Can we get a masssive Delidded thread going again?

[chris89]

I don't think that benchmark is very accurate. Since my 580 score 10.50 giga pixel's at 900mhz and the test is only 1 Second long... I could clock it easily to 1025Mhz for a stable 1 Second long benchmark...

Compute performance is a fairly accurate indicator of a stronger performer...

Even Luxmark V2.0 apparently on the Default scene compared to others...

 

[LTC8K6]

http://www.3dmark.com/3dm11/7301064

GTX480 at 1Ghz 3dMark

 

[chris89]

http://www.3dmark.com/3dm11/7301064

GTX480 at 1Ghz 3dMark

Nice heres a little comparison showing how the 5300Mhz CPU clearly outperforms a 2926Mhz CPU ... 900Mhz 580 vs 1010Mhz 480 clearly beating quite a bit lol... I ran it on a stable system not trying to get high scores... The stretch above 900Mhz on the 580 requires more watts than fps gains... Same applies to 480 above 850mhz the watts go up to ridiculous levels and fps sees a decent gain... Did you volt mod the 480 above 1.213V?

http://www.3dmark.com/compare/3dm11/7301064/3dm11/9799392

This as good as I could get for the moment... having trouble with 975mhz but 950mhz went fine...

http://www.3dmark.com/compare/3dm11/7301064/3dm11/9799410

 

[MrTeal]

You should take a look at the datasheet of the graphite TIM. The 240W/mK rating will be in the x-y direction, while it will have another rating in the z direction. For example, something like the common Laird TGon 800 has a z-axis (through the sheet) rating of 5W/mK. Just be careful with them as many are electrically conductive.

I'd be careful using too much pressure, you really don't need that much if you switch to liquid TIM. You just won't get the performance you're looking for using a pad of any kind. I've played with a lot of TIMs, and it's very tough to beat a nice thin application with direct die contact. I've dissipated 800W+ from 324mm^2 of die area using a Swiftech H220 CLC with some big server fans blowing through it, and die temps stayed around 100C. That's die power, not board power; The board itself was pulling 1kW.

If I were you, I'd give die->paste->shim->Liquid Ultra->HS a try. It won't perform as well as direct die attachment, but it'll still give you excellent performance.

 

[chris89]

Tgon™ 800 Series - I feel that Z Axis is false advertising... The performance when stacking is pretty bad it seems to cut it equally per stack meaning 5w to 2.5w to 0w... Seriously if u stack 3 of them you get 0% thermal transfer 100% Nothing transfers...

I haven't tested the T62 Series --- Since it's Z-Axis is 15W/m K and that Z-axis is really all that matters... It is misleading with that worthless X-Y rating ...
t-Global Technology


What board was pulling 1kw? I have seen my 480 pull more power than I've ever seen anything pull ever before for sure... Extremely high wattage under overclocks of 850-900mhz.

The 480 though can be very efficient and perform very good at the extreme efficiency. I run a 480 for my bro at 608mhz core 1216mhz shader 1700mhz ram at 950mv... Which shows in the temps alone its drawing less than 200 watts my guess. Pretty much runs with no vsync at around 200fps on cs:go 1080 at 144hz for him at like 65C.

For him I run a [FONT=&quot][FONT=&quot][FONT=&quot][FONT=&quot]Chooyu Chemical Co. Ltd.[/FONT][/FONT][/FONT][/FONT] PAKCOOL TP-260(0.5 mm) on the core. Bergquist all the vrm and ram. PAKCOOL is 6W/m K and Bergquist is 5W/m K.

Thanks. My problem with my 480 was the lid came off so I was trying to get it working again.

It appears to contact only on the outer edge of the chip...

Can you tell me a little about these items I'm testing now?

http://www.digikey.com/product-detail/en/0/926-1013-ND

http://www.digikey.com/product-detail/en/0/1168-1413-ND

 

[MrTeal]

What board was pulling 1kw? I have seen my 480 pull more power than I've ever seen anything pull ever before for sure... Extremely high wattage under overclocks of 850-900mhz.

The 480 though can be very efficient and perform very good at the extreme efficiency. I run a 480 for my bro at 608mhz core 1216mhz shader 1700mhz ram at 950mv... Which shows in the temps alone its drawing less than 200 watts my guess. Pretty much runs with no vsync at around 200fps on cs:go 1080 at 144hz for him at like 65C

Thanks. My problem with my 480 was the lid came off so I was trying to get it working again.

It appears to contact only on the outer edge of the chip...

Can you me a little about these items I'm testing now?

http://www.digikey.com/product-detail/en/0/926-1013-ND

http://www.digikey.com/product-detail/en/0/1168-1413-ND

It was a custom Bitcoin mining board I designed using four 6-phase IR3550 based VRMs. Lots of umph to push power into a chip.

The TGON 820 is a 20 mil thick 5W/mK pad. They have amazing x-y thermal properties, so they're really good at taking heat from a chip and spreading it over a large area, but for a straight thermal interface pad they're actually pretty mediocre.

The T62-1 is much better in that regard, at 15W/mK and only 6.3 mil thick.

I'm not sure on surface wetting ability of either of those pads though. The rating is the bulk conductivity of the material. It doesn't say anything about the ability to get heat from the die into the pad, or from the pad out to the shim/IHS. From most non-pad thermal interfaces, minimum bond line thickness and the corresponding wetted area are vastly more important than the bulk thermal conductivity of the TIM.

 

[chris89]

It was a custom Bitcoin mining board I designed using four 6-phase IR3550 based VRMs. Lots of umph to push power into a chip.

The TGON 820 is a 20 mil thick 5W/mK pad. They have amazing x-y thermal properties, so they're really good at taking heat from a chip and spreading it over a large area, but for a straight thermal interface pad they're actually pretty mediocre.

The T62-1 is much better in that regard, at 15W/mK and only 6.3 mil thick.

I'm not sure on surface wetting ability of either of those pads though. The rating is the bulk conductivity of the material. It doesn't say anything about the ability to get heat from the die into the pad, or from the pad out to the shim/IHS. From most non-pad thermal interfaces, minimum bond line thickness and the corresponding wetted area are vastly more important than the bulk thermal conductivity of the TIM.

Thanks man. I have been testing different materials over a period of time now. I ran into Digikey and started testing some stuff I thought was going to blow away all the Silicone material I test from Bergquist, Chooyu Chemical Co Ltd, and Fujipoly.

Come to find out the Graphite did great on a AMD Opteron but it's a cool chip anyway I think the Opteron was super flat too so it was a perfect bond...

I'll try out he T62-1 on my Xeon's to see if temps drop any... I'm running Fujipoly 0.5mm 14w/m K on them right now silicone gap pad. Is it likely I will see zero gains?

The Graphite really requires a perfectly flat surface and a fairly high compression to achieve very good temps. I find however getting very good temps is pretty difficult. I found myself requiring to completely pull the card apart after 10 seconds into boot the interface was not working at all.

 

[chris89]

I just re-did it again looking at 95C within 30 seconds. This is getting aggrivating. I think the diode requires paste because a pad can't hold the insane heat coming off the 480 diode even at 950mv and 607mhz.

 

[sm625]

Why dont you just remove material from the IHS until your temperatures return to normal? After the IHS comes off like yours did, it wont sit flat like it used to, due to the excess material left on the board and on the IHS. So it is probably sitting too high relative to where it used to sit, so the die is not making good contact with the IHS.

 

[chris89]

Thanks yeah I think the real issue now is the shim width needs to be precise and paste needs to be used between the diode and shim and probably between the shim and heatsink. Since we need a gap so tight a 0.5mm pad is too great.

I ordered again 1.25" x 1.25" which is pretty close to the precise size of the diode. Just a little big maybe 1.15" or 1.1" is ideal hell maybe it's 1"x1" I haven't measure.

The problem I had is 1/8" is too thick because the screws aren't long enough and I used screws 25% longer than the original 1/8" is far too great.

So I'm ordering again 1.25" x 1.25" at half of 1/8" thickness (Maybe 2.75mm?) I believe is ideal since 1/8" is 3.175mm so we get the point...

Anyway I also lap the shim to perfect flatness lapping on 600 grit until it's perfectly flat which did make a difference for sure!

The 480 idles with the Pad at 47C but that's 71% fan speed... So I hope to gosh this is resolved within the coming weeks. lmao

 

[soccerballtux]

well idc was the progenitor of the last one and he's been permabanned now so...

 

[xthetenth]

Wait what? When did that happen?

 

[soccerballtux]

lol

 

[chris89]

Well small update picked up longer screws but it didn't matter.

I actually found out how to get unreal temps with the non adhesive graphite pad. The pad is extremely thin like as thin as a piece of paper maybe thinner. It's super cheap too a huge 18"x18" sheet is like $7 with shipping $15.

Anyway I noticed temps no more than 60C under load which is a 20C drop from standard performance. The heat is transferred to the heat pipes in a lightning hurry. Within 5 seconds of post the pipes were working like a mad man.

ANYWAY it doesn't matter because the GTX 480 MPGA Diode had a nano crack from previously and it's gone. RIP GTX 480...

Anyone know where I can have a new MPGA chip soldered on at?

Or would it be cheaper to get a different card... lol

 

[MrTeal]

Well small update picked up longer screws but it didn't matter.

I actually found out how to get unreal temps with the non adhesive graphite pad. The pad is extremely thin like as thin as a piece of paper maybe thinner. It's super cheap too a huge 18"x18" sheet is like $7 with shipping $15.

Anyway I noticed temps no more than 60C under load which is a 20C drop from standard performance. The heat is transferred to the heat pipes in a lightning hurry. Within 5 seconds of post the pipes were working like a mad man.

ANYWAY it doesn't matter because the GTX 480 MPGA Diode had a nano crack from previously and it's gone. RIP GTX 480...

Anyone know where I can have a new MPGA chip soldered on at?

Or would it be cheaper to get a different card... lol

You actually want to replace the chip on it? Heh, even if you could source it, getting it replaced would probably cost you more than the $125AR you'd spend on a new 270X, which is both faster and uses less than half the power of that 480.

 

[chris89]

Yeah sourcing a replacement is the problem right knowing if they are a legit professional company with excellent competitive pricing under $100 hopefully.

I could pick a new 480 up for like $50-75 If I get the right offer or auction. This 480 I paid $35 for and originally had artifacting beyond recognition. Through the use of up to 5 or more bakes requred to get it going 100% stable. Which was great but there was a tiny issue around bake 4 that was an issue that did get fixed with the 5th bake, although the IHS melted and came off after higher than normal temperatures by a long shot. So I basically was working with a perfect operating card prior to the IHS falling off as it was a learning experience. Imperfect screw lengths and pressures were likely the cause.

The 480 does perform amazingly though even today. GTA V for instance runs perfectly smooth all max details and distances. 384-bits of GDDR5 at 200,000 Mega bytes per second helps. It also has 48 ROP's which is equal to for instance a GTX Titan. It's only lacking texture mapping units so a high clock is ideal upwards of 850Mhz on the core clock is entirely possible. That brings the 480 up to over 40,000,000,000 Pixel's per second.

Anyway Thanks bro

R7 270X

compute cores---Texture Mapping Units---ROP's

1280 : 80 : 32

GTX 480

480 : 60 : 48

---R7 270X moves textures around a bit faster but is limited by lower than 48 Pixel Pipelines. Pixel Pipelines are as fast as what seeing on your screen ie "pixel's". So the R7 270X can handle a raw texture sitting in video memory better it's limited by it's 179 Gigabyte per second 256 bit memory. The 480 can run at around 210GB/s with high power consumption it does make up for it.

Good price though for sure with probably very low temps right on the r7 270x?

 

[MrTeal]

You can't really directly compare AMD vs nVidia compute/TMUs/ROPs, or even directly compare them for different architectures. In every gaming/compute benchmark though the 270X is faster, by lots. It even just eeks ahead in pixel fillrate.
http://www.anandtech.com/bench/product/1043?vs=1135

All depends what you want to do, really. If you want to actually game on it, I'd say just buy a new card. You'll almost certainly spend more fixing the 480. If you want a project, buy a new card and then try to get a broken $20 480 and do the change yourself.