AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread

I thought that'd be a given, otherwise why the need for water cooling?

(Who's Gerald Marley and why has he this picture?)

Im not sure, there's a need for watercooling, after all, custom coolers gpu's deal with way more amount of power thermal dissipation than that,

Buts its cool, and allow them more flexibility .. Everyone do AIO water coolers, Corsair, Asetek, Intel, ( since the I7 4900 series ), cooler masters, every brands present their pc with them for the cpu, or the gpu.. just have to look on any brands presentations of their new case, their last motherboard in tech meeting.. Look on every tech presentation what you have his is a cpu cooled by this type of all in one H2o ..

Look like AMD have find some argument with it.. dont need to do much work on the cooling part, customizable, good temps, marketing, this seems a good standard now for enthusiast gamers who dont have the knowledge or dont want set up their own custom h2o loops.

Personally i make my custom H2o loop since more than 15 years .. so the performance of this type of AIO dont interest me..

WIth HBM, it make even more sense for cool the core+vram.. you have 1 waterblock, who cool a closed surface, outside, the only one part to cool is the PWM .. it simplify everytthing...


And like write gamervivek, it will also help keep an higher boost clock ..

Something with custom h2o setup like i use, way better of any AIO, is my GPU's temp difference between idle and full load ( benchmark or vray tracing ) is only of 3-4°C delta .. basically if my GPU's idle at 30°C, the maximum temp at full load is 34°C, and like every parts of my GPU's are watercooled ( core, vram, PWM ), you get the same temps in every parts, whatever it is the ram, the core, the pwm.. with extremely low variation compared to "standard" air system between idle and full. ( where you can have 90°C load for PWM, and 40 for idle, 30°C for the core and 70°C full load ). You end with really good stability let alone longevity.
 
Last edited:
Remember, TDP and thermal dissipation aren't directly linked, and AMD has had trouble with a lot more thermal dissipation per watt than NVIDIA even before Maxwell. My guess is that the stacked HBM, while using less power, builds up a lot of heat in a small area, requiring better heat dissipation via a watercooler. It's the same reason I suspect 2.5/3D silicon stacking isn't somehow going to benefit GPU's other than memory, because GPUs are already super wide by nature and both Nvidia and AMD already get thermally bound at the top end, trying to stack the GPU die itself would probably just cook the entire thing.
 
At the end of the day, we know that 300-350W can relatively easily and cheaply be cooled with conventional cooling.
Maybe there are good reasons to go with an expensive water cooler even though it's not strictly needed, but it's not the most logical case to make.
 
Here are some other possibilities, with a few overarching themes:

This is an expensive interposer-based GPU, so volumes are limited.
The memory capacity is limited, which undercuts it in the professional and compute realms, further reducing its volume.
The pictured card already appeals to an upper tier boutique enthusiast market, where regular power budgets and having sense commensurate with money have a lower incidence than the general population.
There is potentially Nvidia competition that might with decent effort nearly match or beat it, so there needs to be an at-launch first-party exemplar that is hardest to beat, regardless of the small volumes.

If there is a mere-mortal 6+8 pin solution with air cooling, anything above that will be even smaller in volume.
There is pretty much no room between this lower sliver of a niche and anything but the AIW pictured, with a shroud that looks much less flattering in that pic and at that angle. I am not sure if I want to post my grade school drawing on it or eat an ice cream sandwich.

We know what AMD's own air coolers are, currently, and they do not fit that upper niche.
What custom engineering for a super-enthusiast air cooler, with the matching premium noise profile, and necessary performance on product that will likely have the highest power density AMD has ever produced, for the press-sample SKU, is going to be justified for the hundreds or thousands of units?
AMD's manufacturing variances for its air coolers have been disastrous for Hawaii, which appears to be less of a thermal challenge.
If AMD flubbed the physical/thermal/DVFS design, even less leeway may be available.

A water cooler rebranded from a more competent manufacturer would have greater thermal slack, less NRE, better engineering than AMD has managed, and its own volumes outside of this prestige product.

Sure there could be some third party boards, with third-party air coolers AMD has no interest in competing against, and then there are outside water-cooling solutions and full system loops that don't get into launch reviews.

There may be some new air cooler at some point for HBM GPUs, possibly. Possibly, it would not work with a product in this design corner in this product year.
 
ok, ok, I got it. ;)

There's one thing though: AMD has been telling during multiple conference calls that they expect better financials in part due to new GPU released in H2 2015. With a full rebranded lineup and limited volume Fiji, I don't see how that's possible.

With that out of the way: why is that the 8 pin PCIe power supply connectors can carry more power than those with 6 pins? Those 2 extra pins are both carrying GND.
 
Last edited:
The sites I've looked at seem to indicate that the extra pins do not carry current, but their presence serves as an electrical indicator to the GPU that the power supply on the other end can actually handle the additional load.
It may be more of an electrical/physical keying with potentially downwards compatibility if it's 6+2, where the PSU is obligated to meet the requirements if it has that connector available.
 
Maybe there are good reasons to go with an expensive water cooler even though it's not strictly needed, but it's not the most logical case to make.
300-350W air coolers are fucking loud at full load. They have to be, because of the thinness of the fans. If there was more room for the cooler, thicker fans and heatsinks could be used, but cardslots don't lend themselves to that kind of design. Essentially, modern add-in board GPUs have totally the wrong form factor for the amount of power they draw. They should be like old notebook MCMs or these newfangled PCIe flash SSD sticks, laying flat against the mobo so that a tall cooler could be mounted on top. But alas, we're stuck with this shitty old IBM PC/XT structure which was designed for cards drawing mostly single or low double digit power.

So I'd say a watercooler makes excellent sense.

These coolers aren't particularly expensive either. There's nothing complicated about them whatsoever, all the tech in them is many decades old already; radiator is made using same tech that has been used in car production for example since basically forever, pump isn't expensive or complex either and so on. It costs a little more than an air cooler perhaps, but a bigass aircooler with fat heatpipes aren't cheap either. There's a lot of metal in those things, bent into complex shapes, which has to be made and fitted with high precision to work correctly.

The reason these closed water coolers cost a fair bit is because their higher 'value', derived from higher performance at lower noise levels.
 
AMD is probably pushing to beat the titan x and needs the 2 8 pins. They will probably have an air cooled sku below that is more efficient.
 
At the end of the day, we know that 300-350W can relatively easily and cheaply be cooled with conventional cooling.
Maybe there are good reasons to go with an expensive water cooler even though it's not strictly needed, but it's not the most logical case to make.

I don't think it's ever been demonstrated on a card this short, let alone quietly.
 
The sites I've looked at seem to indicate that the extra pins do not carry current, but their presence serves as an electrical indicator to the GPU that the power supply on the other end can actually handle the additional load.
It may be more of an electrical/physical keying with potentially downwards compatibility if it's 6+2, where the PSU is obligated to meet the requirements if it has that connector available.
The 2 pins are both extra grounds
 
I don't think it's ever been demonstrated on a card this short, let alone quietly.
Depends on which was first, the chicken or the egg: they could have chosen a card this short because they were forced to use water cooling anyway.
Anyway, we'll see. It's probably going to be 300W territory...
 
GTX 480 could have been named such for the amount of power it drew in certain scenarios, and it aircooled just fine. :)
 
The 2 pins are both extra grounds

It was claimed that the extra pins could be spoofed for some users that wanted to fit an 8-pin card where there wasn't a connector. My initial impression was that this meant nothing was overly dependent on them, but that might mean that the cards in question were not frequently drawing enough to tax the 8-pin or something undesirable was being done to the remaining grounds.
I personally haven't had the desire to try it, so it isn't a well-vetted data point.
 
Depends on which was first, the chicken or the egg: they could have chosen a card this short because they were forced to use water cooling anyway.
Anyway, we'll see. It's probably going to be 300W territory...

It's probably going to be higher, at least as an option for overclockers, otherwise there would be no need for two 8-pin power connectors.

I think one of the photos of Fiji shows a small switch on the card itself. My guess is this is some kind of power limiter. By default, the card might be limited to 300W, with the option to take it to 375W (or more) by flipping the switch.
 
It was claimed that the extra pins could be spoofed for some users that wanted to fit an 8-pin card where there wasn't a connector. My initial impression was that this meant nothing was overly dependent on them, but that might mean that the cards in question were not frequently drawing enough to tax the 8-pin or something undesirable was being done to the remaining grounds.
I personally haven't had the desire to try it, so it isn't a well-vetted data point.
Not just claimed, it was done, too. You could use metallic paperclip to fool 6-pin for 8-pin, the paperclip was connected to nothing but the 2 extra "slots" in the cards connector
 
Khronos site does have 8800M as a opencl2.0 compliant and apparently it only works on GCN1.1 with 1.0 limited to opencl1.2

And he says later,
"The device ID of the 8870M and M370X is 0x6281."
 
Back
Top