Even More GeForce GTX 980 and GM204 Specs Tumble Out

[rtwjunkie]

I have to admit, despite my reservations about these cards for replacing my 780, I think a 970 would be a huge upgrade for my fiance's GTX 660 Ti FTW. She uses Photoshop, and games, so the whole thing sounds like a big plus for her. I'll be looking forward to the first 970 reviews.

 

[D007]

.

I mentioned this before and I am not a GPU guru, but the GTX980 64 ROP should benefit 4k+ users. The Titan only has 48.

That's what I care about.
Just got a 4k.
ty.

 

[HumanSmoke]

According to SiSoft, the GTX 980 is boosting to 1630MHz from its 1126MHz base clock. No wonder the thing is close to GK 110 !

And it looks like AMD are having second thoughts on the 285X....unless they plan on naming it something else

 

[sergionography]

According to SiSoft, the GTX 980 is boosting to 1630MHz from its 1126MHz base clock. No wonder the thing is close to GK 110 !

And it looks like AMD are having second thoughts on the 285X....unless they plan on naming it something else

Regardless of clock speed the Maxwell cores if I rememer correctly should be about 30-35% faster than Kepler that's before we even look at power use. That being said 2048 Maxwell cores should be a match to the 2880kepler cores on the gtx780ti, this is at the same clock, so this having over 25% clock boast over it will sure give it an advantage but then let's factor in bandwidth and what not then I won't be surprised if gtx980 was faster than gtx780ti by 10% and maybe reaching 20% in some best case scenarios.

As for amd they also have some tricks up their sleeves, except for the power use. even though they keep improving efficiency in graphics, instead of turning it into better power saving they seem to just use that real estate to add compute features. as for the r9 285 it's obvious amd crippled the chip as much as possible to kind of show off how they can match the 280-280x with much less resources and bandwidth, so take into consideration that it's a harvested chip with disabled units so only the full 285x will show us what this new amd architecture (or GCN tweak I shall say) Is capable of. However to target gtx760 as a competitor is just a very poor choice indeed timing wise. the only problem with releasing the r9 285x now is that it will most likely beat Hawaii especially if the chip turns out to have a 384bit bus. So amd won't release it yet

 

[HumanSmoke]

Regardless of clock speed the Maxwell cores if I rememer correctly should be about 30-35% faster than Kepler that's before we even look at power use.

There is no strict baseline to work from since GM 107's analog is GK 106 (wider bus), and the GK 107 isn't built for performance - it has no boost capability. But if you compare bus width to bus width (128-bit) then
GM 107 (GTX 750 Ti) vs GK 107 (GTX 650) : 640 core vs 384 (66.67% advantage), both feature 16 ROPS, both have a 128-bit bus width, 40 TMU vs 32 for the GTX 650 ( 25% advantage) , memory clock/bandwidth : 5400 effective vs 5000 / 86.4GB/s vs 80 ( 8% advantage), clocks are nominally similar 1020 base/ 1085 boost (actual average 1140MHz) vs 1058 for the GTX 650 ( 7.8% advantage). Power consumption is similar.
Actual performance difference: 79.4% at 1920x1020, and 83.9% at 2560x1600.............so, not quite as pronounced as it would seem given the difference in specification

This is obviously a gaming-only scenario and is only applicable to the two GPUs - it doesn't account for many other metrics, or how the cache hierarchy scales on a larger die. That aside, I believe the core speed increase will be a major factor in the performance boost of the GM 204 WRT GK 110. We'll see when GTX 980 review benchmarks surface and how overclocking a 780 Ti against it affects the differential. The only real measure of gains in architectural efficiency will need a clock-for-clock (within reason) comparison.

 

[Sony Xperia S]

I think AMD Radeon R9 285X with full Tonga with good optimised drivers and proper frequencies can compete very successfully at least with GTX 970.

 

[sergionography]

There is no strict baseline to work from since GM 107's analog is GK 106 (wider bus), and the GK 107 isn't built for performance - it has no boost capability. But if you compare bus width to bus width (128-bit) then
GM 107 (GTX 750 Ti) vs GK 107 (GTX 650) : 640 core vs 384 (66.67% advantage), both feature 16 ROPS, both have a 128-bit bus width, 40 TMU vs 32 for the GTX 650 ( 25% advantage) , memory clock/bandwidth : 5400 effective vs 5000 / 86.4GB/s vs 80 ( 8% advantage), clocks are nominally similar 1020 base/ 1085 boost (actual average 1140MHz) vs 1058 for the GTX 650 ( 7.8% advantage). Power consumption is similar.
Actual performance difference: 79.4% at 1920x1020, and 83.9% at 2560x1600.............so, not quite as pronounced as it would seem given the difference in specification

This is obviously a gaming-only scenario and is only applicable to the two GPUs - it doesn't account for many other metrics, or how the cache hierarchy scales on a larger die. That aside, I believe the core speed increase will be a major factor in the performance boost of the GM 204 WRT GK 110. We'll see when GTX 980 review benchmarks surface and how overclocking a 780 Ti against it affects the differential. The only real measure of gains in architectural efficiency will need a clock-for-clock (within reason) comparison.

Well I didn't get too fine with the calculation and just went by what nvidia claimed about the claimed improvement in the Maxwell architecture and it's performance vs Kepler so in other words I made a general assumption similar to what nvidia might've expected out of this chip when they started the design process. On another note the die size I think is the perfect size where u have plenty resources yet u don't miss on clock speed unlike those huge chips of the titan and 780 ti. And if this gm204 is truly the successor of gk204 then it's a great step forward for sure

 

[Sony Xperia S]

Maxwell is to nvidia what Core 2 was to Intel (when Pentium M's qualities were transferred to the desktop CPUs).

Only AMD is yet to figure it how to improve on that and finally put Intel and nvidia to their knees.

Looking hopeful and optimistic to 2015 and 2016 for AMD.

 

[64K]

I ran across this while looking at some pics of GTX 980s. I noticed the dual 8 pin power connectors so I followed the source and read about it. It will also have an updated Titan cooler with a TDP of 275 watts. This card may end up being able to do some crazy high clocks on air.

http://wccftech.com/colorful-unveil...rce-gtx-980-graphics-card-triple-fan-cooling/

 

[Vayra86]

first post?

Gotta start somewhere amirite?

 

[Sony Xperia S]

From the review http://wccftech.com/nvidia-maxwell-...0-15-faster-r9-290x-gtx-970-10-faster-r9-290/

I see that GTX 970 is on par with R9 290X.

AMD is in a deep trouble and need to urgently slash the price from $459 to $349 max.

 

[HumanSmoke]

Only AMD is yet to figure it how to improve on that and finally put Intel and nvidia to their knees.

AMD is in a deep trouble and need to urgently slash the price from $459 to $349 max.

Well, that changed fast.

BTW: That scenario of cost cutting will put AMD closer to their minimum cash reserve requirement ($600 million) to stay solvent since the price cut you envisage will need a price adjustment down the product stack - and of course the graphics business is likely the only one that turns a significant profit for the company.

 

[Sony Xperia S]

Well, that changed fast

Very "funny". I am yet to figure it out where and how exactly you see that both quotes contradict each other.

Perhaps in your hating fantasy?

 

[Vayra86]

Well, let's see what GTX 970 will be priced at first...

No doubt NV will market these chips as the new high end for a while, especially since those benches equal the current high end. I'm quite sure they'll put a 290x +10/20 dollar price tag on them. They can sell these for more as total cost of ownership will be somewhat lower due to power consumption.

 

[HumanSmoke]

Very "funny". I am yet to figure it out where and how exactly you see that both quotes contradict each other.

Um, maybe because you haven't figured out that the second statement calls for a course of action that will directly impact it's ability to attain the first.

AMD has just under $950m in cash and securities. It needs to maintain a $1 billion dollar level to sustain itself (R&D). Every dollar under that figure DIRECTLY removes AMD's ability to develop product. At $600m AMD will cease to be viable (their estimate not mine). The GPU division made $82 million income last quarter while the processor division made $9 million.

Perhaps in your hating fantasy?

No, simple mathematics and the ability to parse a financial statement. Try it some time, there's a wealth of information you might learn from.

Well, let's see what GTX 970 will be priced at first...
No doubt NV will market these chips as the new high end for a while, especially since those benches equal the current high end. I'm quite sure they'll put a 290x +10/20 dollar price tag on them. They can sell these for more as total cost of ownership will be somewhat lower due to power consumption.

Very true. Driving AMD into oblivion is something that Intel (and Nvidia) could have achieved some time ago, but that brings more problems than it solves - monopoly and anti-trust issues, and a potential AMD buyer that has the will and financial muscle to trouble the other two (Samsung for instance who are looking to get into the GPU business). Intel's size and resources are pretty well known, but Nvidia also has in excess of $4 billion in cash/securities reserves and carries NO debt (AMD's debt burden is increasing and costs around $200m in interest alone annually)- it has the ability to prosecute a price war much more effectively than does AMD - it also has a larger market and an elevated brand awareness than the Sunnyvale company.

All that would achieve is to drive AMD into the arms(!) of a company better suited financially, and decrease Nvidia's financial base. Hardly an advantageous situation.
It's a pity the "big picture" isn't as alluring as swinging from unfounded optimism to outright pessimism for some people.

 

[Sony Xperia S]

...

I understand you completely. I see the financial report and it is indeed very unlikely that AMD will survive very long these pressures, they have 2.2 B dollar debt.

Guess that ATi purchase did more harm than expected benefits.

But... I still believe that a 'miracle' can happen and with small budget for R&D and projects, they can achieve a lot!

Well, let's see what GTX 970 will be priced at first

GTX 970 is expected around $329. See the link if you will and wish.

 

[HumanSmoke]

But... I still believe that a 'miracle' can happen and with small budget for R&D and projects, they can achieve a lot!

Well you know what Nietzsche said about hope
"Hope in reality is the worst of all evils, because it prolongs the torments of man"

AMD aren't going anywhere. By the same token, I doubt that they will become a significant player (nor will Nvidia for that matter IMO) - too much competition from companies with significantly more resources, and a more determined and ruthless approach.
AMD is already swimming in red ink, so I doubt Nvidia would initiate a price war just to make the swimming pool a bit larger and cut it's own revenue into the bargain.

 

[Vayra86]

I understand you completely. I see the financial report and it is indeed very unlikely that AMD will survive very long these pressures, they have 2.2 B dollar debt.

Guess that ATi purchase did more harm than expected benefits.

But... I still believe that a 'miracle' can happen and with small budget for R&D and projects, they can achieve a lot!



GTX 970 is expected around $329. See the link if you will and wish.

The unfortunate reality is that most miracles are bad for AMD, or they just don't manage to really cash out on them. Look at Mantle... Nobody will tell you to take an R9 so 'you have Mantle'.

 

[GhostRyder]

The unfortunate reality is that most miracles are bad for AMD, or they just don't manage to really cash out on them. Look at Mantle... Nobody will tell you to take an R9 so 'you have Mantle'.

I disagree on that since games are starting to show up more and more with Mantle support recently (GTA V is a big name). But yea most people do not buy a GPU based on gameworks, mantle, physx, or what not but the tech itself is seeing a great adoption rate.

Guess that ATi purchase did more harm than expected benefits.

But... I still believe that a 'miracle' can happen and with small budget for R&D and projects, they can achieve a lot!

Well the point of the purchase has started to get more and more ground as APU's become more pronounced and the mobile market continues to shoot up. The biggest problem has just been the gambles they have taken with multi-core being king over single threaded and the delay if pushing HSA now. Future can only tell if they push it to the right people what will happen in the future.

Back to the actual subject, it seems that depending on the real price the 970 could be the king if all deals on GPUs replacing the R9 290 unless price cuts ensue. I wish more information about the 980 or benchmarks would surface because I find it hard to believe based on its specs it's going to be where it sits in the leaks. Going to be an interesting few weeks...

 

[HumanSmoke]

Maxwell launch slide deck courtesy of Videocardz, and some very interesting advances away from the whole f.p.s. and power use debate