Radeon R9 295X2 Press Deck Leaked

[btarunr]

Here are some of the key slides from AMD's press-deck (presentation) for reviewers, for the Radeon R9 295X2 dual-GPU graphics card, ahead of its April 8 launch. The slides confirm specifications that surfaced earlier this week, which describe the card as bearing the codename "Vesuvius," having two 28 nm "Hawaii" GPUs, and all 2,816 stream processors on the chips being enabled, next to 176 TMUs, 64 ROPs, and 512-bit wide GDDR5 memory interfaces. Two such chips are wired to a PLX PEX8747 PCI-Express 3.0 x48 bridge chip. There's a total of 8 GB of memory on board, 4 GB per GPU. Lastly, clock speeds are revealed. The GPUs are clocked as high as 1018 MHz, and memory at 5.00 GHz (GDDR5-effective). The total memory bandwidth of the card is hence 640 GB/s.

The Radeon R9 295X2 indeed looks like the card which was pictured earlier this week, by members of the ChipHell tech community. It features an air+liquid hybrid cooling solution, much like the ROG ARES II by ASUS. The cooling solution is co-developed by AMD and Asetek. It features a couple of pump-blocks cooling the GPUs, which are plumbed with a common coolant channel running through a single 120 mm radiator+reservoir unit. A 120 mm fan is included. A centrally located fan on the card ventilates heatsinks that cool the VRM, memory, and the PCIe bridge chip.



The card draws power from two 8-pin PCIe power connectors, and appears to use a 12-phase VRM to condition power. The VRM appears to consist of CPL-made chokes, and DirectFETs by International Rectifier. Display outputs include four mini-DisplayPort 1.2, and a dual-link DVI (digital only). The total board power of the card is rated at 500W, and so AMD is obviously over-drawing power from each of the two 8-pin power connectors. You may require PSUs with strong +12V rails driving them. Looking at these numbers, we'd recommend at least an 800W PSU for a single-card system, ideally with a single +12V rail design. The card is 30.7 cm long, and its coolant tubes shoot out from its top. AMD expects the R9 295X2 to be at least 60 percent faster than the R9 290X at 3DMark FireStrike (performance).



View at TechPowerUp Main Site

 

[Naito]

Beast of a card! Certainly offers more kit than nVidia's offering

 

[Xzibit]

Just noticed. Its called Project Hydra.

 

[LAN_deRf_HA]

Why oh why would you try to cheaply imitate the styling of your competitor? This card looks like a Chinese knockoff of a high end Nvidia card. It's so bizarre they'd go that direction, look at the final product, and then just not see the glaring problem there. I mean they even have the glowing side logo.... It's like they decided grey metal and rivets must be part of a new DX standard.

 

[hanzawhtet7]

WOOHOO! AMD is giving best shot! Looking forward to it

 

[jateruy]

What a fugly shroud and fan, I would much prefer the original R9 styled cooler instead of this cheapy plastic looking LED fan.

 

[manofthem]

Sounds like these gpu hybrid coolers are becoming increasingly popular, but I wouldn't want to bother with it.

This card deserves a full water block.

 

[Nordic]

They should sell these cards with no cooler for cheap.

 

[qu3becker]

This card reminds me of the XFX recent cards like this one:
http://www.newegg.ca/Product/Product.aspx?Item=N82E16814150590
As for me, the card looks okay. At least it's not glossy plastic like the 7990.

 

[Xzibit]

Looking at the numbers

295x2 - 11.5 TFLOPS
Titan Z = 8 TFLOPS

 

[HumanSmoke]

Looking at the numbers

295x2 - 11.5 TFLOPS
Titan Z = 8 TFLOPS

What a joke.
Point #1 : 295X2 is based on continuous operation at 1018MHz (and the actual number would be 11.467*) - how likely is that scenario
And since you're cherry picking a bullet point in isolation, how about this one:
Point #2 : FP64.....295X2 : 1.433 TFlops (AMD artificially limits double precision to 1/8 of single precision with Radeon Hawaii) -and which is actually ~25% less than the HD 7990) and 54% less than Titan Z if the Z's 8TF FP32 is correct
* (1018 * 5632 * 2)

 

[damric]

I love posting this pic

 

[RCoon]

Overdrawing power from two 8pins is less than stellar. I'm aware most psu's from reputable brands can handle it, but it's outside specs. Would it really have hurt just banging on a third 8 pin just for safety? It's not like the people who buy this won't have one available.

 

[Patriot]

Overdrawing power from two 8pins is less than stellar. I'm aware most psu's from reputable brands can handle it, but it's outside specs. Would it really have hurt just banging on a third 8 pin just for safety? It's not like the people who buy this won't have one available.

PCIE 3 slot can give 300w... why mobo needs multiple 8pins.

 

[seronx]

PCIE 3 slot can give 300w... why mobo needs multiple 8pins.

PCIe 3.0 Electromechanical.
PCIe slot = 75 watts
PCIe 2x4 = 150 watts

75 + 150 + 150 = 375 watts

The PEG cables can be told to go above and beyond the official specification but not the slot.

75W + 300W + 150W or 300W + 300W with no power coming from the slot.

 

[RCoon]

PCIE 3 slot can give 300w... why mobo needs multiple 8pins.

The total board power of the card is rated at 500W, and so AMD is obviously over-drawing power from each of the two 8-pin power connectors

Says in the OP it's going to be overdrawing.

 

[Patriot]

PCIe 3.0 Electromechanical.
PCIe slot = 75 watts
PCIe 2x4 = 150 watts

75 + 150 + 150 = 375 watts

The PEG cables can be told to go above and beyond the official specification but not the slot.

75W + 300W + 150W or 300W + 300W with no power coming from the slot.

So it seems... looks like 300w is still max for pcie spec... dual 8pin config not official lol ... Thanks for the correction.

 

[HumanSmoke]

PCIe 3.0 Electromechanical.
PCIe slot = 75 watts
PCIe 2x4 = 150 watts

75 + 150 + 150 = 375 watts

The PEG cables can be told to go above and beyond the official specification but not the slot.

75W + 300W + 150W or 300W + 300W with no power coming from the slot.

Well, that sounds fine and dandy.......what about AMD's own range of 900 (990FX/990X/970) series chipset boards? They aren't specced for PCI-E 3.0 operation are they? That being the case then an overdraw past 75W via the slot isn't a given.

 

[seronx]

Well, that sounds fine and dandy.......what about AMD's own range of 900 (990FX/990X/970) series chipset boards? They aren't specced for PCI-E 3.0 operation are they? That being the case then an overdraw past 75W via the slot isn't a given.

The PCIe Electromechanical specification is forward and backwards compatible.

 

[HumanSmoke]

Would it really have hurt just banging on a third 8 pin just for safety? It's not like the people who buy this won't have one available.

Other than thumbing its nose at the PCI-SIG, it might have been a tight fit to accommodate a third power plug. According to the slide deck, the card is 305-307mm long. I believe the PCI-SIG limit is 312mm.

 

[RCoon]

Well, that sounds fine and dandy.......what about AMD's own range of 900 (990FX/990X/970) series chipset boards? They aren't specced for PCI-E 3.0 operation are they? That being the case then an overdraw past 75W via the slot isn't a given.

Kaveri/FM2+ does though right?

 

[HumanSmoke]

Kaveri/FM2+ does though right?

Yup.

 

[btarunr]

NextFX CPUs, which could be FM2+ chips with four Steamroller modules, no IGP, but integrated PCIe gen 3 root complex, should overcome that. Current AMD 7-series APUs support PCIe gen 3.

 

[radrok]

Would've loved to see it as a 3x8 pin just in case someone wants to abuse it for overclocking, 2x8 doesn't leave much room for power

 

[adulaamin]

I'm glad AMD came out with this card although if I were ever to buy one I'd ditch that hybrid cooler and build a custom loop. I don't like how it looks like but if it performs well and is priced right then it's a win.