Hardware stub

Corsair & MSI Change Sea Hawk Coldplate to Protruded Model

Posted on August 31, 2016

In Corsair's recently released Hydro GFX marketing video, we noticed that the video card on display used the protruded coldplate that we've been talking about since the 980 Ti Hybrid. That plate was recently put to the test in our GTX 1080 Hybrid vs. Sea Hawk review, where we found the protruded unit performs marginally better than the flat plate shipping with the Sea Hawk / Hydro GFX. We reached out to Corsair to discuss the change spotted in the marketing video, hoping to understand why the unannounced* (officially) modification was made, and have outlined the email responses below.

This seems largely to be a non-issue for users who purchased their cards from the official Corsair website, though we do have some contingencies for MSI Sea Hawk buyers. Note also that the temperature difference we spotted between the coolers is partially a result of new information we received regarding the Hydro GFX, primarily that the coldplate had its standoffs machined down by MSI prior to shipment. These machined standoffs have a larger tolerance (~0.2mm) for height than we've seen in from-factory Asetek CLCs (~0.05-0.08mm), which means mounting pressure could contribute to marginal thermal differences.

The video breaks things down most readily, but continue reading if preferred.

Our emails to Corsair included the below Q&A:

GN: Why make the change?

Corsair: “Some of the first units of Sea Hawks and Hydro GFX used the flat base with machined down standoffs to provide better contact with the GPU surface. We changed immediately to the raised GPU cold plate. We had used the CPU cold plate in the previous version with no issues. That did not work for the new 10 series as nVidia changed some dimensions for production silicon that were not seen on pre-production parts.”

GN: Why not make the running change in a way that just shortens the stand-off height? Why change the copper coldplate rather than have the standoffs built to a shorter spec?

Corsair: “It’s a trade-off between tooling a new standoff or taking an off the shelf cold plate that already existed for the solution. Taking the off the shelf solution was the best choice when looking at the options available at the time.”

GN: Will Corsair publicly state this change?

Corsair: “As the function and feature of the card has not changed and both cold plate solutions meet the stated spec of the card, we don’t plan to split them as a different SKU. We’ve not upgraded or cold plate – it’s just a slight tweak to accomplish the same result given the resources available at initial dev, and extended production.”

GN: If I buy this card today, what cold plate will I receive in the mail in three days?

Corsair: “Honestly, could be either. Without MSI giving us the serial numbers and lot codes of units produced with which material we would not know. The sample for the CORSAIR video and your review sample were from the same batch. Only the initial product run use the machined stand-offs. All later units going forward use the GPU cold plate.”

GN: Was the shaving modification made to all affected production standoffs?

Corsair: “Every production unit that shipped with the CPU cold plate had the shaved standoffs.”

GN: How many users purchased the first production run?

Corsair: “85 of the first 100 CORSAIR Hydro GFX GTX 1080 went to end users.”

GN: As we saw in our tear-down video, the Hydro GFX had OK contact with the coldplate (looking at TIM), but not as flush as we've seen from other units. Either this factor or the copper protrusion -- or both -- is contributing to the performance swing we have seen in the coolers. Not to fight semantics, but "minimal" impact is not "impact within real-world use;" we are seeing a repeated, measurable difference between the plates. It is perhaps insignificant at the end of the day -- talking a degree, here -- but I do want to make clear that there is performance improvement by sticking an EVGA cooler + Corsair fan onto our Hydro GFX card. As for whether that difference is made by the protrusion or because of poorer contact by the Corsair coldplate to the GPU, I am not presently sure. I could perform additional testing with a larger sample size if I had access to one.

Corsair: “In our own cooling using a TTV we saw very little difference between the CPU cold plate and the GPU cold plate for degrees C per watt cooling performance. The contact of the cold plate to the GPU die is the only difference between the two. Mounting will affect performance every time. We’re confident every Hydro GFX GTX 1080, and I’m sure MSI would agree re: SEAHAWK cards, meets the performance specifications that we and MSI publish for the card. There will be variance in performance from card to card and some will perform better than others, but every card meets its published spec guaranteed.

While I appreciate you might be able to find some small differences between the cold plates, there’s minimal difference to the end user from either. These small differences between initial and extended production are just the reality of getting a product to market as soon as we could to meet demand, and then optimising the solution to meet Nvidia’s moving target silicon specs.”

Thoughts

Based on these answers, Corsair and MSI encountered an issue where the silicon was not tall enough (an issue we saw with the 1060 Hybrid mod) to meet the flat coldplate used in the initial review unit. Corsair, for this first production run, worked with MSI to machine-down the standoff height to close the gap. The contact, as we stated in our review, appeared acceptable but not quite as even/flush as we're used to seeing. MSI was responsible for truncating the standoffs. Here are the values we measured:

The standoff heights on the H55 CLC are:

  1. 5.77mm

  2. 5.71mm

  3. 5.91mm

  4. 5.82mm

The standoff heights on a from-factory Asetek cooler for GPUs:

  1. 6.32mm

  2. 6.33mm

  3. 6.38mm

  4. 6.35mm

Because of this change, Corsair was able to resolve any potential issue with their engineering oversight and ensure that, theoretically, end users do not receive a product which has a 0.5mm gap between the GPU silicon and the coldplate. This is an unfortunate oversight by Corsair and MSI, but one which appears to have been resolved while running. We would chalk this up to a “non-issue” for the most part, with thermal differences likely seeing marginal improvement on the protruded copper coldplate (second production run onward). Corsair disagrees with us here, based on these emails, but we stand by our tests and would argue that – minimally, if not because of the thicker copper plate – the protruded model will outperform a unit with 0.2mm height tolerances on the standoffs. That difference could create a mounting pressure difference which contributes to the ~1C performance improvement with the protruded coldplate.

For owners of MSI cards, we recommend installing AIDA64 or GPU-Z, then running a stress test (3D Mark or a game will suffice) for ~20 minutes. Check the average load temperatures during this time. If your unit is exceeding 60C, given a reasonable ambient of ~20-25C, we advise that you contact MSI to discuss the issue. Alternatively, contact GN, and we will help you troubleshoot. There should be no issue here, from what we're told, but Corsair did not answer for MSI's units shipped, and so we have no hard data on whether any Sea Hawk units made it out of the factory without the shaved-down standoffs.

Note that this change means our review of the Hydro GFX card will not feature identical cooling components to what is found on later production models. This will change thermal results slightly, though not necessarily significantly.

To reiterate: We see the matter as largely a non-issue, but do encourage early MSI Sea Hawk owners to validate their temperatures. Corsair Hydro GFX owners appear to be in good shape, based on the replies to our email.

* = Original word choice was "silent," as in "silent announcement," as the change was not accompanied by a public statement and MSI has not yet provided insight to the change on their end. Upon further consideration, we have changed this word to "unannounced," because Corsair has been rather vocal to us about the change; we appreciate that dialogue.

Editorial: Steve “Lelldorianx” Burke
Video: Andrew “ColossalCake” Coleman