GA-EP43T-UD3L SpeedFan Issue

Hi all.

Just a question inquiring about a SpeedFan (Latest Version) Issue that i have noticed.

My GA-EP43T-UD3L Rev 1.0 (Bios Version F5) Seems To have a Temp issue On one of the sensors.

During stability testing (3.33Ghz Intel E7600 & GB Samsung RAM 8-8-8-22 (CPU-Z) (333 x 10 FSB : DRAM 1:2)) I am getting a Temperature warning from the MOBO's Speaker Which is set to go off @ 70'C

The interesting thing is that its not the core temps setting it off, its some other Mobo Sensor IT8718 Which I have Labeled North bridge. as you can imagine this has gotten me Rather Frustrated. Core temps are stable, voltages are stable but the MOBO is whacking out. :pfff:

Screen shots Click to enlarge images

I tried To bring down Temps By attaching A 45mm Fan to the NB With no avail, it is Pulling Air through the NB

Anybody have any ideas on what this temp Reading is Actually Representing?
Why is it so high? And is there anything i can do (Besides Water cooling) To cool it down?

Thank you in advance

PS. What is TJ MAX? :heink:

Message edited by fisshy on 05-06-2010 at 01:11:48 PM

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

Quote :

some other Mobo Sensor IT8718 Which I have Labeled North bridge

The IT8718 is the board's LPCIO chip (low pin count input/output), and is responsible for a huge number of funtions, among which are 'off-CPU' temperature measurements, floppy operation, fan control, PS2 kbd/rodent hook-up, com ports, and on and on... We ( use the 'royal' we - meaning I, nor anyone I know) don't know the exact position of the sensor - the closest we've been able to determine is "somewhere above the right end of the primary PCIeX16 slot, which puts it near or under the northbridge"... Here is something pertinent from GB tech:

Quote :

"CPU_FAN connector needs a CPU cooler with four wires were the PWM modulator is included in the cooler. The mobo senses the CPU temperature and managing the pin 4 signal, the cooler's fan will automatically be increased or decreased, following the CPU temp. Fan's rotation feedback returns via pin 3 for BIOS alarm feature.

SYS_FAN2 connector needs any fan, with three or four wires (don't care) since the speed is controlled via on-board PWM controller that supply pulse-width-modulated 12v for the fan via pin 2. So, if using a four wire fan, its internal PWM is locked to full speed (this explain that fixed 5v on pin 4). This time, the mobo senses the North Bridge (in your case the P43) temperature and will increase or decrease the speed of the fan connected in this socket. Feedback is sent to pin 3 for BIOS alarm purposes.

SYS_FAN1 and PWR_FAN are three pin socket with fixed full speed fan. No controlled rotation at all from the mobo. Rotation feedback also monitored via pin 3 only for BIOS alarm purpose."

Bridge chips are made to 'run hot' - most have a Tmax at, or above, 100°C; that does certainly not mean that they benefit from running hot - indeed, if the memory controller is run fast, overvolted, has overvolted memory, or more than one DIMM per channel, or various combinations of these factors, it can always benefit from 'active cooling'!

You might want to take a look at the third last post here:
http://www.tomshardware.com/forum/ [...] d3l-issues
He, thankfully, found the problem without much effort, but the "semi-locked retainer, missing spring, bent sink assembly" keeping the NB from properly 'spilling' heat is always a possibilty in these situations...

Message quoted 1 times
Message edited by bilbat on 05-07-2010 at 02:46:44 AM

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

bilbat wrote :

Hmmm - odd, very odd. Try having a look at this: your northbridge sink is the big one, right next to your CPU. It is held in place by a pair of 'spread-pins' that snap through the board, and underneath the top 'heads' of the pins should be a couple little springs - try to get a look at 'em (dental mirror might help...) and see if either one looks damaged, and whether the sink itself is 'down' squarely on the bridge itself... You're only running two dimms, and they should be 'installing' (LoadOpt) at the stock JEDEC 1.8 - shouldn't be enough to run the bridge very hot, but your symptoms are exactly what happens when overclocking/overvolting a bridge (say, 4 DIMMs @ 2.1V and 1066) - runs OK 'till you 'push it'... I'm wondering if somehow the sink isn't being held in flat, complete contact with the bridge's thermal surface? Next step, my guess, might be to pull the sink (by carefully 'squeezing' the bottoms of the pins to release them [and be careful! don't let the eetsy-beetsy springs fly off into the room - or you'll be crawling on the floor for three hours - ask me how I know :pt1cable: ] with a needle-nose...) to get a look at the thermal pad beneath... I always pull 'em, toss the thermal pads, lap the sinks, and use TIM paste on 'em, but then you've got to replace the pins with some nylon 4-40 screws, as the 'stack height' is smaller with the pad gone...

I'm taking you meant this post?
it Makes sense to me, it did seem to be on a little loose when i was attaching the 45mm fan.
i didn't check the pins, i just pushed down the cooler firmly and used a jewlers screwdriver to apply pressure to the center of the pins located inside the cooler with no effect

I would re-seat the cooler but as my general rule I only re-seat any cooling component with Fresh TIM. Since the only TIM i have at the moment is uber expensive HP server solution enterprise grade stuff, I'm kind of stuck...

Do you thing Lapping the NB would really decrease temps that much?

Thanks for your reply
are you like some official Gigabyte techie?

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

Quote :

Do you think Lapping the NB would really decrease temps that much?

Nah - lapping is only good for a couple degrees; I just do it 'cause I'm freaking obsessive [:fixitbil:9]

I'm mainly suspicious of 'thermal pads'; twice now (never on a GB board BTW) I've found 'pad problems' with 'misbehaving' graphics cards; once, had a totally missing pad on a voltage regulator; second time, found a pad 'folded over' diagonally on a VRAM chip, which caused enough of a 'stack height' problem to cause the sink to not contact the row for about two chips in either direction...

Now that I've seen some VidCard cooling 'oopses', I'm scanning the news daily, waiting for the first heatsink failure on an nVidia GTX-480 to cause a 'China Syndrome'-type 'meltdown', and take a small neighborhood with it into a molten hole in the ground [:bilbat:9]

I also have some concerns about the mounting 'hold-down' springs. Some years back, I did a high-speed assembly machine that made neon sign tube mounts. They are a little gizmo, sort of like a shock-absorber, that have a spring-loaded plunger with a u-shaped clip on the plunger end to hold the neon tube, and an oval base with a mounting hole on the body of the tube to screw it onto the sign's frame. We used a couple vibratory feeders to supply the springs to the 'inserter' station, and had to add a few 'sortation' stops to the feeders to 'intercept' malformed springs - and never were able to 'catch' the perfectly straight, but incompletely formed (my assumption - the 'spring wrapper' ran out of wire part way through) ones - if I recall correctly, we had about a one in five or ten thousand failure rate. Now, GB may 'squat-down' harder on their suppliers for QC, or, in an application like that, it may be worthwhile to do automated optical inspection of each piece - but my main workstation board ~~has~~ (had) ten of these babies - if one in ten thousand is 'malformed', it would mean every thousandth board has a loose sink corner somewhere [:fixitbil]

As far as TIM goes, I usually recommend TIM Consultants' T-C Grease™ 0098, which can be ordered here, and is reviewed here. In case you don't want to be bothered to read the whole 'comparo', the trick to the stuff is that it's 'thixotropic', i.e., it becomes less viscous when shear forces are applied, so when you 'wiggle' the sink down into place, the stuff flows out better. Not particlarly expensive, and the thing that got me was that they threw it in a padded envelope, mailed it to me, and charged, like, a buck and a half, or two and a half for shipping! None of the usual "put it in a protective tube, and blister pack the tube, and put the blister pack into a box with a bunch of advertising crap on it, and wrap that box in bubble wrap and throw it into a larger UPS box, and charge more for shipping than the stuff cost in the first place!!" [:boudy]

Quote :

are you like some official Gigabyte techie?

Naahhh - just a regular guy - though I do levitate in the morning to put my pants on both legs at a time [:jaydeejohn:4]

I'm just a really old fart, [:bilbat:6] who's been 'pushing' computers for a loooonnnng time! Learned FORTRAN IV on an IBM mainframe as a junior in highschool in '69, using programs one statement to a punch-card, when a disk drive was a 'luxury item' the size of a small chest freezer; first overclock was an '87 KayPro Z-80, and I could cry :cry: for those days - for, like, $5.95 (and fifty-cents shipping and handling!) you got an actual schematic of your actual motherboard - and you could 'hack at it' with an actual oscilloscope to see actual signals!! [:isamuelson:8]

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

Cool stuff. i wish more companies would do things like that (Basic packaging) just to bring down prices of goods, seriously ive already bought it, why are they still advertising!

Looks like some Good TIM Ill weigh up the option later tonight, Im only having temps issues under Prime 95 Cause' of the temps. in gaming and normal everyday operation everything is fine.

That fact about the spring clip system scared me into opening everything up and checking, but alas no problems there...

Im also looking forward to some gtx 480 pulling a 'Melt through bootom of case and desk and ceramic tile syndrome's'.

It would be nice to get Schematics with your MOBO's it would make things interesting.

Just on another completely unrelated topic

GA-X58-UD7
or GA-X58-UD3L?

Which would you choose?

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

Ya know - that's the same thing I think every time I get one of these 'gussied-up', advertising-laden boxes: "You don't have to market to me, for god's [:lorbat:5] sake, I already bought your crap!! And, in a day or two - I'll know whether it works, or not?!?" Do they really think that there is a significant market of people who 'wander through' their local computer shop and pick out a motherboard based on how 'remarkable' the box is?? (Wait - now that I think [:grahamlv:3] of some of the questions we see here - that might explain a lot of things [:bilbat:9] !!) Gotta admit, though, the Sapphire vidcard boxes make good 'pin-up porn'

As for the MOBO comparison question, brings up nothing but more questions. First off, what market (continent, country) are you in? For US market, I see neither part number... The UD7 I'd assume is X58A-UD7? And we neither show an 'L' version of the UD3; nor any 'first gen' boards with a 'plain' X-58 designation - they were all EX-58-UDx...

Assuming were talking about one of three boards here, the GA-X58A-UD3R, the GA-X58A-UD7, and maybe the GA-EX58-UD3R, you can see a feature by feature comparison here...

I always tell people that questions like this kind of resemble "how high is up?" At the risk of sounding like the pimply kid in the 'big-box' hardware store, I gotta ask "what is it that you're trying to do?"

If I were building a production system to edit video, or even game, and thought I might need a couple or three grahics card slots, I'd stick with the 'A' suffix boards - ditto for three-channel RAM - ditto,if I gave a damn at all about USB3 or SATA3... (Actual, physical hard drives are years out from being able to use a 6G interface; SSD's could - but, so far, are disappointing - there's but two from Crucial, and they have 'semi-broken' firmware - though there's future promise from SandForce - and the USB makers are finding that, to make a flash drive fast enough to use the USB3 spec, they're having to 'double up' already too-hot controller chips, which means they need to use metal cases [which are pricey, and no one really wants...], and you might burn your thumb removing them [which no one really wants, either!] - so, as I've noted here before, when I see USB3 and SATA3 supported on an Intel chipset, I'll know it's officially time to 'transition'!) The older board is the only 'first gen' board I'd ever consider buying, as it's actually second-gen (the 1.6/1.7 revs)! There were PCIe bus termination problems in the first gens - probably only noticeable if you're a rabid overclocker... I would also never buy an 'L' suffix board, if there's an 'R' suffix of the same board available - the 'L's have an ICH10, the "R's, an ICH10R - never know when you're gonna wanna RAID - and it's nice to be able to just decide - not have to consider another board! (And, price difference should be negligible - can't remember the wholesale differential from Intel, but it's a 'pittance'...)

Message quoted 1 times
Message edited by bilbat on 05-07-2010 at 07:59:23 PM

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

bilbat wrote :

Do they really think that there is a significant market of people who 'wander through' their local computer shop and pick out a motherboard based on how 'remarkable' the box is?? (Wait - now that I think [:grahamlv:3] of some of the questions we see here - that might explain a lot of things [:bilbat:9] !!) Gotta admit, though, the Sapphire vidcard boxes make good 'pin-up porn'

I would also never buy an 'L' suffix board, if there's an 'R' suffix of the same board available - the 'L's have an ICH10, the "R's, an ICH10R - never know when you're gonna wanna RAID - and it's nice to be able to just decide - not have to consider another board! (And, price difference should be negligible - can't remember the wholesale differential from Intel, but it's a 'pittance'...)

Thanks, alot of useful info there. Just refering to the "Pretty Packaging" issue I have seen a man walk into my favorite IT store Look at the Box, Pick one Up, tilt it in the light and the same with the one next to it. He then walked to the counter and bought the first board to the counter... the scary thing is that all the Specs where on the back of the box...SIGH

Sorry bout my Codename mess up, out of sheer luck you managed to target the two boards that i was originally looking for!

I agree with you on the SATA3 And USB3 Issue, i don't think The extra speed would be worth burnt fingers.

I am A casual gamer, my Last GA-X58-UD7 Setup was drowned in a water cooling accident so i was just wondering if there was anything better on offer by GB at the moment (in that price range)Im a pretty Good OC er if i don't say myself, i like to push hardware :sarcastic: .

I would like to Go crossfire int the future, also looking at Virtualization on desktops at the moment

Im Living in South Africa At the Moment. My main supplier is Rectron SA. Their prices are descent but their customer service is outstanding, they will be able and have proved competant at aquiring specialist hardware.

Again thank You.

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

I think the X58A-UD7 is pretty much the top of the line, currently...

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

^ i thought so... Any idea on any Monstrous Boards currently planned for release?

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

Nah - I don't have any 'inside' contacts (as of yet, anyhow - I'm workin' on it! [:bilbat:2] )

I do know exactly what they should make: 7025

Lessee:
...dump the Aspeed 2050 graphics... (not gonna be a server!)
...dump the AST2050 IPMI/iKVM (ditto!)
...dump the Winbond W83793G monitoring for their usual iTE IT8720... (less real-estate!)
...dump the space 'reserved' for the LSI 1068E SAS controller ... (ditto!)
...dump one useless DIMM slot from each CPU cluster... (ditto!)
...dump the Realtek ALC262 CODEC ('cause it's worse than useless) & fit an eight-channel ALC889A in the same space...
...keep the slot arrangement...
...don't even think about USB3 (anyone's!) or SATA3 (Marvell's)... (just 'cause you've got 72 lanes of PCIE don' mean you gotta waste some!)
...shrink to fit (eight layer MOBO?) regular ATX form factor...

Instant market leader! "Twenty-four gig memory, twenty-four meg cache, twenty-four hyperthreads - brutal symmetry!" - there, I've got the marketing guys' job done for 'em, too!

Message edited by bilbat on 05-08-2010 at 07:30:28 PM

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

That would be a monstrous board...

especially if it was able to function with only one CPU (Low budget systems high quality and good scalibility on the MOBO)

I would keep the SAS controller (Ive hear talk of Solid state SAS drives!)
i would think about adding SATA 3 AND USB 3 On a Later revision board

Mabybe 6 Dimms (DDR3) per CPU Like EVGA's Classified SR2
[img]http://www.google.co.za/imgres?imgurl=http://img.hexus.net/v2/news/EVGA/classified-sr2-1.jpg&imgrefurl=http://www.hexus.net/content/item.php%3Fitem%3D22960&usg=__vNnw0-KQ2sIJFy2jHL0PZsDb5F0=&h=400&w=450&sz=196&hl=en&start=11&sig2=GwUt3SI3wjaToAaJPZB1fw&um=1&itbs=1&tbnid=Y3Bq5XQMLMxQVM:&tbnh=113&tbnw=127&prev=/images%3Fq%3DEvga%2BClassified%2Bsr%2B2%26um%3D1%26hl%3Den%26client%3Dfirefox-a%26sa%3DN%26rls%3Dorg.mozilla:en-USfficial%26tbs%3Disch:1&ei=qYvmS8TiFtaa-AaWg5ToAw[/img]

I agree with the Sound card option

Maybe 2x 1000/100/10MBPS Network Option

I would settle for extended ATX.

It would eclipse all boards out there!

LOL, there we go Gigabytes GA-X58A-UD9!

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

IGNORE PREVIOUS POST (Sorry it wouldn't let me Edit [Damn JAVA])

That would be a monstrous board...

especially if it was able to function with only one CPU (Low budget systems high quality and good scalibility on the MOBO)

I would keep the SAS controller (Ive hear talk of Solid state SAS drives!)
i would think about adding SATA 3 AND USB 3 On a Later revision board

Mabybe 6 Dimms (DDR3) per CPU Like EVGA's Classified SR2

I agree with the Sound card option

Maybe 2x 1000/100/10MBPS Network Option (sorry, i see the board has that already)

I would settle for extended ATX.(E-ATX)

It would eclipse all boards out there!

LOL, there we go Gigabytes GA-X58A-UD9!

Who the heck is TYAN ?

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

Have you looked at that SR2's manual? Read the BIOS Tuning section especially - and we dare to complain about the GB manuals!! They 'cheat their way into the slot count by using an nVidia NF200 muliplexer 'break-out'; and I always say "if you rob Peter to pay Paul for toooo long, you get a 'sore Peter'!! [:fixitbil:1] And I need the [:bilbat:2] PCI slot for embarassing hardware - have a modem in every system - lazy man's autodialer...

Tyan is MiTAC, Taiwanese, (isn't everyone) concentrate mostly on server hardware: MiTAC International Corp. 神達電腦股份有限公司

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

Lol Yes... The Manual is BAD

Unfortunately you have lost me with the NF200 Multiplexer 'Break-Out'. I have never heard of it before. My Ignorance astounds myselfs!

As for the PCI slot, I might agree with you, Only one though, no more is needed

I still Wouldn't Rush out and buy A TYAN Board anytime soon!

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

The NF200 is a bridge chip that 'spreads' PCIe bandwidth, so instead of the shipset/processor 'allowed' bifurcation of ports (the 'bifurctaion simply means 'halving' - i.e., if you have a 'native' sixteen lane physical 'provider', you can get, say, an eight and two fours), you can get ports that 'appear to be' wider than bifurcation allows. A number of the 'high performance' boards made to support three-way or four-way xFire/SLI are made this way... The underlying chipset has two sixteen lane 'providers', which normally would allow a sixteen, and two eights; the NF200 allows them to 'present' aditional ports that are nominally and electrically sixteen lanes, but, of course, have some bandwidth penalty attached... I'm pretty sure that nVidia's twing-GPU vidcards also use them internal to the card. Good article at Tech News Daily!

Besides this little 'tray born' dialer:

I've been 'fiddling' with some stuff that should provide 'pop-up' caller-ID info for incomings, even inside Media Center - but, so far, no workee :heink:

Tyan is kind of the 'obverse' of GB: GB sells a lot of consumer boards; makes, but doesn't sell a lot of server setups; Tyan sells a lot of server boards (and servers); they make, but don't sell a lot of consumer boards... Unless my 'dream board' shows up pretty soon, I think my next workstation 'rebuild' will be 7025 based - want to start experimenting with learning OpenCL, need the x16 slots. A six-monitor vidcard (have four monitors 'wasting' two slots now...), a pair of ATI stream processor cards, and an eight-lane Areca RAID controller...

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

OK, I get what you are saying with the NF200 Stuff

Caller ID Would Be pretty kick ass

I'll Ask some of my Contacts if they have ever Heard OF TYAN Or MiTAC.
A dream Board... The thing with that is, by the time they have gotten their ducks in a row to manufacture the board Technology will have moved... DDR4 Memory Onboard SAS or ScsI Raid Controllers and even PCI-E 24x (Rumored)

My Theory on the dream board topic is that you should grab a top Of the line Mobo When They Come available and not bother wasting time, money, RMA stresses and Faulty Bios's and firmware.

All that being said , a Dual CPU Mobo Sounds Like A good Idea... If they can master Up a new chipset for them to Unlock their full potential... Who knows?

Personally i would love to see GB come out tops. Releasing A monstor before any other manufacturers know what hit them and then copyrighting their Chipset design...

Open CL sounds Like Fun but isn't it only available on nVidia Cards?
Googled that raid controller and... WOW.
Very nice server grade Hardware.

What do you Mean by 7025 Based?

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

The 7025 is the Tyan 'server board family' I pointed out earlier. The chipsets (5520) exist, but few 'do it right'; the 5520 has a pair of QPIs, like the dual-CPU Xeons, so you wind up with this:

Trick is, you gotta use two of 'em, which not many do! ASUS makes one, which contains way to much 'proprietary tech' to appeal to me, and SuperMicro makes one, but it's the size of medium county in Texas - need a proprietary case, which are really rack-mounts that you 'can' tip on end! CUDA only works on nVidia, OpenCL is, well, 'open', and ATI makes some 'equivalents' to the Tesla (and maybe Fermi, if they haven't completely gutted it by the time it's actually released...

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

Oh, Yes Psh i knew that!

I see on that diagram that CPU0 IS allocated Memory Banks CHA+CHB+CHC

While CPU1 Is Allocated Memory Channel CHD+CHE+CHF

Does this mean that CPU1 Will only activate When CPU0 Is under Intense Load?
Does this also mean that CPU0 Can "Borrow" Ram From CPU1?(I realise this process would be sow as it has to activate CPU1, Process the Memory and then Port it over the QPI link joining the 2 CPU's)

Does any software Support Dual CPU Mobo's?

I also See that THE IOH0 and IOH1 Are joined BY a QPI Link, What is the possible reason for this?

Where Would The SB/BIOS be routed through in a setup like this?

So many Questions!

ATI 100-505584 FireStream 9270 2GB 256-bit GDDR5 PCI Express 2.0 x16 Stream Computing Card

* FireStream 9270
* 2GB GDDR5
* PCI Express 2.0 x16

* Chipset Manufacturer: ATI
* Stream Processors: 800 Stream Processing Units
* Memory Clock: 850 MHz
* Operating Systems Supported: Windows Vista (64-bit and 32-bit), Windows XP (64-bit and 32-bit) Linux (64-bit and 32-bit)
* Model #: 100-505584
* Item #: N82E16814195081
* Return Policy: VGA Standard Return Policy

*
* $1,204.99
*
* Free Shipping!

This Card Costs About 2500USD IN South Africa, Grab it... Quick

What does Open CL even DO?

------------------------------ Please read this before posting your next Question http://www.catb.org/~esr/faqs/smart-questions.html | [:fisshy]

|

!Now To Overclock My Brain!

Reply to fisshy

Quote :

Does this mean that CPU1 Will only activate When CPU0 Is under Intense Load?

Thread distribution (i.e., what core will get what job) is generally done by the OS, though (more and more) programs themselves have OS 'hooks' to allow them to distibute their work across many threads (cores)...

Quote :

Does this also mean that CPU0 Can "Borrow" Ram From CPU1?(I realise this process would be sow as it has to activate CPU1, Process the Memory and then Port it over the QPI link joining the 2 CPU's)
Does any software Support Dual CPU Mobo's?

Pretty much, all modern OS support this - windoze 7 (not sure about Vista) and Linux support NUMA (non-uniform memory access), which allows a piece of software to set a 'processor affinity' function, so it can, whenever possible, use a particular processor. Imagine this: from the diagram above, let's say you have a pair of x16 graphics cards in slots 1 & 3; you'd really like to tell the video driver to run on CPU0, as it has more direct access to those particular slots... Unfortunately, I am not aware of vidcard drivers' support yet - I try to 'nag' someone at least once a week! [:isamuelson:8] One of the reasons dual-CPU supported Xeons cost more is that, in addition to the extra QPI link, they have internal, automated 'cache snooping' features, to 'see' each other's caching operations, and sub-systems to access each other's memory 'stock' more efficiently!

Quote :

I also See that THE IOH0 and IOH1 Are joined BY a QPI Link, What is the possible reason for this?

Simple reason: PCIe is 'bi-directional'; cards can not only 'swap info' both ways to/from the CPU(s), but can 'talk' to each other (that's also why 'xFire' and/or SLI - makes their 'local conversations' faster than having to depend solely on the PCIe link); much quicker for card in slot 1 to 'gab with' the card in slot 7 if it can avoid having to go through three QPI links 'along the way'...

Quote :

Where Would The SB/BIOS be routed through in a setup like this?

The BIOS, by necessity, must 'talk to and through' everyone; if you'll look carefully, you'll see that there is just a little bit of a 'truncated arrow' at the bottom right of IOH0 - that is the ESI link to the southbridge. The IOHs even have a 'register setting' to establish 'where this is'; one of the IOHs in a multiple setup is designated 'Legacy IOH'; that is the one through whom all southbridge traffic is automatically directed...

Quote :

What does Open CL even DO?

...to explain this, I have to explain a bit about how modern graphics processors work... How a scene is 'rendered' is by 'tesellation', sub-division of the elements into smaller-sized triangles. In a three-d scene, you can easily see how each element 'impinges' on each surrounding element - reflections, shadows, etc... GPU guys have figured out that it's infinitely faster to 'compute' these effects by loading an array of these elements into a multi-core GPU, and crank 'em out a 'batch at a time'. GPU multi-cores have, thus, been developed to a finely-honed 'state-of-the-art', driven mostly by the (bless their money-wasting hearts) gamers!

Now, it turns out there are large numbers of 'real-world' processes that lend themselves, as well, to this 'divide-and-conquer' strategy: molecular modeling, computational fluid dynamics, futures trading analysis, weather simulation - all involve 'wide arrays' of particles, atoms, offers and bids, each of which 'impinges' on the surrounding elements. Stream cards, and Tesla/Fermi take an advanced graphics processor, and 'make it available' for 'general-purpose' use - and OpenCL is the language that works with the largest number of these setups. Computational fluid dynamics actually has its own language 'FOAM', that is spreading, and nVidia has their own language, CUDA, which is highly optimized for their specific cards... The 'trick' is that, when 'spreading out' a problem across a hundred, or four-hundred cores, there must be 'mechanisms' to provide for inter-process 'signalling' and 'synchronization', which, in the real world, becomes the biggest issue for implementations - and this is what OpenCL is 'built' to accommodate [:jaydeejohn:3]

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

Quote :

I also See that THE IOH0 and IOH1 Are joined BY a QPI Link, What is the possible reason for this?

<><><><>>>><<<<<><><><><><><><>

Quote :

Simple reason: PCIe is 'bi-directional'; cards can not only 'swap info' both ways to/from the CPU(s), but can 'talk' to each other (that's also why 'xFire' and/or SLI - makes their 'local conversations' faster than having to depend solely on the PCIe link); much quicker for card in slot 1 to 'gab with' the card in slot 7 if it can avoid having to go through three QPI links 'along the way'...

Sorry Stupid question.

Quote :

Pretty much, all modern OS support this - windoze 7 (not sure about Vista) and Linux support NUMA (non-uniform memory access), which allows a piece of software to set a 'processor affinity' function, so it can, whenever possible, use a particular processor.

Should have seen that coming...

Please keep me Updated on those Vidcard drivers, Just please PM me as you get any info

Quote :

Computational fluid dynamics actually has its own language 'FOAM', that is spreading, and nVidia has their own language, CUDA, which is highly optimized for their specific cards...

Very interesting, Classic case of Nvidia Being Themselves and screwing the consumer over... :non:
So If i am not mistaken OpenCL is A programing language? :??:

Message edited by fisshy on 05-12-2010 at 07:10:56 PM

Reply to fisshy

Quote :

Very interesting, Classic case of Nvidia Being Themselves and screwing the consumer over...

Yup - they're taking lessons from Apple - the main people who use CUDA currently are researchers in university labs - and they've got 'grad students to burn!' nVidia is making 'em deals on getting Teslas into the schools cheap, on the theory that once they've learned how to do everything in CUDA, they'll be too 'canalized' to ever want to xfer their programming, libraries, and skillset over to OpenCL, or anything else! Not a bad plan, assuming they can ever make an actual Fermi card!

Quote :

So If i am not mistaken OpenCL is A programing language?

Yup - I should have said right away - info here, at Khronos...

Message quoted 1 times

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

bilbat wrote :

Quote :

Very interesting, Classic case of Nvidia Being Themselves and screwing the consumer over...

Yup - they're taking lessons from Apple - the main people who use CUDA currently are researchers in university labs - and they've got 'grad students to burn!' nVidia is making 'em deals on getting Teslas into the schools cheap, on the theory that once they've learned how to do everything in CUDA, they'll be too 'canalized' to ever want to xfer their programming, libraries, and skillset over to OpenCL, or anything else! Not a bad plan, assuming they can ever make an actual Fermi card!

What a Scary Thought... The ATI fanboy inside me is busy panicking! :cry:

I cant see myself going out and Buying An Nvida Vid card anytime soon...(Been ripped off one to many times! :non: )

My theory is that IF Nvidia Wants to go ahead and Spawn a generation Of CUDA programmers then ATI should retaliate with OpenCL.

It would be cheaper to implement: The SDK is FREE ...

But as you State this all hinges on Fermi...

Watch this, it made me laugh (Just a Note... it has very strong language)

Learnt so much More Than Why My NB was over heating, Thanx :hello:

Reply to fisshy

Look What i Just Found GA-X58A-UD9 :bounce:

Press Release GIGABYTE Unleashes GA-X58A-UD9, Unlocks Monster Performance
GIGABYTE TECHNOLOGY Co., Ltd, a leading manufacturer of motherboards, graphics cards and computing hardware solutions is proud to unleash their latest monster performance GA-X58A-UD9 motherboard, featuring a revolutionary new 24 phase Unlocked Power design, 4-way graphics support including NVIDIA SLI and ATI CrossFireX, as well as a host of unique GIGABYTE features such as 333 Onboard Acceleration and On/Off Charge.

"GIGABYTE set out to completely redefine what is possible performance and feature-wise on the X58 platform, and the GA-X58A-UD9 truly delivers on both fronts," commented Tim Handley, Deputy Director of Motherboard Marketing at GIGABYTE Technology Co. Ltd. "Providing the industry's highest caliber CPU power delivery with our unique 24 phase Unlocked Power design and including all of the features which set GIGABYTE motherboards apart from the competition, the GIGABYTE GA-X58A-UD9 fits into a whole new first class category all its own."

Well That Sucks...
I expected more :??:

Found it here

Reply to fisshy

Saw that new little honey at Semi-Accurate 'bout simultaneously! Notice: it uses a pair of those nVidia NF200 'bridges' to get the 'big slots'!

------------------------------ An optimist sees a glass half full; a pessimist, a glass half empty; an engineer sees a glass that's twice as large as required...
Reply to bilbat

Forum Motherboards & Memory : Gigabyte GA-EP43T-UD3L SpeedFan Issue

Motherboards

Memory

Chipsets & Bios

Abit

Aopen

ASRock

Asus

Chaintech

Corsair

DFI

Elitegroup

Epox

FIC

Foxconn

G.SKILL

Gigabyte

Jetway

MSI-Microstar

Shuttle

Soyo

Supermicro

Tyan

Please mind