Swiftech Releases Komodo HD5800 F/C Waterblock For ATI

[Philwong]

What is this, http://www.moderatesystems.org?

Unfortunately, the Komodo is not the most "extreme" block as far as VRM temp is concerned. It's all about expectation and you ought to be looking at EK for the lowest possible temp. Like I mentioned earlier, if Swiftech can bring down the temp by such a huge degree just by using paste, there's no reason why EK will not benefit from this tweak as well.

Phil

[dejanh]

Unfortunately, the Komodo is not the most "extreme" block as far as VRM temp is concerned. It's all about expectation and you ought to be looking at EK for the lowest possible temp. Like I mentioned earlier, if Swiftech can bring down the temp by such a huge degree just by using paste, there's no reason why EK will not benefit from this tweak as well.

Phil

My comment was only intended as a rebuttal to your statement of pushing the card to the extreme. I hardly pushed the card to the extreme, and personally I even stated that I do not think the card died because of VRM temperatures but rather all of the mounting and remounting. Anyway, the main reason I took everything down again last night is so that I can get data for Gabe to show him the problem. I got the data, but unfortunately I also got a dead card now. I had no choice anyway though but to take down the loop again last night since the VRM temperatures were out of control. I was intending to replace the paste with pads again but once everything was done the card would no longer start. I'm loading the images and doing the write up now, they will be up shortly.

Edit: @Philwong, I forgot to mention something. Furmark got nerfed by ATI in their drivers after I reported a problem with it overheating the VRMs at launch of HD 4800 series cards several years ago. I realize that Furmark is not representative of normal load however it is a great test tool and shows problems quickly. After the VRM debacle that happened with the HD 4800 series cards I did not expect a repeat of the problem, and as you can see below I was right. Furmark or not, the block is not making contact as it should be so the VRMs burn, burn, burn

[dejanh]

Ok, I sorted through the photos and my notes and I think this one illustrates the best the lack of contact. I needed to put on a lot of paste to get anything to come across onto the VRM cooling area. The block was compressed as much as it can be to the graphics card.



Notice that the GPU die and the single VRM in the lower right corner have good compression, but the VRMs at the rear of the card barely make contact, and definitely do not have any compression.

And another screenshot following maximum compression of block to card, this time with abut 1/2-3/4 of the thermal paste; notice one of the VRMs has no more contact at all. Also notice the scratches around standoffs as a result of trying to tighten them down as much as possible to get the block to make proper contact in vain.



Now, I have measured the PCB height, VRM height, standoff height when fully screwed in and standalone, VRM plateau height, and determined that the VRMs cannot be cooled with the paste method on either of my Komodo blocks with this card. All measurements have a margin of error +/- 0.05mm due to human error and were done with a digital caliper.

PCB, VRM dimensions:

PCB thickness: 1.55mm - 1.6mm
VRM (rear) thickness: 0.85mm
VRM (front, single) thickness: 1.1mm

Block dimensions, around rear VRM area:

Top-left corner + standoff: 15.30mm
Bottom-left corner + standoff: 15.1mm
VRM plateau height (average): 14.15mm
Difference: 0.95mm - 1.15mm

As you can tell from these dimensions it is not possible to make good contact with the VRM using paste only as the difference between the VRM thickness and where the plateau sits once mounted ranges from 0.1mm to as much as 0.3mm.

The second block is more or less the same, having dimensions 15.05mm/14.75mm/13.9mm (avg); net difference 0.85mm - 1.15mm. Furthermore, on the second block even the single VRM would not make proper contact as the net difference in height between the standoff and plateau height is 1.2mm - 1.3mm, larger than 1.1mm height of that VRM.

Furthermore, the screw-in depth for standoffs and standoffs themselves are not equal. They range in height from 3.85mm to 4.1mm in no practical order. The screw in depths appear to do the same and are uneven. I took a 4.1mm standoff and placed it into the center hole by the DRAM cooling pad near the inlet/outlet and the PCB would still warp here because the screw in depth was deeper by about 0.3mm than other areas. This would happen every time but if the fastener was loosened to eliminate the warping insufficient contact with the DRAM would be made. Finally, around the single VRM area the board warps significantly once the card is mounted in the case as the perpendicular forces exerted by the weight of the block pulling down and away from the mounting slots on the PCB.

Two blocks later I would say that using this block without thermal pads and instead using thermal paste is strongly not recommended, at least with MSI R5850. I will test things out on the Asus EAH5850 as well once it is here in a few more days. The MSI R5850 is now off for an RMA and I really, really hope I actually get the replacement.

[jaredpace]

Hello,

The indention you have marked, "Bad contact, no compression" with a red line with two arrows on each end, can you remove the pad sitting inside that indention to give your block more height/clearance to successfully cover the components on the pcb marked by the white sticker, "CPL2-3". Then you may have greater compression/pressure on the 5 VT1165 slaves to the left which are not being touched by your cooling block. Also, can you make your cooling block mount tighter to that area of vrm slaves?

Why not use a highly conductive thick thermal pad of good quality to mount on those 5 components. These 5 slaves need *MUCH* more cooling than the parts under the white sticker CPL2-3, which require little cooling, and are not even touching the stock cooler.

[dejanh]

sorry for not being able to get back to you faster, but I am in China right now, and facing very hectic schedule.

You are essentially correct: the key here is standoff height. So please make sure that all standoffs are completly fastened tight, particularly in the VRM's area. One or two more turns of fastening the standoff will make the difference between poor contact and good contact. We are indeed dealing with very small distances here, of the order of 0.1 to 0.2mm. anmd since we do not want to warp the PCB in excess, tolerances are tight indeed.

I officially implemented the "no pad" tweak after successfully testing it myself with 5850, 5870 and 5970, so I felt very comfortable recommending it.

BUT making sure that all standoffs are fastened tight is something that I do automatically without even thinking twice about it, and in hindsight I should have added this in the instructions. Using pads gave us sufficient margin to allow for larger tolerance gaps, but not using pads is obviously reducing the margin of tolerance that we have, and therefore absolute repect of the specified standoff height relative to PCB becomes obviously critical.

The conendrum we find ourselves into is this: using standoffs to prevent excessive warpage of the PCB, in other words user safety, and the desire to provide a perfect TIM between VRM's and copper plate. In theory, these two goals are mutually incompatible. In practice, they are possible, but reserved to advanced users who have a solid understanding of the underlying difficulties associated with this kind of mechanical challenge. This is the very reason why I included the tweak in an "advanced users" section of the installation guide.

Other vendors (who do not use standoffs) do provide a better TIM, but this is at the cost of safety. The users of these competing solutions typically qualify themselves as advanced, and so they do not really mind taking the risk. We on the other hand, want to provide safety as well as performance, and it would appear that we still have either some tweaking or better maybe explaining to do in order to accomplish this.

Doing a hard mount (without pads) is quite possible with the Komodo's, I have done it, and I'll stick to my guns about it. Performance is exceptional, and pcb warpage was minimal. Should I revise my standoff heights and therefore reduce the tolerance gap without pads is something I now need to consider carefully.

Hi Gabe, thanks for taking the time to quickly reply. Unfortunately I have automatically done exactly what you have suggested. All of the observations and measurements that I have provided have been taken after tightening the standoffs as much as physically possible on the whole block, and as I have indicated previously, even replacing "longer" standoffs for "shorter" standoffs in critical areas by swapping around standoffs on the block. There is simply no way to make the VRM contact the plateau with paste. Furthermore, maximum compression of the block to the card was done in that there were no gaps at all between any standoff and the PCB, or simply put the card was as flush to the Komodo block as it can physically get.

I agree completely that you should try to revisit the standoff heights to lower the tolerance gap without pads. As it is now you have a fair amount of tolerance, at least in my testing. Furthermore, I would ensure that the machining height deviation tolerances on standoffs and VRM plateau areas be virtually eliminated as not doing so results in potential problems with contact due to the fine measurements we are dealing with here. Basically, you should strive to have all standoffs be the exact or as close to exact height that they need to be as possible as well as ensuring even and correct height of the VRM cooling plateaus of each block.

I am waiting for the RMA MSI R5850 to come back as well was for an Asus EAH5850 card to arrive. I will repeat my testing with those cards as well to ensure that this was not just a rare occurrence with some MSI R5850 cards. However, having said that this should not be taken as a reason to not address this at all. I realize that you have successfully tested this method in house, but both you and I know that not all cards will be identical with identical heights and to recommend a blanket method based on a gross assumption that it would work in every case is unrealistic at best. Even with reduced tolerances there should be a notice indicating that the performance may not be as desired.

Finally, one option that you do have is to include some longer fasteners that could be used to mount the block without standoffs and indicate that this is considered advanced installation with higher risk of damaging the card but would still leave the option for those people who may want to mount the whole block without pads and standoffs to do so. After all, in terms of its thermal design Komodo is really a great block and can pretty well be a top performer if it were not for a few of these issues.

It is imperative as you already know to be extremely careful here, no matter which path is chosen, as crushing digital VRMs is very easy to do and that's the last thing you want people to do after trying to get better temperatures out of their VRMs.

Hello,

The indention you have marked, "Bad contact, no compression" with a red line with two arrows on each end, can you remove the pad sitting inside that indention to give your block more height/clearance to successfully cover the components on the pcb marked by the white sticker, "CPL2-3". Then you may have greater compression/pressure on the 5 VT1165 slaves to the left which are not being touched by your cooling block. Also, can you make your cooling block mount tighter to that area of vrm slaves?

Why not use a highly conductive thick thermal pad of good quality to mount on those 5 components. These 5 slaves need *MUCH* more cooling than the parts under the white sticker CPL2-3, which require little cooling, and are not even touching the stock cooler.

I think you misunderstood what the issue here is actually. Komodo has no problem contacting the VRMs with the stock pad or with any other pad thicker than (I believe) about 0.3mm-0.5mm. However, as of recently Swiftech has revised their mounting instructions to include an advanced step that involves removing the thermal pads from the VRM cooling areas and replacing the same with paste for better thermal interface. However, unlike in Swiftech testing, I have not found it possible to use this method with at least one HD 5850 and two Komodo blocks as the dimensions of the standoffs do not allow for proper contact to be made between the block and the card even if fully compressed.

In terms of removing pads to give more clearance, I have already responded to this earlier, and then Gabe responded to it as well. The clearance on Komodo is not controlled by pads, it is controlled by standoffs. The only way to make the block more flush to the card is to use smaller standoffs or to file down the existing ones, both of which requires a physical change to the current design.

[gabe]

Finally, one option that you do have is to include some longer fasteners that could be used to mount the block without standoffs and indicate that this is considered advanced installation with higher risk of damaging the card but would still leave the option for those people who may want to mount the whole block without pads and standoffs to do so. After all, in terms of its thermal design Komodo is really a great block and can pretty well be a top performer if it were not for a few of these issues.

Excellent suggestion, I will look into it. I am still quite puzzled by your problem. Do you want a set of reduced height standoffs? I can't do it til i get back next week, but it's feasible. email me at gabe@swiftech.com if you do.

[dejanh]

Excellent suggestion, I will look into it. I am still quite puzzled by your problem. Do you want a set of reduced height standoffs? I can't do it til i get back next week, but it's feasible. email me at gabe@swiftech.com if you do.

Email sent. In terms of the problem it is very well possible that this was, is, and will be one of those rare outlying cases, that the card was just not quite the same as other reference HD5850 cards. There are variations between manufacturers sometimes as was for example the case with HD5970 cards. Some of those had some components higher on some PCBs than others creating a contact issue. I will redo all the tests as soon as I have the new card(s) in hand to verify the problem with other cards. In the meantime, those standoffs would be helpful since it would (a) allow me to do a proper mount even if I still have the same problem, and (b) directly compare the result with stock standoffs to reduced standoffs.

[Philwong]

My comment was only intended as a rebuttal to your statement of pushing the card to the extreme. I hardly pushed the card to the extreme, and personally I even stated that I do not think the card died because of VRM temperatures but rather all of the mounting and remounting. Anyway, the main reason I took everything down again last night is so that I can get data for Gabe to show him the problem. I got the data, but unfortunately I also got a dead card now. I had no choice anyway though but to take down the loop again last night since the VRM temperatures were out of control. I was intending to replace the paste with pads again but once everything was done the card would no longer start. I'm loading the images and doing the write up now, they will be up shortly.

Edit: @Philwong, I forgot to mention something. Furmark got nerfed by ATI in their drivers after I reported a problem with it overheating the VRMs at launch of HD 4800 series cards several years ago. I realize that Furmark is not representative of normal load however it is a great test tool and shows problems quickly. After the VRM debacle that happened with the HD 4800 series cards I did not expect a repeat of the problem, and as you can see below I was right. Furmark or not, the block is not making contact as it should be so the VRMs burn, burn, burn

If running Furmark at such extreme settings is not pushing your card to the limit with all the warnings of average VRM temps delivered by the Komodo, I don't know what else to say. I will be pissed too if I zapped my card so I do feel for you. But what else can we Komodo owners do when the coolant is not flowing anywhere close to the VRMs? It's not some rocket science that the Ek shines in this area, while the Swifty aced in the GPU department with its micro pins.

Phil