NVIDIA GF100 & Friends speculation

[gongo]

I think the length of the card is also important buying factor....Caymans are really longer than "normal" cards...

 

[Dave Baumann]

Yes, this makes sense. More careful routing is probably needed, and high frequency noise might be more of a problem. It is unclear though to me if the pcb is actually more complex in this case (I was primarily basing this on power draw) but it could be.

And 4-5Gbps aren't that different.

I think the length of the card is also important buying factor....Caymans are really longer than "normal" cards...

Partner Cayman PRO's are generally smaller.

 

[Pressure]

I think the length of the card is also important buying factor....Caymans are really longer than "normal" cards...

As long as they adhere to ATX specifications, there shouldn't be a problem unless the user bought a case that doesn't live up to the aforementioned specifications.

 

[mczak]

And 4-5Gbps aren't that different.

Well there has been much talk about redwood vs. juniper mc phy size, which is also only the difference between 4 and 5 gbps...
Also, the pcb is pretty much the same for HD6970, so it looks like the HD6950 PCB might be made for more than 5Gbps. I guess though your answer means on the pcb level there wasn't really anything special necessary to make those higher speeds possible.

 

[Sxotty]

As long as they adhere to ATX specifications, there shouldn't be a problem unless the user bought a case that doesn't live up to the aforementioned specifications.

There are tons of cases where long video cards are a PITA. It is pretty silly to dismiss the problem with some hand waving about ATX specifications. Just like some video cards coolers make it nearly impossible to remove the card from the PCI-e slots b/c the lock is impossible to get to. Makes me wonder sometimes if the companies bother installing the cards in their cases at home

 

[Silent_Buddha]

It's still not AMD's problem if a case manufacturer makes an ATX case but doesn't correctly implement the ATX specifications. In other words, blame the case manufacturer's for cutting corners and ignoring the specs. The specifications are there specifically to make sure that anything correctly implementing the spec should work with other things correctly implementing the spec.

Now if someone tries to jam an ATX sized card into an Mini ATX case, well, then it's down to the consumer.

Regards,
SB

 

[A1xLLcqAgt0qc2RyMz0y]

GeForce GTX 590 in February with dual GF110

http://www.nordichardware.com/news/...orce-gtx-590-in-february-with-dual-gf110.html

Quote: The new graphics card comes with two GF110 GPUs and NVIDIA has chosen to go all in with the shaders. It goes with the top model of GF110 with 512 CUDA cores, which results in 1024 CUDA cores in total and 2 x 384-bit memory bus.

 

[Kaotik]

http://www.nordichardware.com/news/...orce-gtx-590-in-february-with-dual-gf110.html

Quote: The new graphics card comes with two GF110 GPUs and NVIDIA has chosen to go all in with the shaders. It goes with the top model of GF110 with 512 CUDA cores, which results in 1024 CUDA cores in total and 2 x 384-bit memory bus.

No, no and once again no.

There won't be official 590, simple as that, they will not break the PCIE specs.

The "leaked pic" is from AIB custom project similar to ASUS MARS/ARES projects, and projects similar to those are the only possible "dual GF110" that might ever come

 

[A1xLLcqAgt0qc2RyMz0y]

No, no and once again no.

Just keep repeating that to yourself over and over again and maybe just maybe it won't get released next month. And if it does just close your eyes and maybe it will go away.

So far rumors on nVidia's 500 series have all come true. First the 580 then the 570 and just recently the 560 TI.

In less than 30 days we will know if the 590 rumor also comes true.

 

[kresek]

what are the merits of a dual GF110 solution comparing to one using dual GF114's?

a single GF114 approaches the GTX 470 performance territory, using less than three fourths of the GF110 die size, (therefore) still having manageable power consumption - could it be just easier to bin commercially viable amounts of fully enabled, but lower clocking GF110 chips?

 

[Topman]

what are the merits of a dual GF110 solution comparing to one using dual GF114's?

a single GF114 approaches the GTX 470 performance territory, using less than three fourths of the GF110 die size, (therefore) still having manageable power consumption - could it be just easier to bin commercially viable amounts of fully enabled, but lower clocking GF110 chips?

I couldn't agree with you more.

Dual GF110 downclocked chips VS dual GF114 'ok' chips:
Dual GF114 far cheaper, less complex PCB(256 vs 320/384 buses), ~same performance.

The only one (official) reason for dual GF110 is 4-way SLI.

Yes, a dual GF110 has more performance potencial too
(HD5970 has a 400w TDP capable cooler, overclocking friendly: in stock, has 300w PCI-E limit)

waiting...

 

[Kaotik]

Just keep repeating that to yourself over and over again and maybe just maybe it won't get released next month. And if it does just close your eyes and maybe it will go away.

So far rumors on nVidia's 500 series have all come true. First the 580 then the 570 and just recently the 560 TI.

In less than 30 days we will know if the 590 rumor also comes true.

The "500 series rumors" were known to be true when GTX480 and GTX460 were launched - fully enabled "b revision" chips (as it's known, GF110 is really called 100b and GF114 is surely called 104b) would come sooner or later.

GF110/100b dual chip just isn't happening without heavily castrating it, and at that point dual GF114 would make more sense and at that point it wouldn't be winning anything anymore most likely (assuming 5970 vs 6870 CF difference is same still as it was when 6870CF was reviewed, 6870 CF in fact beats GTX560Ti SLI in TechPowerUps test suite, which is (by far) the most comprehensive testing suite out there

 

[Sxotty]

It's still not AMD's problem if a case manufacturer makes an ATX case but doesn't correctly implement the ATX specifications. In other words, blame the case manufacturer's for cutting corners and ignoring the specs. The specifications are there specifically to make sure that anything correctly implementing the spec should work with other things correctly implementing the spec.

Now if someone tries to jam an ATX sized card into an Mini ATX case, well, then it's down to the consumer.

Regards,
SB

You can fufill specs, but still make life hard. When your HDD is running into the Graphics card power connectors, or the SATA ports are hidden under a cooler and things it is a PITA. When your old case that worked great suddenly doesn't it is annoying. I think a lot of users don't bother to check out the dimensions in solid works to see how well everything will play in advance. I know I don't.

I don't think anyone is bad for making a long board, but at the same time a shorter one can be desirable to a segment of the market and there is nothing wrong with that either. Just like a power hungry GPU isn't necessarily bad for all segments. They are just characteristics of design that a consumer can make a choice on, just like the cost, acoustics, thermals etc.

 

[Silent_Buddha]

The "500 series rumors" were known to be true when GTX480 and GTX460 were launched - fully enabled "b revision" chips (as it's known, GF110 is really called 100b and GF114 is surely called 104b) would come sooner or later.

GF110/100b dual chip just isn't happening without heavily castrating it, and at that point dual GF114 would make more sense and at that point it wouldn't be winning anything anymore most likely (assuming 5970 vs 6870 CF difference is same still as it was when 6870CF was reviewed, 6870 CF in fact beats GTX560Ti SLI in TechPowerUps test suite, which is (by far) the most comprehensive testing suite out there

Well that article does go on to say...

GeForce GTX 590 will instead get considerably lower clock frequencies and voltages than GeForce GTX 580 that use a single GF110 GPU.

So it's entirely possible it could be clocked lower than GTX 570 with lower voltage. Which would enable full shader cores at respectable power use.

So, potentially performance around dual GTX 560's or maybe even dual GTX 570's depending on workload.

Of course, that then begs the question of why not use cheaper GF114 chips? Unless that would require so much power that a dual GF114 would actually be slower and use more power than a clock castrated dual GF110.

Regards,
SB

 

[Silent_Buddha]

You can fufill specs, but still make life hard. When your HDD is running into the Graphics card power connectors, or the SATA ports are hidden under a cooler and things it is a PITA. When your old case that worked great suddenly doesn't it is annoying. I think a lot of users don't bother to check out the dimensions in solid works to see how well everything will play in advance. I know I don't.

I don't think anyone is bad for making a long board, but at the same time a shorter one can be desirable to a segment of the market and there is nothing wrong with that either. Just like a power hungry GPU isn't necessarily bad for all segments. They are just characteristics of design that a consumer can make a choice on, just like the cost, acoustics, thermals etc.

Sure but again that's a problem with the case manufacturer. Hell, full ATX Motherboards are often a PITA to work on with certain "ATX" cases even if you had a short video card. With those same HDD's which could block a longer ATX sized graphics card also blocking the SATA ports on the MB. Or optical drives blocking the use of memory slots.

Sure it's obviously a PITA. But again, IMO, this would be more a case of case manufactures either making non-standard ATX sized cases (where drives can overhang the
MB) or just making plain unfriendly to work on cases.

I certainly understand the frustration (I have a P-183 which is a PITA to work on even with a short video card, but the results for me are worth it. ) but I think the blame is misplaced.

Regards,
SB

 

[Sxotty]

I certainly understand the frustration (I have a P-183 which is a PITA to work on even with a short video card, but the results for me are worth it. ) but I think the blame is misplaced.

Regards,
SB

I have a P-180 myself hehe. I guess that explains something, but it was really the mobo with sideways SATA ports that drove me bonkers. It got them out of the way of stuff on the motherboard, but made connection inside the case a huge hassle.

 

[Silent_Buddha]

Yup yup, exact same thing here with the sideways connectors. The P-183 made things very very slightly easier as you could route some of the cables around the back of the case under the MB tray to the bottom HDDs. Routing power cables that way also makes thing slightly easier and a LOT less cluttered. And without the power cable clutter, removing and installing a video card is a lot less of a pain.

Still a PITA to work with, but as mentioned the end result is worth it. I just had to redo things recently when I upgraded to Sandy Bridge from Lynnfield. Much easier, but still a PITA this time since I already had many of the cables pre-routed.

Regards,
SB

 

[Kaotik]

Well that article does go on to say...



So it's entirely possible it could be clocked lower than GTX 570 with lower voltage. Which would enable full shader cores at respectable power use.

So, potentially performance around dual GTX 560's or maybe even dual GTX 570's depending on workload.

Of course, that then begs the question of why not use cheaper GF114 chips? Unless that would require so much power that a dual GF114 would actually be slower and use more power than a clock castrated dual GF110.

Regards,
SB

The GTX570 uses ~50W more power than even OC'd 560Ti (560Ti@900MHz), sure you can cut that a bit by voltages and clocks, but when the difference in performance of the 2 cards is aroun 10-20%, you don't have that much room to go down before it would just make more sense to use 114.

And even GTX560 at stock speeds is using too much power to be put in dual-config without cutting clocks/units/voltages

 

[Arty]

The "500 series rumors" were known to be true when GTX480 and GTX460 were launched - fully enabled "b revision" chips (as it's known, GF110 is really called 100b and GF114 is surely called 104b) would come sooner or later.

GF110/100b dual chip just isn't happening without heavily castrating it, and at that point dual GF114 would make more sense and at that point it wouldn't be winning anything anymore most likely (assuming 5970 vs 6870 CF difference is same still as it was when 6870CF was reviewed, 6870 CF in fact beats GTX560Ti SLI in TechPowerUps test suite, which is (by far) the most comprehensive testing suite out there

I gotta see this.

 

[AlphaWolf]

I gotta see this.

If you check out their 6870 crossfire review the 6870s beat out the 5970 by 10% or more in the higher resolutions (1680, 1920 and 2560). The 560TI sli review has the SLI setup barely edging out the 5970 by 2-4% in those same resolutions.

So ya, it looks like crossfire is scaling better and 2x560 probably isn't going to compete with 6990.