LAB501

AMD Radeon R9 295X2 – Nuclear launch detected

AMD Radeon R9 295X2

…the 4K display was showing the same message for 20 min now… activate the Hydra Weapon System. The red line of characters was repeating itself over and over again while I was assembling the weapon. It was just a few minutes after I have entered the correct code to unlock the metal case I have received from the agency when the high-tech 4K monitor started to go wild. First, a series of artifacts of different colors made me think that I was dealing with a broken monitor or with some sort of electromagnetic interference.

Shortly after the artifacts started to show, the screen turned black and I could see the first message of that day – Nuclear launch detected – AMD launch imminent, followed by the obsessively repeating phrase “activate the Hydra Weapon System”. The timing was too precise to be just a coincidence… the 4K monitor didn’t show any artifacts until now so I had all reasons to believe that this odd behavior was directly related to the metal case I have received that day…

After all, the weapon had to be connected to the monitor in order to be activated… and this was the reason I have gotten that monitor years ago. Well…honestly speaking, after the green side made their announcement I was hoping I could prepare both weapons simultaneously so I can test them side by side. However, I cannot say I was too shocked when I learned that the “Z” was not ready. It was not the first time a thing like that happened. Honestly, I wasn’t even expecting to see a green weapon too soon, which was an extra reason for me to be a bit surprised when I received the metal crate with the AMD logo from the red team.

While I was thinking about this odd situation, I was mounting the rig with mechanical gestures. I did that so many times in the past that the cables almost went in their connectors on their own and I didn’t even finish thinking about all this and the weapon was already set-up. Cute beast, I said to myself before pressing the button…

A dim red light started to embrace the room, followed by the sound of liquid circulating through the cooling pipes. Hmm… the beast was much more silent then I expected… I said to myself just before the red light transformed into a giant dome that was covering our lab. Six days later the weapon system was operating at full specs and the red dome was covering the entire northern hemisphere.

Intro

Specs

 

 R9 290XHD 7990R9 295 X2GTX 780 Ti
CoreHawaii

MaltaVesuviusGK110
GPU clock1000 MHz1000 MHz1018 MHz876 MHz
GDDR clock1250 MHz1500 MHz1250 MHz1750 MHz
Boost clock1000 MHz1000 MHz1018 MHz928 MHz
Memory4096 MB2 x 3072 MB2 x 4096 MB3072 MB
TypeGDDR5GDDR5GDDR 5GDDR5
Bus512 bit2 x 384 bit2 x 512 bit384 bit
Shader processors 28162 x 20482 x 28162880
ROPs642 x 322 x 6448
Process28 nm28 nm28 nm28 nm
Power6+88 + 88 + 86 + 8
TDP290 W375 W500 W250 W
Die size438 mm22 x 365 mm22 x 438 mm2550 mm2
Transistor count6,2 billion 8,6 billion12,4 billion7.1 billion

Just a couple of hours before I have received the metal case I have also received a highly classified communication in which the parameters for the new weapon were disclosed. The name was AMD Radeon R9 295X2 and besides using a single PCB and a different cooling solution, the specs detailed in the technical paper seemed very familiar. After all, on the same PCB I could now find two cores which I knew very well from AMD R9 290X. And even if the new weapon was called Vesuvius instead of Hawaii, the specs were the same, except for the clocks.

And even the clocks weren’t that different. The memory was running at the same speed, while the cores on AMD Radeon R9 295X2 were running at 1018MHz (compared to 1000MHz on AMD R9 290X). I know, only a 2% speed bump, but we have to take into consideration that we now have 2 cores on the same PCB, so that was not an easy task. Under this circumstances, the 500W declared TDP was coming as no surprise, especially when we were talking about a weapon with 12,4 billion transistors.

GPUZ295

 

Packaging

 

Unfortunately, the metal briefcase was destined only for secret agents who were testing the first prototypes of the weapon before the actual launch. Once it was going to become available for regular soldiers, the weapon was to be delivered in normal casing, according to each manufacturer and not in a metal briefcase. On the other hand, those packages were most likely going to contain more things that the metal case I have received. Besides the weapon, all I could find inside was the motto of the secret agency – “Melius est ergo duos esse simul quam”… Two are better than one (Eclesiast, 4.9)

Ambalaj-01

Ambalaj-02

Ambalaj-03

Ambalaj-04

Ambalaj-05

The weapon

The first time I took the weapon out if it’s crate I was a bit surprised – this was not what I expected from the red team. If the old guns were painted in black and red, AMD Radeon R9 295X2 was wearing a different combination of war paints – grey and black, with a touch of red on the fan.

Two 8 pin connectors were required in order to feed power to the beast, while the monitor could be connected through one DVI connector or 4 mini DisplayPort 1.2 connectors. The really interesting part from the cooling system was those two black hoses coming out of the regular casing.

Constructie-01

Constructie-02

Constructie-03

Constructie-04

Constructie-05

Cooling system

 

“You will witness something for the very first time…” a fellow from the agency told me. “Guys, I have seen such things before” I wanted to reply. Anticipating my answer, the agent continued “…the first time when somebody is using such a cooling system on a reference card”. Well… he had a point – I have seen water cooled custom designs since 2009, but nobody dared to use such a solution for a reference design. Well, when you have to cool of 500W, there is no surprise they had to bring in the heavy guns.

AMD’s solution was elegant and efficient, and I suppose this had something to do with the fact that they worked with some experts in the field of watercooling, namely Asetek. The two cores were cooled down using two Asetek waterblocks. Each block had its own pump and both were connected in series in order to circulate the cooling liquid through the 120mm radiator. There, the liquid was cooled down with the help of a 120mm fan, before it was sent back to the waterblocks to repeat the thermal cycle.

Of course, the cores were not the only components that were cooled properly. On the back of the card we could find a backplate which was taking the heat of the GDDR 5 IC’s, while on the front of the card we could see two black radiators that were cooling the components surrounding the cores. The VRM was cooled with a separated copper radiator and with the help of the red fan positioned in the middle. The whole set-up looked pretty interesting and soon I was going to find out what it was capable of.

 

Racire-01

Racire-02

Racire-03

Racire-04

Racire-05

 

Under the hood

 

Taking into consideration the power of the monster cores that were soldered to the PCB, the built quality could only be top notch. The first thing that got my attention were the good quality parts, from the Hynix T2C IC’s to the tantalum caps, Copper Bussmann coils and metal fets.

Each core had a 5 phase VRM, comprised of Copper Bussmann coils, ceramic and poscap caps and metal fets, controlled by the 81022 PWM. The VRM for RAM, PLL and the PLX IC were comprised of one phase, with a Copper Bussman coil, polimeric and poscap caps and two metal fets. It was obvious that AMD had made no compromise in the development of this beast.

 

PCB=01

PCB=02

PCB=03

PCB=04

PCB=05

PCB=06

 

Testbed

 

  
CPUIntel Core i7 4770K
MotherboardGIGABYTE GA-Z97X-OC Force
RAMGEIL Evo Corsa 8GB DDR3-2133 CL10
CoolingNoctua NH-D14 + 2 x Coolink SwiF 120P
HDDKingston HyperX 240GB
PSUCooler Master UCP 1100W
CaseHSPC Top Deck Station
Room temp25 oC
OSWindows 7 Ultimate x64 SP1
AMD DriverCatalyst 14.4 Press BETA
Nvidia DriverForceware 335.23 WHQL
CPU clock4500 MHz
DDR clockDDR3 2133
Timings9-11-9-28 1T

 

In order to get the best performance from such a beast, we used a PCI-E 3.o platform, based on a Haswell CPU. The Intel Core i7 4770K CPU was clocked at 4500MHz, being cool and cozy with 1.16v and a Noctua NH-D14 on top. We used a GIGABYTE Z87X-OC Force motherboard and a dual-channel GEIL Evo Corsa 8GB DDR3-2133 CL10 memory kit set at DDR3 2133 9-11-9-28. The PSU was our old Cooler Master UCP 1100W while the “case” was a HSPC Top Deck Station bench table.

The OS, Windows 7 Ultimate x64 SP1. was installed on a Kingston HyperX 240GB SSD. Of course, we used the latest AMD driver at the time of the test (Catalyst 14.4 Press BETA), and the latest Nvidia driver available at the time of the testing (Forceware 335.23 WHQL). We are using only original benchmark and games, regularly updated.

We tested all graphic cards with our usual test resolutions (1920×1200 si 2560×1600), with two quality settings for each (8xAA and 2xAA), replacing them with 4xAA and NoAA when the game only offered this option. All details were set to maximum, as long as this meant that we were using identical settings for AMD and Nvidia cards. Because we are testing the very best of the best, we also added a third resolution to this test, the all mighty 4K (3840×2160).

OLYMPUS DIGITAL CAMERA

 

3DMark

 

As you can imagine, we couldn’t have launched the weapon without a series of pre-launch tests and comparisons with other solutions from our arsenal. Unfortunately, as much as we would have loved to compare AMD Radeon R9 295X2 with it’s direct rival from the green company, the Z wasn’t delivered to us even until now, so we will not take it into consideration. Our lab has been supplied so far with all existing weapons in the last years, so if something does not get to us… well, it simply does not exist.

We first gave AMD Radeon R9 295X2 the task of destroying frames in a synthetic environment, namely 3DMark. In all the situations I have tested, the difference between HD 7990, GTX 780 Ti and AMD Radeon R9 295X2 was substantial, as expected.

3DM1

3DM2

3DM3

 

Far Cry 2

 

FC2-3

FC2-4

The first real-world test shows us why 4K is mandatory in such a test – the difference between AMD Radeon R9 295X2 and it’s rivals is less than 3% in the classical resolutions.

FC2-5

FC2-6

 

Only in 3840×2160 we can start to see what AMD Radeon R9 295X2 is made of, and the gap between our weapon and the old models starts to widen. When we use 8xAA, the results are 50% better then the old generation.

 

FC2-1

FC2-2

 

Crysis 2

 

C3

C4

 

In Crysis 2 we only see differences between HD 7990 and GTX 780Ti and so far AMD Radeon R9 295X2 can only show it’s power in huge resolutions, like 4K.

 

C5

C6

 

As I was saying, here is where you can see the full power of the beast that is called AMD Radeon R9 295X2. Also, GTX 780Ti makes a nice move in this situation, with 4xAA.

 

C1

C2

 

Hitman Absolution

 

H3

H4

Hitman comes to confirm what we already know – the higher the resolution and the higher the AA setting, the bigger the distance between the red monster and its competitors gets.

H5

H6

 

And it is obvious in 3840×2160, where AMD Radeon R9 295X2 and GTX 780Ti get a measly 1fps minimum frame rate with 8xAA, while 7990 can’t even finish the test.

 

H1

H2

 

Alien vs Predator

 

A3

A4

 

AvP is the first test that shows a bit of a challenge for our competitors, and AMD Radeon R9 295X2 can demonstrate it’s superiority because of this.

 

A5

A6

 

As we go into high-resolution territory, the difference between AMD Radeon R9 295X2 and GTX 780Ti & HD 7990 becomes more obvious, the first getting almost double performance numbers.

 

A1

A2

 

DIRT Showdown

 

D3

D4

 

Need more pixels..nothing to see here…

 

D5

D6
 

 

D1

D2

 

Unigine Heaven 4.0

 

U3

U4

Just as we saw in AvP, as long as the engine is complex enough, the difference between AMD Radeon R9 295X2 and it’s competitors is obvious even in lower resolutions.

U5

U6

 

U1

U2

 

Sleeping Dogs

 

S1

S2

Just like Battlefield 4, Sleeping Dogs refused to be stable in 3840×2160, AMD Radeon R9 295X2 being the only card that could successfully finish the test.

S3

S4

FarCry 3

 

FC3-3

FC3-4

 

FarCry 3 does bring a bit of a challenge, there for the results are indeed interesting.

 

FC3-5

FC3-6

 

So interesting that in fact HD 7990 can’t properly run in 3840×2160, while AMD Radeon R9 295X2 has triple performance compared to GTX 780 Ti. Even so, if you really want to play this game at this resolution, my advice would be not to use AA.

FC3-1

FC3-2

 

Battlefield 4

 

B1

B2

 

Read them and weap boys… quality gameplay in 2560×1600 and 8X AA in Battlefield 4. Not bad!

 

B3

B4

 

Metro Last Light

 

Remmeber? The higher the… yeah… you’ve noticed this already.

 

M1

M2

M3

 

Power consumption

 

CI

CL

 

In order to measure the power consumption, we used a dedicated tool, leaving the system in idle for 10 minutes. Then we ran Crysis 2 in 1920×1200 4xAA. The numbers you see above represent the power consumption of the whole system, not just the VGA, and we are talking about the highest value recorded during our test.

That means that the average power consumption is considerably lower, but we are looking for the peak value, the worst case scenario, the maximum possible stress for our PSU. Obviously the total power consumption of the system can be even higher if you stress even more components (storage, RAM, etc), but we are looking for a maximum value relevant to daily use scenarios, aka gaming.

When I first saw the TDP declared by AMD (500w) I though I might melt some wires in my power grid. I say that because in the last years manufacturers were optimistic when it came to declare the power consumption of their graphic cards. This time, things are the other way around. In 1920×1200, as we usually test, the system only drew 538W from the wall. On the other hand, when I tested in 4K, the meter was showing a power consumption of 638W. Bottom line – if you want this card you’d better get a reasonably sized good quality PSU.

 

Temperature

 

TI

TL

 

The temperature levels we reach are not representative for the temperature levels you could get in a normal system, in an enclosure. If you have a poorly ventilated enclosure you could get much higher temps, while if you have a very well ventilated enclosure you could get better temp numbers. The purpose of this chapter is to compare one graphic card to another, not to get absolute numbers.

We used EVGA Precision to measure the temperature levels of the GPU and the fan speed, both in idle and in full load, in the same situation we used in the power consumption test. Knowing that we have 2 monstrous GPUs on the same PCB and taking our hot experience with R9 290X into consideration, I was expecting to see really impressive numbers here. And I did, but not in the way I expected. The cooling system AMD developed together with Asetek is really something, and the cores didn’t get past 61 oC in our testing conditions.

 

Noise levels

 

ZI

ZL

 

We measure the noise level using a professional Voltcraft meter, namely SL200. The meter is fixed on a tripod at 85cm from our test platform, above the CPU cooler. For this test, the CPU runs fanless at 3GHz. The only fan besides the one from the graphic card is the one from the PSU.

The tests are done in a soundproof room with a noise level bellow 32dB. Remember – the numbers you see here are relevant to our test platform and you will get different results at home, in a case, etc. Our reference system is built so we can compare various graphic cards in the same conditions. The sound level is measured during 4 runs of Crysis 2 in 1920×1200 with 8xAA. The numbers you see in the graphs are the maximum levels measured during these tests.

Well, with a high power consumption and low temperatures we would have expected this test to have a noisy outcome, but AMD and Asetek manage to impress once again and the noise levels are pretty low for a high-end graphic card. And definitely lower than those of the AMD R9 290X…

Final thoughts

 

Well, the nuclear threat has been neutralized, the weapon is armed and ready, and today we had the chance to see how it is built and how it performs compared to the ex top of the line from AMD (HD 7990) and the actual Nvidia top of the line (GTX 780 Ti). And unfortunately we couldn’t test against a modern Nvidia dual GPU because… we didn’t have one.

At the moment things seem to be crystal clear – GTX 780Ti is the most powerful single GPU solution out there while the title of “most powerful graphic card ever built” goes to AMD Radeon R9 295X2. The red behemoth is literally a beast and it is the most powerful thing I have personally tested. And the price is on par with the performance (MSRP 1499 USD).

Leaving that aside, I think the most important thing we must understand today is that in 1920×1200 (or 1920×1080) the difference between AMD Radeon R9 295X2 is not that high. The difference increased with each step we increment the resolution, the true power of this card being truly obvious only in 4K (3840×2160). I can’t say how the 4K technology will evolve this year, but I do know that an entire industry is focused on bringing new and more affordable solutions (24″ Dell at 850 USD, 28″ Samsung at 700 USD and so on). I am going to start paying attention when I will see a 32″ Benq or Acer under 1000 USD…

Honestly, 4K under 32″ makes no sense, especially if you care about your eyesight. Think that on a 24″ screen you will have images exactly twice as small compared to 1080p, which is not at all comfortable for the users. Because of that, AMD Radeon R9 295X2 is a solution for those who can afford not only the graphic card but also a good quality 4K monitor. If you really want to enjoy this beast you will need something bigger than full HD, and this is a first step towards the dawn of 1080p gaming on PC’s…

Outro

 

6 comments la: AMD Radeon R9 295X2 – Nuclear launch detected

    1. GLC wrote on:

      A little typo in the second pharagraph: “The timing was to precise to be just a coincidence…”. The correct form of the first “to” is “too”, not “to”.

      Reply
      1. Done, thanks!

    2. GLC wrote on:

      Also you wrote “before” as “be fore” on the third page.

      Reply
    3. Thanks for the proofing!

      Reply
    4. GLC wrote on:

      I’m glad I could help you!

      Reply
    5. Monstru Post author wrote on:

      Thank you for the help guys!

      Reply

    Leave a Reply