Nvidia finally loses.

[Werne]

Apparently the 295x2 is faster at 4k resolution.

25% more ROPs on a 1GHz GPU strapped to a 33% wider memory bus, of course it's better at 4K. But because of the high-resolution design and lack of TMUs as well as lower OC clocks (due to power consumption and heat), it lacks on lower resolutions where it gets absolutely hammered by GK110 which can pull off higher clock speeds on an architecture optimized for more mainstream 1080/1440p resolutions.

 

Hawaii is designed for high resolutions like 4K but compromises lower resolution performance to reach that goal.

[Thor.]

Good old amd fighting a good fight, we need a cheap ti equivalent gpu, not dual just single.

Edited June 4, 2014 by Thor.

[dave1029]

 

Apparently the 295x2 is faster at 4k resolution.

25% more ROPs on a 1GHz GPU strapped to a 33% wider memory bus, of course it's better at 4K. But because of the high-resolution design and lack of TMUs as well as lower OC clocks (due to power consumption and heat), it lacks on lower resolutions where it gets absolutely hammered by GK110 which can pull off higher clock speeds on an architecture optimized for more mainstream 1080/1440p resolutions.

 

Hawaii is designed for high resolutions like 4K but compromises lower resolution performance to reach that goal.

 

Your definition of "hammered" is amusing. It's on average 5-10 fps per faster than the 295x2 at 1080 and 5 fps faster at 1440. Remember, two 780 Ti's/Titan Blacks cost more than a single 295x2- not to mention the 295x2 won't throttle.

[Werne]

Your definition of "hammered" is amusing. It's on average 5-10 fps per faster than the 295x2 at 1080 and 5 fps faster at 1440. Remember, two 780 Ti's/Titan Blacks cost more than a single 295x2- not to mention the 295x2 won't throttle.

I guess you didn't read the articles well enough. While R9 290X throttles at 95C, R9 295X2 is set to throttle at 75C, which some reviewers reached with the AIO supplied on it. Plus, VRMs hit close to 105C after some time, that would also throttle it down.

 

And what I meant by "hammered" is not on stock clocks, it's after the cards are overclocked. See, GK110 is a monster in that regard and GK110 can clock up to 10-15% higher than Hawaii, especially on water, and it's even enough to tie with it on 4K. That's what's killing Hawaii chips the most. Also, don't compare enthusiast equipment on stock clocks.

 

Finally, 780 Ti SLI costs less than a single R9 295X2, so does R9 290X CF. Current dual-chip cards are incredibly overpriced, used to be that they are cheaper than two cards from which the chips are derived, but now it costs 50% more for no benefit at all.

[dave1029]

 

Your definition of "hammered" is amusing. It's on average 5-10 fps per faster than the 295x2 at 1080 and 5 fps faster at 1440. Remember, two 780 Ti's/Titan Blacks cost more than a single 295x2- not to mention the 295x2 won't throttle.

I guess you didn't read the articles well enough. While R9 290X throttles at 95C, R9 295X2 is set to throttle at 75C, which some reviewers reached with the AIO supplied on it. Plus, VRMs hit close to 105C after some time, that would also throttle it down.

 

And what I meant by "hammered" is not on stock clocks, it's after the cards are overclocked. See, GK110 is a monster in that regard and GK110 can clock up to 10-15% higher than Hawaii, especially on water, and it's even enough to tie with it on 4K. That's what's killing Hawaii chips the most. Also, don't compare enthusiast equipment on stock clocks.

 

Finally, 780 Ti SLI costs less than a single R9 295X2, so does R9 290X CF. Current dual-chip cards are incredibly overpriced, used to be that they are cheaper than two cards from which the chips are derived, but now it costs 50% more for no benefit at all.

 

780 Ti SLI is like $50 cheaper than a single 295x2. And no, the 295x2 does NOT throttle. It took an hour for furmark to throttle the card with only its reference cooler. Combine a properly cooled case with big fans and it could take hours to throttle. Any SLI or xfire combo will throttle quicker because the pcb's are right next to eachother making cooling tougher. A single card can be overclocked more than a dual gpu, two GK 110's can NOT be overclocked as much as a single card because it WILL throttle.

[Werne]

780 Ti SLI is like $50 cheaper than a single 295x2.

It's actually $100-200 less expensive and overclocks up to 15% higher than two separate Hawaii cards, two GPUs on a single PCB will overclock even lower due to VRM heat emission limitations and the VRMs are bound to heat up like crazy due to high power draw (VRMs have 80-85% efficiency meaning the VRM heatsink would have to dissipate 75-100W of power). Those massive motherboard VRM heatsinks on AMD AM3+ boards like Sabertooth can barely sustain a mere 50-70W of dissipation power before hitting their thermal limit (that's with active cooling), a graphics card stands little chance despite having more power phases.

 

Also, two MSI Twin Frozr R9 290X cards cost $1120 and they are no slower (actually faster, they are higher clocked out-of-the-box). R9 295X2 is a useless product, designed for no one other than die-hard fanboys at $1500, and that Titan Z thing is much, much worse at $3000 cause it's designed for... no one, actually. They are both useless cards and they serve a niche market.

 

And no, the 295x2 does NOT throttle. It took an hour for furmark to throttle the card with only its reference cooler. Combine a properly cooled case with big fans and it could take hours to throttle. Any SLI or xfire combo will throttle quicker because the pcb's are right next to eachother making cooling tougher. A single card can be overclocked more than a dual gpu, two GK 110's can NOT be overclocked as much as a single card because it WILL throttle.

Blower cooler cards can be placed directly on top of each other without the top card sucking on the bottom card's hot air, and nVidia actually coupled their series with very capable blowers. AMD, well, I feel sorry for the guys who bought the reference 290/290X models.

 

By the way, I ran Crossfire with axial fan cards, if you have decent airflow the variation between cards will be under 8C. However, you apparently give airflow to dual-GPU cards but deprive two separate cards of air (that's what I understand from your post). Not to mention most respectable CF/SLI boards actually have spacing between the cards which means they don't lay on top of each other but have a whole PCI slot worth of room between them. Even my cheapo GA-970A-UD3 has lots of spacing for CF. Seems to me like you're just making up weird scenarios to prove a point, don't do that.

 

And no, a single card will overclock worse despite having higher-binned chips due to aforementioned problem with having two sets of VRMs crammed next to each other on a standard-sized PCB, R9 295X2 is the same size as an R9 290X but with twice the power draw and twice the power delivery heat that needs to be dissipated. VRMs aren't water cooled on R9 295X2, they hit nearly 100C in gaming scenarios and 103C in Furmark test while throttling at 105C and dropping boost clocks down lower as soon as hitting 95C. Even an 80MHz overclock on the GPUs would make the card hit the ceiling and drop to base. Without a full water block (I think EK has one), you're getting stock speeds, which isn't the case with separate cards.

 

A dual-GPU card never, and I mean never, overclocks as good as two separate cards. As was the case with every dual-GPU card ever produced - 590 OC'd worse than two 580s, 690 OC'd worse than two 680s, 6990 OC'd worse than two 6970s, 7990 OC'd worse than two 7970s, R9 295X2 OCs worse than two R9 290X and Titan Z OCs worse than two Titan Blacks. It always will be this way until these cards come with a phase change unit or an LN2/DICE pot as reference cooling.

[dave1029]

 

780 Ti SLI is like $50 cheaper than a single 295x2.

It's actually $100-200 less expensive and overclocks up to 15% higher than two separate Hawaii cards, two GPUs on a single PCB will overclock even lower due to VRM heat emission limitations and the VRMs are bound to heat up like crazy due to high power draw (VRMs have 80-85% efficiency meaning the VRM heatsink would have to dissipate 75-100W of power). Those massive motherboard VRM heatsinks on AMD AM3+ boards like Sabertooth can barely sustain a mere 50-70W of dissipation power before hitting their thermal limit (that's with active cooling), a graphics card stands little chance despite having more power phases.

 

Also, two MSI Twin Frozr R9 290X cards cost $1120 and they are no slower (actually faster, they are higher clocked out-of-the-box). R9 295X2 is a useless product, designed for no one other than die-hard fanboys at $1500, and that Titan Z thing is much, much worse at $3000 cause it's designed for... no one, actually. They are both useless cards and they serve a niche market.

 

And no, the 295x2 does NOT throttle. It took an hour for furmark to throttle the card with only its reference cooler. Combine a properly cooled case with big fans and it could take hours to throttle. Any SLI or xfire combo will throttle quicker because the pcb's are right next to eachother making cooling tougher. A single card can be overclocked more than a dual gpu, two GK 110's can NOT be overclocked as much as a single card because it WILL throttle.

Blower cooler cards can be placed directly on top of each other without the top card sucking on the bottom card's hot air, and nVidia actually coupled their series with very capable blowers. AMD, well, I feel sorry for the guys who bought the reference 290/290X models.

 

By the way, I ran Crossfire with axial fan cards, if you have decent airflow the variation between cards will be under 8C. However, you apparently give airflow to dual-GPU cards but deprive two separate cards of air (that's what I understand from your post). Not to mention most respectable CF/SLI boards actually have spacing between the cards which means they don't lay on top of each other but have a whole PCI slot worth of room between them. Even my cheapo GA-970A-UD3 has lots of spacing for CF. Seems to me like you're just making up weird scenarios to prove a point, don't do that.

 

And no, a single card will overclock worse despite having higher-binned chips due to aforementioned problem with having two sets of VRMs crammed next to each other on a standard-sized PCB, R9 295X2 is the same size as an R9 290X but with twice the power draw and twice the power delivery heat that needs to be dissipated. VRMs aren't water cooled on R9 295X2, they hit nearly 100C in gaming scenarios and 103C in Furmark test while throttling at 105C and dropping boost clocks down lower as soon as hitting 95C. Even an 80MHz overclock on the GPUs would make the card hit the ceiling and drop to base. Without a full water block (I think EK has one), you're getting stock speeds, which isn't the case with separate cards.

 

A dual-GPU card never, and I mean never, overclocks as good as two separate cards. As was the case with every dual-GPU card ever produced - 590 OC'd worse than two 580s, 690 OC'd worse than two 680s, 6990 OC'd worse than two 6970s, 7990 OC'd worse than two 7970s, R9 295X2 OCs worse than two R9 290X and Titan Z OCs worse than two Titan Blacks. It always will be this way until these cards come with a phase change unit or an LN2/DICE pot as reference cooling.

 

You're also leaving out one thing. Sure it might be a "tad faster" at lower resolutions, but it has more vram. 8GB as opposed to 3 GB. And I just got my 295x2 today... Let's just say it won't overheat. In fact, in runs at 34 C idle. Doesn't get into the 70's while gaming. Ever. I have good fan cooling though along with it being liquid cooled. Also, the actual GPU's are exposed on the card, so the heat can easily escape.

[FMod]

I wonder just how many people buy $3,000 dual-GPU cards - and a common reason to buy dual cards, BTW, is to put 2 of them together, it's much more usable than 4 separate cards - to play in 1024x768 or whatever's en vogue today.

[Werne]

You're also leaving out one thing. Sure it might be a "tad faster" at lower resolutions, but it has more vram. 8GB as opposed to 3 GB.

It's 4GB VRAM, 4GB pool per GPU which is mirrored, same as two 290X in Crossfire since R9 295X2 is permanent Crossfire in a small package (same CF protocols and rules apply). If you have 1GB VRAM used on one GPU, all that data is mirrored into VRAM of the other GPU as well, occupying 2GB out of 8 total. If you have 3GB used out of 4 on a GPU, it occupies 6GB total. You seem to lack fundamental knowledge about how Crossfire and dual-GPUs work - yes, there is physically 8GB worth of GDDR5 on the PCB and no, there's only 4GB usable due to mirroring.

 

To clarify - VRAM in CF operates differently than dual-channel system memory. System RAM increases bandwidth and doubles memory pool since it's designed to strip data and store it across two modules, which reduces access to each module boosting bandwidth and reducing latency. VRAM, on the other hand is mirrored since each GPU needs to be fed with the same data and they can't operate out of the same pool (data rewriting would cause conflicts). And even if they were able to, sharing the pool would effectively halve the bandwidth and increase latency since it would need an insanely fast interconnect bus, meaning that the card would be about the same speed or slower than a single-GPU card due to twice the I/O access to the pool.

 

And FYI, even 4GB of VRAM is unnecessary unless running a console-ported mess like Watch Dogs (which stutters on a dual-GPU cards and CF/SLI anyway, even Titan Black which has 6GB of the stuff) on 4K. VRAM is a marketing gimmick, people actually think having tons of VRAM makes the card go faster. It doesn't, though it does limit memory OC potential due to having more GDDR5 modules which bring potential of one being crap and not allowing the others to reach higher clocks. It's why cards with less VRAM modules generally reach higher memory clocks, a good example being a 1GB 7850 vs 2GB 7850 where 1GB card often reaches up to 10% higher VRAM clocks despite having the same memory module manufacturer and models. Every module has different yields and the worst one determines the highest overclock, that's why people buy dual/triple/quad-channel memory kits instead of separate modules, they are more likely to have similar yields but not necessarily.

 

And I just got my 295x2 today... Let's just say it won't overheat. In fact, in runs at 34 C idle. Doesn't get into the 70's while gaming. Ever. I have good fan cooling though along with it being liquid cooled. Also, the actual GPU's are exposed on the card, so the heat can easily escape.

Seeing as how you lack knowledge about what the card actually is, the most likely scenario is that you are lying. Either that or you got it without knowing what it is, in which case I feel sorry for your wallet. And you mention idle temperatures which were never relevant in any way, and they never will be. My card idles at 28C on GPU and that doesn't mean anything at all, it's the load temperature that matters (62C on GPU, 64C on VRMs).

 

And I never said the GPUs will overheat, I said the VRMs will overheat once the card is overclocked and begin dropping down boost clocks lower at 95C even on stock (and it can easily reach 95C VRMs on stock). It's damn near unavoidable, unless you either have a full water block on it, play games for less than 1h at a time, or run on a 60Hz 1080p screen (but then you're completely wasting money). How well you cool the GPUs will not affect the VRM temperatures since those are air-cooled with the central axial fan on the shroud, completely separate.

 

By the way, I have no idea what you're trying to say with "GPUs are exposed" and "heat can escape". Doesn't make much sense to me without any further details.

 

I wonder just how many people buy $3,000 dual-GPU cards - and a common reason to buy dual cards, BTW, is to put 2 of them together, it's much more usable than 4 separate cards - to play in 1024x768 or whatever's en vogue today.

A guy I know has a 7990 hooked up to a single 720p monitor, so I can honestly see people buying 295X2 and Z for sub-2M pixel resolutions. I sometimes feel bad for laughing at those people, but 99.9% of the time it amuses me.

[dave1029]

 

You're also leaving out one thing. Sure it might be a "tad faster" at lower resolutions, but it has more vram. 8GB as opposed to 3 GB.

It's 4GB VRAM, 4GB pool per GPU which is mirrored, same as two 290X in Crossfire since R9 295X2 is permanent Crossfire in a small package (same CF protocols and rules apply). If you have 1GB VRAM used on one GPU, all that data is mirrored into VRAM of the other GPU as well, occupying 2GB out of 8 total. If you have 3GB used out of 4 on a GPU, it occupies 6GB total. You seem to lack fundamental knowledge about how Crossfire and dual-GPUs work - yes, there is physically 8GB worth of GDDR5 on the PCB and no, there's only 4GB usable due to mirroring.

 

To clarify - VRAM in CF operates differently than dual-channel system memory. System RAM increases bandwidth and doubles memory pool since it's designed to strip data and store it across two modules, which reduces access to each module boosting bandwidth and reducing latency. VRAM, on the other hand is mirrored since each GPU needs to be fed with the same data and they can't operate out of the same pool (data rewriting would cause conflicts). And even if they were able to, sharing the pool would effectively halve the bandwidth and increase latency since it would need an insanely fast interconnect bus, meaning that the card would be about the same speed or slower than a single-GPU card due to twice the I/O access to the pool.

 

And FYI, even 4GB of VRAM is unnecessary unless running a console-ported mess like Watch Dogs (which stutters on a dual-GPU cards and CF/SLI anyway, even Titan Black which has 6GB of the stuff) on 4K. VRAM is a marketing gimmick, people actually think having tons of VRAM makes the card go faster. It doesn't, though it does limit memory OC potential due to having more GDDR5 modules which bring potential of one being crap and not allowing the others to reach higher clocks. It's why cards with less VRAM modules generally reach higher memory clocks, a good example being a 1GB 7850 vs 2GB 7850 where 1GB card often reaches up to 10% higher VRAM clocks despite having the same memory module manufacturer and models. Every module has different yields and the worst one determines the highest overclock, that's why people buy dual/triple/quad-channel memory kits instead of separate modules, they are more likely to have similar yields but not necessarily.

 

And I just got my 295x2 today... Let's just say it won't overheat. In fact, in runs at 34 C idle. Doesn't get into the 70's while gaming. Ever. I have good fan cooling though along with it being liquid cooled. Also, the actual GPU's are exposed on the card, so the heat can easily escape.

Seeing as how you lack knowledge about what the card actually is, the most likely scenario is that you are lying. Either that or you got it without knowing what it is, in which case I feel sorry for your wallet. And you mention idle temperatures which were never relevant in any way, and they never will be. My card idles at 28C on GPU and that doesn't mean anything at all, it's the load temperature that matters (62C on GPU, 64C on VRMs).

 

And I never said the GPUs will overheat, I said the VRMs will overheat once the card is overclocked and begin dropping down boost clocks lower at 95C even on stock (and it can easily reach 95C VRMs on stock). It's damn near unavoidable, unless you either have a full water block on it, play games for less than 1h at a time, or run on a 60Hz 1080p screen (but then you're completely wasting money). How well you cool the GPUs will not affect the VRM temperatures since those are air-cooled with the central axial fan on the shroud, completely separate.

 

By the way, I have no idea what you're trying to say with "GPUs are exposed" and "heat can escape". Doesn't make much sense to me without any further details.

 

I wonder just how many people buy $3,000 dual-GPU cards - and a common reason to buy dual cards, BTW, is to put 2 of them together, it's much more usable than 4 separate cards - to play in 1024x768 or whatever's en vogue today.

A guy I know has a 7990 hooked up to a single 720p monitor, so I can honestly see people buying 295X2 and Z for sub-2M pixel resolutions. I sometimes feel bad for laughing at those people, but 99.9% of the time it amuses me.

 

It's 59 C under load. And fine, Vram doesn't *stack*, but it does. This is because the cards use half as much vram in xfire as opposed to a single card. I've noticed this. AC 4 for instance, with one card it's about 2.5 GB vram. With xfire it's 1.3Gb per card. The card is extremely cool and quiet under heavy load. It is probably because my case has really good cooling, regardless, the vram and gpu's are cooled via liquid cooling. In a benchmark, the card never throttled over a 2h max load period. It's not permanent xfire. I can turn one of the gpu's off. I have to specify it in the application settings as it defaults to xfire always on. So I can play watch dogs. I got it for one reason, 4k gaming. So my wallet is hurting but it is a happy hurt.