AMD Radeon Fury X Series | HBM, Small Form Factor And Water Cooling | June 16th

Hey guys, I have some questions about the Fury X,

1. Do I need to have a fully water cooled system in order to install the Fury X?

2. Since it is a water cooled card, do you need a certain case to install this type of card?
1. No
2. You just need a case with enough room to install that huge radiator, so you should be fine. You might have to remove a side case fan, but that's it.
 
1) No it's an all in one cooler. It doesn't matter how the rest of your system is cooled.
2) You case only needs a 120mm fan mounting position available. Your case should be fine.
 
This bit is interesting:



I think by 'operating' they mean load temp. But this suggests the cards overclock much better than a Titan X, and are also faster at stock if the leaked benches are correct. Downside is power draw is higher. But at least temps look great.

If it can OC like a mother fucker and still stay cool and quiet, it will be the card to beat, bar none.
 
1. No
2. You just need a case with enough room to install that huge radiator, so you should be fine. You might have to remove a side case fan, but that's it.

1) No it's an all in one cooler. It doesn't matter how the rest of your system is cooled.
2) You case only needs a 120mm fan mounting position available. Your case should be fine.

Thanks for the replies guys, that sounds really easy. Just got to wait for those benchmarks though.
 
Is it true that the Fury X is supposed to have a fairly small memory clock speed, despite the 512 GB/s bandwidth? Some places have been tossing around 500-1000 MHz, and that kinda makes it sound like the GPU's bandwidth would vary greatly with overclocks.
 
Is it true that the Fury X is supposed to have a fairly small memory clock speed, despite the 512 GB/s bandwidth? Some places have been tossing around 500-1000 MHz, and that kinda makes it sound like the GPU's bandwidth would vary greatly with overclocks.
The physical memory clock is rather low, but the bus is extremely wide. Basically, the card simply has "enough" BW ;)
 
I just hope the availability is good. Nothing like launching a (seemingly) desirable product and there's only a small number to go around. We've heard rumors about yield issues and I'm guessing this is why the Fury is launching next month and the Fury Nano at some point after that (I'm guessing end of August).
 
I just hope the availability is good. Nothing like launching a (seemingly) desirable product and there's only a small number to go around. We've heard rumors about yield issues and I'm guessing this is why the Fury is launching next month and the Fury Nano at some point after that (I'm guessing end of August).

In the UK, I think OverclockersUK is a good bet to get one, as they took pre-orders on all the 980 Ti cards. I've contacted them about when they're taking pre-orders.
 
I hope MicroCenter will have it. I'll order it for a pick up right as it shows up on the site and pick it up the next day (like if it's midnight or something).
 
I always thought the big hitch with 4K gaming was the amount of VRAM. Will faster vram at 4gb be better for 4K than more albeit slower vram?
 
While it is true that PSU can supply that amount, you definitely shouldn't be drawing that amount of power on a regular basis. One should target at least a 20% margin between the rated power supply and ones own power requirements. At 50~60% load the PSU is the most efficent, typically.

Of course, but a modern unit at 85-90% that is sitting at 650w and 85% is going to cost 764w at the wall rather than 722w at 90%. 40w isn't nothing, but then you're computer also isn't going to be sitting at 650w permanently while gaming if you're only running a 650w PSU.

With a 290X and an intel CPU your entire system might only draw 450w even if you've overclocked a bit, and even then only at peak loads. A 550w PSU in that instance is more than sufficient.
 
Would a Fury Nano be workable space/heat/power-wise in a Silverstone Sugo SG-05? It has a 300W 80-rated PSU, and the only components I have in there atm are an i5-3570, 8 GB RAM, and an SSD.
 
I guess. In a year or two (more like 6 months) my SLI TITAN X setup will be sold off and running in someone else's secondary box. Future proofing is STUPID. This is one of the primary arguments of the "moar VRAM" crowd.

4K/5K/8K slideshows...no thanks. Give me 1440p/144Hz.

Not necessarily, sometimes people have different situations. I am a software developer so when I buy a new PC, its very high end (eg 32 GB RAM, huge storage, high end CPU) and lasts years. So its an expensive PC anyway, so incremental costs for future proofing make more sense in this situation (since the % price increase is relatively smaller vs building a pure economical gaming PC)
 
Would a Fury Nano be workable space/heat/power-wise in a Silverstone Sugo SG-05? It has a 300W 80-rated PSU, and the only components I have in there atm are an i5-3570, 8 GB RAM, and an SSD.
lets see ... 77W for CPU, 10W for RAM, a watt or two for SSD, so maybe 100 Watts with fans n' stuff + headroom.
That leaves 200ish Watt for the GPU. The Nano has 175W TDP. Cutting it close.
 
Would a Fury Nano be workable space/heat/power-wise in a Silverstone Sugo SG-05? It has a 300W 80-rated PSU, and the only components I have in there atm are an i5-3570, 8 GB RAM, and an SSD.

I've got a 600W PSU. Do you think that will be enough to run a Fury X comfortably then?
 
lets see ... 77W for CPU, 10W for RAM, a watt or two for SSD, so maybe 100 Watts with fans n' stuff + headroom.
That leaves 200ish Watt for the GPU. The Nano has 175W TDP. Cutting it close.

Damn... what about with some underclocking? It seems like the main alternative is still the 750ti, haha.
 
I always thought the big hitch with 4K gaming was the amount of VRAM. Will faster vram at 4gb be better for 4K than more albeit slower vram?

VRAM is good enough as long as you don't use it up. Performance will usually fall off a rock when a game exceedes it and has to put stuff on the system memory (which is much slower). If a game is fine with 4GB though adding more memory won't change a thing though. Memory bandwidth is very important for good performance at higher resolutions and has been a bottleneck lately. You could with Kepler vs GCN in 2012 (where AMD put bigger memory interfaces on competing GPUs) see how Kepler was superior at lower resolutions and AA while GCN did better when those were cranked up.
Framerate-1920x1080-FPS.png

Would a Fury Nano be workable space/heat/power-wise in a Silverstone Sugo SG-05? It has a 300W 80-rated PSU, and the only components I have in there atm are an i5-3570, 8 GB RAM, and an SSD.
You could test how much your system draws without it at load (like prime95). Remember that if the PSU is 80% efficient then ~20% of the power drawn from the wall isn't from the system itself.
 
I've got a 600W PSU. Do you think that will be enough to run a Fury X comfortably then?
Very likely yes, with 300 ish Watt TDP or so. But depends on what else you might have of course.
Damn... what about with some underclocking?
Doubt it will change much, the Nano is already underclocked and the rest doesn't need that much.
Probably better to buy a new PSU, they don't cost a ton compared with a new Fury graphics card.
 
What's really awesome about these cards is that the memory stacks in crossfire. So 2x fury is a much better set up than 2x 980 ti.

Just bought a 980 ti, might be selling it.

A very interesting possibility would be a 2x r9 nano set up, especially if they come out at around $400. Would completely destroy the titan x.
 
What's really awesome about these cards is that the memory stacks in crossfire. So 2x fury is a much better set up than 2x 980 ti.

Just bought a 980 ti, might be selling it.

A very interesting possibility would be a 2x r9 nano set up, especially if they come out at around $400.

It stacks? So it would be working as 8GB instead of two separate 6GB sets?
 
Damn... what about with some underclocking? It seems like the main alternative is still the 750ti, haha.

I'd wait for reviews to come out so we have an idea of boards made and power consumption.

It will probably be possible to underclock it and it might be that you won't hit the PSU's limit. I see posts from people having fitted 5870s and 560tis inside the SG05 with the 300W PSU and they are around 175W as well. CPUs back then also used more power than the 3570.
It stacks? So it would be working as 8GB instead of two separate 6GB sets?
No. It will be marketed as 8GB, but each card has to use their own bank of memory so effectively they have 4GB.
 
What's really awesome about these cards is that the memory stacks in crossfire. So 2x fury is a much better set up than 2x 980 ti.

Just bought a 980 ti, might be selling it.

A very interesting possibility would be a 2x r9 nano set up, especially if they come out at around $400. Would completely destroy the titan x.

Care to link? I've never heard anything about it stacking in crossfire.
 
What's really awesome about these cards is that the memory stacks in crossfire. So 2x fury is a much better set up than 2x 980 ti.

Just bought a 980 ti, might be selling it.


A very interesting possibility would be a 2x r9 nano set up, especially if they come out at around $400. Would completely destroy the titan x.

Huh? This isn't true.

Memory doesn't stack in crossfire, it stacks in DX12, which means that it will stack with both Nvidia and AMD cards. So 2x 980 ti's would still be more memory.
 
Huh? This isn't true.

Memory doesn't stack in crossfire, it stacks in DX12, which means that it will stack with both Nvidia and AMD cards. So 2x 980 ti's would still be more memory.
The game has to be specifically designed for it though. The defacto way of using multiple GPUs has for a long time been Alternate Frame Rendering so I don't think it's going to change over night unless multiGPU setups become really popular.
 
The game has to be specifically designed for it though. The defacto way of using multiple GPUs has for a long time been Alternate Frame Rendering so I don't think it's going to change over night unless multiGPU setups become really popular.

Oh, of course. I was just pointing out that memory stacking isn't an AMD feature but a DX12 feature.
 
Oh, of course. I was just pointing out that memory stacking isn't an AMD feature but a DX12 feature.

Ah, sorry my bad. I thought it was a mantle feature. I hope AMD will make it work on a driver level though because of the dual fury chip coming out this year.

Do we know if the 8gb dual fury will be using some kind of hardware interposer for full 8gb? Is it possible to enable hbm stacking across crossfire on a driver and hardware level?
 
Hate to be lazy, but there are so many new card options that I'm a little overwhelmed.

Which of these cards should I be looking at if I'm wishing to build a VR-capable machine? Power vs. Price would be ideal, as I imagine price will be outrageous given Aussie' Pricing.
 
Hate to be lazy, but there are so many new card options that I'm a little overwhelmed.

Which of these cards should I be looking at if I'm wishing to build a VR-capable machine? Power vs. Price would be ideal, as I imagine price will be outrageous given Aussie' Pricing.

Whatever the best card you can afford is.
 
Ah, sorry my bad. I thought it was a mantle feature. I hope AMD will make it work on a driver level though because of the dual fury chip coming out this year.
It's not a "Mantle feature", and it's not a "DX12 feature" either. It's certainly not an AMD feature or an Nvidia feature for that matter.

If a developer manually supports multi-GPU (which they have to in low-level APIs) then they can decide how to distribute their data, and they may decide, depending on their rendering setup, not to replicate everything. But it's also certain that you won't get a straightforward additive combination of available memory. Some assets/buffers will always need to be replicated.
 
It's not a "Mantle feature", and it's not a "DX12 feature" either. It's certainly not an AMD feature or an Nvidia feature for that matter.

If a developer manually supports multi-GPU (which they have to in low-level APIs) then they can decide how to distribute their data, and they may decide, depending on their rendering setup, not to replicate everything. But it's also certain that you won't get a straightforward additive combination of available memory. Some assets/buffers will always need to be replicated.

Do we know if the 8gb dual fury will be using some kind of hardware interposer for full 8gb? Is it possible to enable hbm stacking across crossfire on a driver and hardware level? Is hbm different in that sense at all? Other people in this thread and elsewhere on the net people seem to think so. Is that just speculative misinformation?
 
Do we know if the 8gb dual fury will be using some kind of hardware interposer for full 8gb? Is it possible to enable hbm stacking across crossfire on a driver and hardware level? Is hbm different in that sense at all? Other people in this thread and elsewhere on the net people seem to think so. Is that just speculative misinformation?
I don't see any reason why HBM would be different from any other memory with regard to how it scales with multiple GPUs.
 
I am so glad I waited. I want a low-power, low-temp upgrade for my HD7770 that will be capable of running stuff in 1080p, or sub-1080p, on low settings. Graphical fanciness does not interest me at all.

So this will be right at my alley.
 
Do we know if the 8gb dual fury will be using some kind of hardware interposer for full 8gb? Is it possible to enable hbm stacking across crossfire on a driver and hardware level? Is hbm different in that sense at all? Other people in this thread and elsewhere on the net people seem to think so. Is that just speculative misinformation?

Code:
[IMG]http://www.techpowerup.com/img/15-06-16/159a.jpg[/IMG]

This is what a dual Fiji looks like. It's pretty much the same as any other dual GPU, just smaller. So yeah, just two packages instead of one big interposer. I don't think they can even be big enough to fit two GPUs and a bunch of HBM stacks.
 
Let's recap a bit.

Nah, let's not. I see now what Durante means.

What's really awesome about these cards is that the memory stacks in crossfire. So 2x fury is a much better set up than 2x 980 ti.

Just bought a 980 ti, might be selling it.

A very interesting possibility would be a 2x r9 nano set up, especially if they come out at around $400. Would completely destroy the titan x.

It's the same as right now with Fury. The effective memory pool is split between the number of chips. The interposer tech may lead to something else eventually but not in this case.
 
I am so glad I waited. I want a low-power, low-temp upgrade for my HD7770 that will be capable of running stuff in 1080p, or sub-1080p, on low settings. Graphical fanciness does not interest me at all.

So this will be right at my alley.

The Fury Nano would be the best pick for that. Overkill for sub 1080p though.
 
Be very careful about the vram across gpus becoming one.

Its only on DX12 and so far we havent saw any real world examples of this.

Pick your salt rock.
 
Top Bottom