mrb
Legendary
Offline
Activity: 1512
Merit: 1028
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 09:06:58 AM |
|
They have *your* money, not mine ![Wink](https://bitcointalk.org/Smileys/default/wink.gif)
|
|
|
|
bitpop
Legendary
Offline
Activity: 2912
Merit: 1060
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 09:08:00 AM |
|
No they have yours too
|
|
|
|
Meizirkki
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 09:31:28 AM |
|
I too was looking at the simulation.. Looks more like 125 than 150 but yea, still way too high. KnC =BFL_ng ![Grin](https://bitcointalk.org/Smileys/default/grin.gif)
|
|
|
|
erk
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 10:30:06 AM |
|
Exactly. Even companies like AMD or Nvidia, with the best thermal engineers and ASIC designers on the planet, barely manage to approach 200-250W consumed by the GPU ASIC chip only (even then, a good 50W+ is consumed by the rest of the card, so the GPU ASIC itself is closer to 200W, not 250W).
KnCMiner is so ridiculously underestimating the complexity of their technical choices... There is absolutely no way they will ship a ~250W 100Gh/s chip (their claims) in October 2013. Mark my words.
It is mind-boggling to see the number of people who accept KnCMiner's claims of feasibility without blinking an eye.
KnCMiner will either spectacularly fail to deliver anything. Or they will have to underclock their chips and increase the number of chips per device to match their performance numbers per device (like BFL did with the Single SC).
GPUs have power constraints, look at server grade processors if you want to what's possible. eg. The Intel Itanium (Hondo) CPU is 260watts TDP, a Sun SPARC T4 240watts, and KNCminer are quoting 250watts for the complete Mercury system not just the CPU.
|
|
|
|
erk
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 11:05:50 AM |
|
I too was looking at the simulation.. Looks more like 125 than 150 but yea, still way too high. KnC =BFL_ng ![Grin](https://bitcointalk.org/Smileys/default/grin.gif) Armchair experts again. I am quite sure the foundry and chip makers will tell KNCminer what is too hot based on the properties of the materials they are using.
|
|
|
|
mrb
Legendary
Offline
Activity: 1512
Merit: 1028
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 11:06:54 AM |
|
GPUs have power constraints, look at server grade processors if you want to what's possible. eg. The Intel Itanium (Hondo) CPU is 260watts TDP, a Sun SPARC T4 240watts, and KNCminer are quoting 250watts for the complete Mercury system not just the CPU.
Exactly, you need to be Intel or Oracle (or AMD or Nvidia) to be able to design such a high TDP chip, especially in such a short timeframe. By the way Hondo has 2 dies, not 1; not even Intel could manage 260 watt on a single die... There is a reason neither Intel nor Oracle continued to sell such high TDP chips. They have to bin the top 5% of their CPUs from the foundry, to find the ones that are stable despite the heat, and it's just too expensive and complex to design systems able to power and cool such chips. And in Mercury, at least 80% of the power will be consumed by the chip alone. 20% will be lost to AC/DC and DC/DC power conversion steps, and maybe 1% in ancillary support components.
|
|
|
|
Meizirkki
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 11:11:42 AM |
|
I too was looking at the simulation.. Looks more like 125 than 150 but yea, still way too high. KnC =BFL_ng ![Grin](https://bitcointalk.org/Smileys/default/grin.gif) Armchair experts again. I am quite sure the foundry and chip makers will tell KNCminer what is too hot based on the properties of the materials they are using. my armchair expertise is indeed expecting some bad news from chip makers https://bitcointalk.org/index.php?topic=259319.msg2766019#msg2766019
|
|
|
|
erk
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 11:21:28 AM |
|
GPUs have power constraints, look at server grade processors if you want to what's possible. eg. The Intel Itanium (Hondo) CPU is 260watts TDP, a Sun SPARC T4 240watts, and KNCminer are quoting 250watts for the complete Mercury system not just the CPU.
Exactly, you need to be Intel or Oracle (or AMD or Nvidia) to be able to design such a high TDP chip, especially in such a short timeframe. By the way Hondo has 2 dies, not 1; not even Intel could manage 260 watt on a single die... There is a reason neither Intel nor Oracle continued to sell such high TDP chips. They have to bin the top 5% of their CPUs from the foundry, to find the ones that are stable despite the heat, and it's just too expensive and complex to design systems able to power and cool such chips. And in Mercury, at least 80% of the power will be consumed by the chip alone. 20% will be lost to AC/DC and DC/DC power conversion steps, and maybe 1% in ancillary support components. The number of dies is irrelevant, that's just a cost cutting measure to save having to make another type of wafer. What's relevant is the 12 year old PAC611 package could handle a TDP higher than the total KNCminer system is expected to draw, so it's not venturing into the unknown like you are trying to make out with your armchair FUD effort.
|
|
|
|
mrb
Legendary
Offline
Activity: 1512
Merit: 1028
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 11:46:39 AM |
|
The number of dies is irrelevant, that's just a cost cutting measure to save having to make another type of wafer. What's relevant is the 12 year old PAC611 package could handle a TDP higher than the total KNCminer system is expected to draw, so it's not venturing into the unknown like you are trying to make out with your armchair FUD effort.
Wrong, the PAC611 package cannot handle 260W. (This theoretical number is merely published because Hondo has 2 dies that could in theory draw 130W each.) In practice, Intel had to throttle the dies so that Hondo never draws more than 170W total: The challenge was to design a dual Madison (Itanium 2) processor module that would use no more then a single madison/power pod reference design provided by Intel (ie, 170W).
The mx2 module is designed to run each processor so that it consumes about 65 percent of the max power that it could theoretically draw.
|
|
|
|
plethora
Member
![*](https://bitcointalk.org/Themes/custom1/images/star.gif)
Offline
Activity: 113
Merit: 10
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 12:34:16 PM |
|
rangedriver: Please stay in the music industry. Don't buy a KnC miner.
|
|
|
|
rangedriver
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 12:36:13 PM |
|
rangedriver: Please stay in the music industry. Don't buy a KnC miner.
You know what guys.. I think you're right. Moment of revelation. I'm outta. End.
|
|
|
|
nightengale
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 12:39:50 PM |
|
Wasn't the original device supposed to be water-cooled? Would that make a difference?
|
|
|
|
Nemo1024
Legendary
Offline
Activity: 1680
Merit: 1014
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 12:40:49 PM |
|
People, stay on topic and kindly keep your personal banter to PMs. It's getting tedious receiving new post notifications and opening a topic only to find a completely irrelevant post with some personal insult thrown in someone's direction. Now, where was that Ignore button, I might need to use it more often. In a way, KnC were right in not releasing too much info. Damned if they do, damned if they don't. Wasn't the original device supposed to be water-cooled? Would that make a difference?
I think they always showed air-cooled schematics on their web site.
|
“Dark times lie ahead of us and there will be a time when we must choose between what is easy and what is right.” “We are only as strong as we are united, as weak as we are divided.” “It is important to fight and fight again, and keep fighting, for only then can evil be kept at bay, though never quite eradicated.”
|
|
|
Bitcoinorama
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 01:03:13 PM Last edit: July 20, 2013, 01:13:50 PM by Bitcoinorama |
|
People, stay on topic and kindly keep your personal banter to PMs. It's getting tedious receiving new post notifications and opening a topic only to find a completely irrelevant post with some personal insult thrown in someone's direction. Now, where was that Ignore button, I might need to use it more often. In a way, KnC were right in not releasing too much info. Damned if they do, damned if they don't. Wasn't the original device supposed to be water-cooled? Would that make a difference?
I think they always showed air-cooled schematics on their web site. They could never release enough info to keep peeps happy. They can control what info they keep from competitors. It's hilarious watching trolls (with alternative vested interests?!) fall to pieces. Twitchy bums, much? *I think* with more than a little optimism Marcus and crew are versed and skilled enough to prepare for such eventualities as heat. The huge mf heatsink they are using is overkill. He said this. Also 1000w is max, again, overkill to ensure it works. There is no watercooling as hosting sites won't permit water cooling in data centres. Watercooled options will be developed for consumers at a later date if you want less heat and noise. Still, please continue. Observation of sudden negative armchair 'experts' with 'professional' insight makes for an interesting take as to why you prowl upon a thread you have no vested interest in...or do you?? (Note: none of this is directed at nightingale or nemo1024) ![Roll Eyes](https://bitcointalk.org/Smileys/default/rolleyes.gif)
|
Make my day! Say thanks if you found me helpful ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) BTC Address ---> 1487ThaKjezGA6SiE8fvGcxbgJJu6XWtZp
|
|
|
erk
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 01:15:19 PM |
|
Wasn't the original device supposed to be water-cooled? Would that make a difference?
Water cooling is not anymore efficient, unless the radiator is passive convection cooled. Most setups have pump and fans which consume power, and kind of spoil the efficiency. Good for tight areas though. I can see why WC would be handy on a GPU mining rig where the cards obstruct airflow from each other. The KNCminer drawings show it's definitely air cooling with a large fan on each ASIC heat sink, plenty of room for fans, and the ASIC modules seem to be placed so they don't block each others airflow.
|
|
|
|
ujka
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 01:21:44 PM |
|
MarsThis device will be the first mass produced product and is available for purchase in the next few days. The key details around this device are as follows: Over 6GH/s 48x IC CYCLONE IV FPGA 115K on board chips See the sites below for more details on the chips we will be using in our Mars product. 30% Improvement in hashing rate over standard Implementation Fan cooled with 48 heat sync’s and fans plus two case fans producing a cooling tunnel across the chips Modular design (we will carry this design throughout all of our products) USB attachment (software will be provided) Shipment begins on mass in just over 2 months. Earn coins while you are waiting in the queue for Jupiter External ATX power will be required (we will provide a recommended list) Upgrade protection built in to any Mars purchase with vouchers for up to $2000 off of any Jupiter purchase. Last but not least, any purchase of the Mars product automatically reserves you the same place in the order queue for the Jupiter product. The queue reservation will last 30 days. For more information about the chips used you can use the following links http://www.buyaltera.com/scripts/partsearch.dll?Detail&name=544-1468-ND http://www.digikey.com/product-search/en?x=18&y=12&lang=en&site=us&KeyWords=EP4CE115F23C8NJupiterThis device is being designed in parallel by the OrSoC engineers. It will offer industry leading performance and power consumption per GH. The final specifications for this device are being ironed out now but we can confirm the following: Minimum 250GH/s per device 7000-8000 USD predicted price 48 Chips (which are being designed exclusively for KnCMiner by ORSoC). An additional gain of 30% more hashing when the advanced algorithms provided by ORSoC are applied. Water cooled setup. Modular Design allowing for expansion at a later date. Embedded Linux device to allow for standalone mining. Shipment begins in early autumn.
|
|
|
|
Kuroth
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 01:23:05 PM |
|
5 Month old Artical.... Silly Trolls.. I too was looking at the simulation.. Looks more like 125 than 150 but yea, still way too high. KnC =BFL_ng ![Grin](https://bitcointalk.org/Smileys/default/grin.gif) Armchair experts again. I am quite sure the foundry and chip makers will tell KNCminer what is too hot based on the properties of the materials they are using. my armchair expertise is indeed expecting some bad news from chip makers https://bitcointalk.org/index.php?topic=259319.msg2766019#msg2766019
|
|
|
|
SirWizz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 01:29:55 PM |
|
More dog poo... I think some of the folks here can not interpret the graphics correctly. I can not either because I don't know if it shows a single core simulation or the whole chip and frankly it is not my current area of expertise however, I highly doubt KNC would release an OrSoc simulation that shows they are unable to understand basic thermal chip simulation and therefore properly account for the heat that needs to be dissipated.
Based on the 4-core design and depending on how many engines are running in each core I would expect the chip thermals to be spread fairly evenly across the chip. Also, this being a big chip you get more real estate for dissipating the heat. So, let's dumb it down for the rest of us, if you partition this chip into four sections and assume 60W of heat needs to be dissipated on an area of 765 mm^2, I would not think that is outrageous at all, in fact it is quite reasonable (I remember that some of the old Thunderbird AMD chips had a die size of around 120 mm^2 and were dissipating 60-70W of heat).
|
|
|
|
erk
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 01:31:01 PM |
|
MarsThis device will be the first mass produced product and is available for purchase in the next few days. The key details around this device are as follows: Over 6GH/s 48x IC CYCLONE IV FPGA 115K on board chips See the sites below for more details on the chips we will be using in our Mars product. 30% Improvement in hashing rate over standard Implementation Fan cooled with 48 heat sync’s and fans plus two case fans producing a cooling tunnel across the chips Modular design (we will carry this design throughout all of our products) USB attachment (software will be provided) Shipment begins on mass in just over 2 months. Earn coins while you are waiting in the queue for Jupiter External ATX power will be required (we will provide a recommended list) Upgrade protection built in to any Mars purchase with vouchers for up to $2000 off of any Jupiter purchase. Last but not least, any purchase of the Mars product automatically reserves you the same place in the order queue for the Jupiter product. The queue reservation will last 30 days. For more information about the chips used you can use the following links http://www.buyaltera.com/scripts/partsearch.dll?Detail&name=544-1468-ND http://www.digikey.com/product-search/en?x=18&y=12&lang=en&site=us&KeyWords=EP4CE115F23C8NJupiterThis device is being designed in parallel by the OrSoC engineers. It will offer industry leading performance and power consumption per GH. The final specifications for this device are being ironed out now but we can confirm the following: Minimum 250GH/s per device 7000-8000 USD predicted price 48 Chips (which are being designed exclusively for KnCMiner by ORSoC). An additional gain of 30% more hashing when the advanced algorithms provided by ORSoC are applied. Water cooled setup. Modular Design allowing for expansion at a later date. Embedded Linux device to allow for standalone mining. Shipment begins in early autumn. Those specs are nothing like what you can order, so I would disregard them totally. For starters the Jupiter is 400GH/s not 250GH/s. Secondly it's 4 chips not 48.
|
|
|
|
ujka
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
July 20, 2013, 01:38:08 PM |
|
Sorry, that was an answer to the guy asking about water cooling in 'original design'. And yes, first published spec. were with water cooling.
|
|
|
|
|