can apply to the standard BAMT image without reducing the existing BAMT functionality, otherwise anything goes.
The first person to complete the challenge will be awarded a fantastic prize package which includes unspecified items of practically undefinable value. You will be as a god among men, celebrated and feared throughout the land. Don't let this amazing opportunity pass you by.
Hints:
My suspicion is that the drivers we use are just broken. Limited experiments with other versions have revealed that certain versions, such as 10.11 and 11.12 behave differently than our version (11.6 iirc). Other versions, 11.7-11.10 seem to lock up just the same as 11.6.
However, 10.11 and 11.12 seem to introduce their own problems, which may or may not be something that can be resolved.
The challenge has been completed, I have seen the light and it is BAMT with 8 GPUs descending from heaven. I'll post more details on that later, but an issue has come up... Unfortunately, there doesn't seem to be any way to get more than 7 working with the current 32bit version. A 64bit kernel/driver is what makes 8 gpus happy. So.. who's going to be sad if we abandon 32 bit machines? Who's going to be sad if we never support 8 gpus (i.e. go to 64bit)? It's one or the other, as far as I am concerned. No freakin way I am maintaining two builds. Speak now, if you care. PS - please no "omg i want to mine this other thing that needs 64 bit blah". I don't care about alternate and scam chains ever working in bamt. This is about supporting the people who want to use it for the things it is designed to do. If you do have 8 GPUs, or if you do have hardware that requires 32bits, I'd love to know. I think the number of people running 8 cards on a single rig is vastly outnumbered by the people running 32bit only CPUs so I think stick to i686 for now. The fact BAMT supports i686 is a major reason to go for that and avoid Linuxcoin ( which is amd64 only and does not work on older machines ). If BAMT goes to amd64 as well then what are we to use for those older machines that are still capable of mining but just lack a amd64 capable CPU ? Certainly not Winblows. Thank you for the hard work and I certainly hope BAMT will stay i686 ( if it ain't broke don't fix it ) as it is now ! |
|
|
|
Yeah. I am pretty sure they launched this like last year and with some impossible specs to ward off any potential competitors from developing anymore. But that fully custom ASIC miner that gets 10 ghash/s at 10W and costs $100 is still on the horizon so BFL beware  ! |
|
|
|
|
OK.
So what are the VRMs that will get very hot on a 5870 reference that is mining at 960 core and 300 memory for example ?
Thanks ArtForz !
|
|
|
|
|
LOL. At least we got the chips data !
|
|
|
|
5970/5850 3 phase, 5870 4 phase gpu. The memory phase is located above the gpu phase on 5870. You can undervolt the memory controller using GPUTool. .95v is idle default.  Nice. I will check that out for my workstation. The bad news is my other 4 rigs are linux and GPUTool doesn't support Linux.  Yeah. We need this tool for Linux. Maybe it can be done with the BIOS as well  |
|
|
|
What is the meaning of this ?
BitcoinHorn Bitcoin Horn
5HyEbr5DEiKmJUTR2FPR75QLX1wNHgbhZS95JZQKWjcLndm2hHU #bitcoin
15 minutes ago
Thanks !
That looks to me like a WIF privkey. You can try claiming it with something like blockchain.info's My Wallet, or importing it into your wallet.dat with pywallet? Yeah. Figured that out myself but it seems like it is already claimed |
|
|
|
|
What is the meaning of this ?
BitcoinHorn Bitcoin Horn
5HyEbr5DEiKmJUTR2FPR75QLX1wNHgbhZS95JZQKWjcLndm2hHU #bitcoin
15 minutes ago
Thanks !
|
|
|
|
I think all of you are so concerned with BFL technology because you are located in China and wanna counterfeit and replicate their design to sell for cheaper  If they got such a great deal on 65nm FPGA why not sell it straight away at normal market price ? more profit the reason it was cheap is probably because nobody wants it anymore except miners Yeah, I kinda figured that one out but it still amazes me the insisting people on here that want to copy BFL's design  |
|
|
|
I think all of you are so concerned with BFL technology because you are located in China and wanna counterfeit and replicate their design to sell for cheaper If they got such a great deal on 65nm FPGA why not sell it straight away at normal market price ? |
|
|
|
Since everyone seems to be farting out ridiculous 'social sharing' that people treat like blogs, and blogs that people treat like their own personal diaries, I thought I might capitalize on the entire thing by cutting to the chase (people love honesty!) Introducing  I'm selling shares now for $1,000,000USD per share, with a trillion shares available. Is everyone ready to get in on this bleeding edge synergy? OK. How can I buy !? |
|
|
|
I am 100% sure Bitscalper is 100% genuine and mybitcoin will never happen again as people have learnt their lessons. Put all your money into it and you will get rich  ! After all, this is the community that thinks one can make a ton of money while said one is sleeping In the end we are all getting scalped anyway so why not deposit some BTC now  |
|
|
|
take a stock clocked GTX590 for example. it uses about 370W on full load, and with my own cuda miner, it puts about 470MH/s. with pocl it's just under 200.
given that the 590 is comparable to a 6990(price, wattage, dual-GPU), I'd say it meets your bet!
Shenanigans. Code or it didn't happen. And a 6990 will still smoke those numbers. Yeah I think we can call BS on this one. GTX 590 doing almost 500 mhash/s ? LOL |
|
|
|
on ref 5870 that VRM slave with a missing brother in the top left is the VDDCI, its vt1165 controller is in the extreme top left corner. memory VDD is the 2 above the missing 5th VDDC phase on the right (looks like those 2 are actually set up to run as a 2-phase... hard to tell from tiny pics) on nonrefs... well, anything can be anywhere edit: nope, those 2 are 2 single vt243s, most likely memory VDDC and VDDQ Yes I am interested in this VRM. What does it do exactly ? |
|
|
|
Can you use the Win32DiskImager.exe tool to re-image a fully patched. .04b before customizing the config files? I'm just trying to save time with having to make 10+ USB keys for all the rigs. Maybe you have a .05 image coming out soon?
Thank you.
Yes you can and should.. I highly recommend to update and then read the image back, then use that for making keys. Save even more time by preconfiguring as much as you can on your master image (bamt.conf and pools file, etc). If your rigs use mostly the same hardware, you can make an image that is practically ready to go. And yes there is a new image coming "soon". Not sure how soon. Yeah a new image would be very nice indeed but if we can use the fixer and just read the image off the USB with dd like we put it there in the first place to make another fixed image then it is not that bad. I really like the fact that this is a hybrid image which gives us a lot of options ( no need to be messing with Unnetbootin or any other such tools just use dd ) ! Thank you for the updates and great work ! |
|
|
|
A little more VRM pr0n.  If you look closely, you'll find several more VRMs on the 5970 first the obvious ones: VRM1 area = VT1165 + 3 slaves, GPU1 VDDC (i2c slave addr 0x70 on gpu1) VRM2 area = VT1165 + 3 slaves, GPU2 VDDC (i2c slave addr 0x70 on gpu2) the square chip at the top/right of vrm2 area = another vt1165 controller, belongs to the 2 slaves above the VRM2 slaves, it's a 2-phase providing both GPUs VDDCI (i2c slave addr 0x71 on gpu1) now, there's avolterra integrated controller (vt237) top/right of the 1 of "VRM1" and another vt237 under the R of "VRM2" I suspect those 2 guys are our memory VCC (one for each GPUs memory?). now on to the non-volterra stuff a AOZ1024D above the VDDCI inductors and output caps... no clue. the 2 FETs below the GPU2 memory chips/left of the suspected GPU2 memory VRM, controller is a uP6101 on the backside ... again, random aux supply. that should be it for major switchers, there's probably another 3 or 4 linear regs strewn about... What about the leftmost VRM on the 5870s ? What is that used for ? What about the 2 smaller VRMs on the right side above ? What about the 4 bigger VRMs near the inductor thingy ? Thanks ! |
|
|
|
It would be "weird" (but not impossible) for an aftermarket cooler to not cool the memory VRM as you would just move the bottleneck from GPU to memory. Yes that is clearly a design error by the aftermarket cooler producer. Check this heatsink ( little red on on the left ) made specifically to cover that memory VRM on the 5870 and the nearest memory chip !  Picture of the VRM locations on a reference 5870 PCB  |
|
|
|
Is that a 5870 PCB but it doesn't look "right" for a 5970? Still I am 99% sure regardless of where it physically is located all AMD card use a different bank of VRMs for memory voltage because they run at a different voltage.
Actually that is a picture from a reference 5850 card I found on the net and edited myself to show the VRM location. I have checked on one of my 5870 that the VRM is also there on the reference 5870s. It is possible but if it there is only 1 VRM = 1 phase power w/ no external inductor. I doubt that would be sufficient to power the memory. On 5850 maybe but 5870s have up to 1GB of memory clocked very high. I would imagine it needs a beefier setup. See above for my theory on VDDC system for 5970. I will look at 5870 PCB and see if there are any other alternatives. Yes I am very interested what that left single VRM does but as far as I can see there is no inductor but there is another one of those Volterra controllers ( 2 in total for a 5870 ) along the single VRM next to the memory chips. |
|
|
|
Is that a 5870 PCB but it doesn't look "right" for a 5970? Still I am 99% sure regardless of where it physically is located all AMD card use a different bank of VRMs for memory voltage because they run at a different voltage.
Actually that is a picture from a reference 5850 card I found on the net and edited myself to show the VRM location. I have checked on one of my 5870 that the VRM is also there on the reference 5870s. I have no 5970s so I cannot check first hand. |
|
|
|
|
Show Posts
