AMD - A New Era In PC Gaming

[Opcode]

That's quite a clever way to cool the VRM, if I'm honest. No exposed moving parts on the GPU to get clogged up with dust, you just need to clean the rad every now and again

Yeah, when I seen it yesterday only one question popped into my head. Does the block feed the VRM or does the VRM feed to block. I presume water flows to the block first as VRM doesn't take much to cool. Either way AMD states the card runs at < 50C under a normal load. Hopefully their tweaked architecture has a lot of overhead as this thing's ready to be toyed with. I'd be impressed to see people hitting even 1.3 GHz core with it. What interests me is HBM overclocking. Even the slightest 50 MHz overclock will make a marginal difference in bandwidth.

[Sakkura]

Yeah, when I seen it yesterday only one question popped into my head. Does the block feed the VRM or does the VRM feed to block. I presume water flows to the block first as VRM doesn't take much to cool. Either way AMD states the card runs at < 50C under a normal load. Hopefully their tweaked architecture has a lot of overhead as this thing's ready to be toyed with. I'd be impressed to see people hitting even 1.3 GHz core with it. What interests me is HBM overclocking. Even the slightest 50 MHz overclock will make a marginal difference in bandwidth.

 

On the other hand, it's not like memory bandwidth is going to be constraining performance much at all. It's got redonkulous bandwidth at stock settings already, along with the bandwidth-saving compression of the newer GCN revision.

[Opcode]

On the other hand, it's not like memory bandwidth is going to be constraining performance much at all. It's got redonkulous bandwidth at stock settings already, along with the bandwidth-saving compression of the newer GCN revision.

I know the card is far from bandwidth starved, although I'm just interested in how much bandwidth you can squeeze out of the memory technology with overclocking.

[shawno]

I know the card is far from bandwidth starved, although I'm just interested in how much bandwidth you can squeeze out of the memory technology with overclocking.

Very valid point re: overclocking the memory.

It will be interesting to see how much you can overclock hbm and wether it handles the extra pressures. Will there be a higher dead card rate or will it handle more stress than gddr5?

[Sam Z Man]

Very valid point re: overclocking the memory.

It will be interesting to see how much you can overclock hbm and wether it handles the extra pressures. Will there be a higher dead card rate or will it handle more stress than gddr5?

Well because it's stacked it has to cool the bottom stack by transferring that heat to the second stack, then transfer the heat of thoes two stacks to the next stack, and so on, so I feel like it's going to have heat problems if overclocked significantly, it's like trying to cool a DDR5 sandwich by putting a cooler on only one end.

[Notional]

Well because it's stacked it has to cool the bottom stack by transferring that heat to the second stack, then transfer the heat of thoes two stacks to the next stack, and so on, so I feel like it's going to have heat problems if overclocked significantly, it's like trying to cool a DDR5 sandwich by putting a cooler on only one end.

 

Actually HBM has thermo micro bumps between the layers of memory in the stack, that disperses heat.

[Sam Z Man]

Actually HBM has thermo micro bumps between the layers of memory in the stack, that disperses heat.

How would that change anything if the only thing cooling it is what is on the top of the stack.

[stconquest]

How would that change anything if the only thing cooling it is what is on the top of the stack.

 

On top, or surrounding?  I could see the housing acting like a heat sink.

 

..oh, and I have absolutely no idea BTW

[Sam Z Man]

On top, or surrounding?  I could see the housing acting like a heat sink.

 

..oh, and I have absolutely no idea BTW

It's okay I have no idea either, just trying to think about it logically.

[Notional]

How would that change anything if the only thing cooling it is what is on the top of the stack.

 

Have a shit google link, because frack google: http://www.google.dk/url?sa=t&rct=j&q=&esrc=s&source=web&cd=7&cad=rja&uact=8&ved=0CEYQFjAG&url=http%3A%2F%2Fwww.hotchips.org%2Fwp-content%2Fuploads%2Fhc_archives%2Fhc26%2FHC26-11-day1-epub%2FHC26.11-3-Technology-epub%2FHC26.11.310-HBM-Bandwidth-Kim-Hynix-Hot%2520Chips%2520HBM%25202014%2520v7.pdf&ei=FRuCVaGYCsiOsgHSnaTwDA&usg=AFQjCNHaQNpNWJQd40GAaGeW_8d1Yrj58g&sig2=s__V2lsstZgIZGnLYhG3Zw

 

Watch page 21 on thermal dissipation. The micro bumps are used to channel heat, so the memory chip itself, won't absorb all the heat.

[Opcode]

Here's the driver for Fury X and the 300 series.

 

The Catalyst 15.15 drivers bring:

• FPS Targeting: The ability to save power by setting a maximum FPS target
• Virtual Super Resolution: The ability to downscale a game from a high resolution to a lower resolution, giving a user higher-quality textures in game (also known as supersampling)
• Performance optimizations: Adding game performance optimizations to AMD R9 Fury X and other graphics cards

Catalyst 15.20 will be the release WHQL driver with also DirectX 12 and Windows 10 WHQL support.

 

How would that change anything if the only thing cooling it is what is on the top of the stack.

It's a little more complicated than you would expect.

 

Well you cannot forget that HBM has it's own thermal management and dummy bumps for thermal dissipation. HBM has built in sensors and as the stack starts getting hotter and hotter the more often it will cycle its heat to the PCB. Temperatures around 85C are normal for HBM stacks according to Hynix.

[stconquest]

 

Here's the driver for Fury X and the 300 series. It's 15.15 so it might have some other game tweaks in it for the 200 series.

 

It's a little more complicated than you would expect.

 

 

insert <itshappeningronpaul.gif> 

 

[Aniallation]

With Fury X2 they'd still be exceeding PCIe spec as they did with the 295X2... two 8-pin (2x150W, 300W) plus what can be provided by the slot (75W) adds up to less then what the theoretical TDP would be, at or around 500W. 

[Opcode]

With Fury X2 they'd still be exceeding PCIe spec as they did with the 295X2... two 8-pin (2x150W, 300W) plus what can be provided by the slot (75W) adds up to less then what the theoretical TDP would be, at or around 500W. 

It's not like another company hasn't done the same. The TITAN Z has a ~500w (2x 250w) power requirement even though Nvidia is lying up and down with 375w on their product page. Fiji sucks 25w more power than the R9 290X and the Fury X2 will suck 50w more than the R9 295x2. Although the performance gain is so drastic at the cost of a few extra watts which is why Fiji is still extremely good in performance per watt. I haven't heard of anyone blowing up their PSU from running either a TITAN Z or the R9 295x2.

[shawno]

All r3xx series have just come up for sale on pccasegear.com

Prices are not too bad. Pretty much the same as the r2xx counterparts.

[Opcode]

All r3xx series have just come up for sale on pccasegear.com

Prices are not too bad. Pretty much the same as the r2xx counterparts.

The XFX R9 390X 8GB is $649.99... that's more than what the R9 290X launched for... It's actually $220 more than AMD's own SEP.

[shawno]

The XFX R9 390X 8GB is $649.99... that's more than what the R9 290X launched for... It's actually $220 more than AMD's own SEP.

Its in aud though. Add the 200 dollar i live in australia tax and its spot on

[Opcode]

Its in aud though. Add the 200 dollar i live in australia tax and its spot on

Meh, I rather Newegg

 

R9 390X = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r9+390x&N=-1&isNodeId=1

R9 390 = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r9+390&N=-1&isNodeId=1

R9 380 = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r9+380&N=-1&isNodeId=1

R7 370 = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r7+370&N=-1&isNodeId=1

R7 360 = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r7+360&N=-1&isNodeId=1

 

Just no pricing yet.

[shawno]

Meh, I rather Newegg

R9 390X = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r9+390x&N=-1&isNodeId=1

R9 380 = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r9+380&N=-1&isNodeId=1

R7 370 = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r7+370&N=-1&isNodeId=1

R7 360 = http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=r7+360&N=-1&isNodeId=1

Just no pricing yet.

Well based on pccg prices i would say usd would be 200 off. Considering the rebrand they could have just got the old stock and drawn the 2 into a 3 and been done with it.

[Sakkura]

Well because it's stacked it has to cool the bottom stack by transferring that heat to the second stack, then transfer the heat of thoes two stacks to the next stack, and so on, so I feel like it's going to have heat problems if overclocked significantly, it's like trying to cool a DDR5 sandwich by putting a cooler on only one end.

 

Except GDDR5 runs at clocks in the 1500 MHz range, while HBM runs at more like 500 MHz. That's one of the main reasons HBM is more power-efficient than GDDR5, which makes thermals less of a concern.

[Zerreth]

ATTENTION PEOPLE IN EUROPE

I actually did the check which sites already got the 390 & 390X for sale as a test for the Fury X availability, here are my findings:


Directly Available (++)
 

Mindfactory.de
Scan.co.uk (but is more expensive than all other retailers in europe)

 
Up for order, expected in a week (+)

Caseking.de
Hardwareversand.de
Alternate.de

Unavailable for Order or Unlisted (-)

Overclockers.co.uk
All Retailers from France, Belgium, Netherlands (Coolblue, Bol, Alternate, Bytes@Work, Pixmania, ...)
This includes ALL Amazon sites, UK, DE, FR, IT, ES

 

I'm also happy to say that the prices are lower than the leaked MSRPs on WCCFTech:

WCCF MSRP 390X: 500-550 € VS Actual Pricing: 450-460 €
WCCF MSRP 390: 420-460 € VS Actual Pricing: 350-370 €

Compare this to actual US pricing and it's basically 1$ == 1 €, so Fury X will likely retail at 650 €

[Sakkura]

ATTENTION PEOPLE IN EUROPE

 

A subset of people in Europe, you mean.

 

For Scandinavia, Komplett seems to have a full set of 300 series cards in stock.

[Zerreth]

A subset of people in Europe, you mean.

 

For Scandinavia, Komplett seems to have a full set of 300 series cards in stock.

Sure, but usually when I check in Scandinavia, prices are way higher, I lived in Finland for half a year and PC hardware prices were ridiculous. It was way cheaper to order via amazon.

Also, why a subset perse ? As far as I know, all retailers I listed ship to all EU countries ? I just checked the Major ones I know of.

[Sakkura]

Sure, but usually when I check in Scandinavia, prices are way higher, I lived in Finland for half a year and PC hardware prices were ridiculous. It was way cheaper to order via amazon.

Also, why a subset perse ? As far as I know, all retailers I listed ship to all EU countries ? I just checked the Major ones I know of.

 

Hardware prices in Denmark are generally a little lower than in the UK, for example. Compared with Germany it varies, but often slightly cheaper in Germany. But then consumer protection is stronger here. I don't know about Finland, it's outside Scandinavia so it might be different there.

[Zerreth]

Hardware prices in Denmark are generally a little lower than in the UK, for example.

Could you please provide me with Retailers in Denmark ? I'm always interested in having more sources to buy from.