Tuesday, March 6th 2012

GK104 (GTX 680) Has 17% Higher Compute Power Than Tahiti (HD7970): Report

A report by 3DCenter.org, which takes into account the specifications of NVIDIA's GK104 GPU, based on clock-speed and specs. leak estimated its shader compute power to be 17% higher than that of AMD's Tahiti in its Radeon HD 7970 avatar (4.46 TFLOPs single-precision floating point). Based on the said specifications, the report also hypothesized things such as memory bandwidth, ROP performance, and texture performance; and compared it to those of the Radeon HD 7970 and Radeon HD 7950.

Source: 3DCenter.org
Add your own comment

48 Comments on GK104 (GTX 680) Has 17% Higher Compute Power Than Tahiti (HD7970): Report

#1
Crap Daddy
Inceptor said:
Yes, but that depends on whether those benchmarks are maximum possible for a 680 and whether or not the 680 will be a single sku or multiple skus. No way to know; if the naming scheme is changed and there are multiple 680 skus, those benches could be for a '680 ultra' for all we know.
Right. That would be the fully enabled GK104 named GTX680. I wouldn't make assumptions on performance based on 3dcenter speculation on a speculation of specs. What we do know is that now NV has one card ready based on the "performance" chip of the Kepler series namely the GK104 which apparently can match and surpass Tahiti XT.
Posted on Reply
#2
Inceptor
Speculation and assumption is what threads like this are all about. And the key word in what you just said is "apparently".
Posted on Reply
#3
bmwmaster
I hope it will support 3 Displays connection. This is the only feature which ATI has and Nvidia only has on 590 GTX.

It would be nice for multimonitor gaming.
Posted on Reply
#4
Crap Daddy
Again, regarding this, the only "speculation" comes from Kyle a while back when he said the card will support NV surround out of the box.
Posted on Reply
#5
n0tiert
as long as W1zzard dont get his hands on it , i belive none of the green fanboys rumours
Posted on Reply
#6
Inceptor
bmwmaster said:
I hope it will support 3 Displays connection. This is the only feature which ATI has and Nvidia only has on 590 GTX.

It would be nice for multimonitor gaming.
I think it would be better to hope that power consumption is reasonable vs the AMD gpus.
I'd like to purchase an NV gpu, but only if power consumption is not too high.
Performance is performance, whether you go NV or AMD, games play well enough; the consoles haven't made their next surge.
What's important is power consumption and feature sets; like having multi-monitor as an option, whichever card you choose. Options are good. Limitations are bad, unless you're on a very restrictive budget.
Posted on Reply
#7
Dj-ElectriC
If we assume theoretically that the GTX680 will be 17% faster then the HD7970. If it wont be at least a decent overclocker... then.... you know the rest.
Posted on Reply
#9
Aquinus
Resident Wat-man
Let's keep in mind how well the 7970 over-clocks. If Kepler over=clocks just as well, it will give the 7970 a run for its money (if the source is true, we should still take it with a grain of salt) but if Kepler doesn't over-clock as well, forget it because Tahiti has proven that it can over-clock very well. Just imagine when the 7990 gets released and if it does have a "Turbo" bios switch, remember what happened with the GTX 590?

I really hope that nVidia learned from their mistakes in the past, there is a lot of hopes and hype for Kepler and I can only imagine how disappointed people will be if it doesn't perform, but keep in mind, with the failure of Bulldozer AMD has clearly went full force forward with their GPUs.

The waiting game isn't good for business (for nVidia, that is anyways).
Posted on Reply
#10
xenocide
I take all these numbers with a grain of salt... I don't think Nvidia is going to release a flagship card that isn't at least as powerful as AMD's offering...
Posted on Reply
#11
1c3d0g
radrok said:
If this is true then the GK100/110 will be a compute monster, paired with a wider memory bus, well... it makes a lot of sense because Nvidia has always made big dies in favour of compute power, a refined Fermi (aka Kepler) will make for a nice CUDA monster :) that's what I'm looking for.
I God damn hope so! Nothing makes my heartbeat go faster, except when I see a next-generation GPU completely pulverize BOINC/Folding@Home projects! :rockout:
Posted on Reply
#13
Aquinus
Resident Wat-man
1c3d0g said:
I God damn hope so! Nothing makes my heartbeat go faster, except when I see a next-generation GPU completely pulverize BOINC/Folding@Home projects! :rockout:
Touché! Got to love Compute power!
Posted on Reply
#14
Shihabyooo
xenocide said:
I take all these numbers with a grain of salt... I don't think Nvidia is going to release a flagship card that isn't at least as powerful as AMD's offering...
+1.
Nvidia might try to push it's CUDA computing further, but not in expense of 3D performance. I don't see Geforce branded cards being sold for Folding much.

Another issue with all these speculations/leaks matter, is the naming schemes. Assuming the 680 succeeds the 580, I don't see why it'd be powered by the GK104 with only a 256bit wide memory bus. Unless there's been some announcements that I missed.
Posted on Reply
#15
Crap Daddy
Shihabyooo said:
Another issue with all these speculations/leaks matter, is the naming schemes. Assuming the 680 succeeds the 580, I don't see why it'd be powered by the GK104 with only a 256bit wide memory bus. Unless there's been some announcements that I missed.
This is the odd part. NV is "adapting" to the market. GK104 was designed to replace the GF114 as the "performance" chip. Due to the "great" performance of Tahiti they realize this chip can compete with AMD's high-end. It can even be better. Since the big boy, GK100/110 whatever, is still not ready they decide to launch the GK104 under the name GTX680 since it outperforms TahitiXT (how and by what margins remains to be seen). The bad news is that by naming it 680 it will de facto replace the 580 at a similar price point (hope I'm wrong) bringing an increase in performance (based on what the 7970 has above the 580) of around 30% at maximum.

As for the GK110 which should have been the real successor of the 580, there's no need for it now in this GPU market. The Teslas and the Quadros can wait a little longer.
Posted on Reply
#16
m1dg3t
jmcslob said:
So what does that mean when the two are compared all around?
Doesn't mean squat because it's still all speculation!! I can speculate the winning loto #'s for tonight's lottery too if i want :rolleyes:

All of these gk104 thread's make me :roll: They are FUDtastic! :D Maybe in 2 or 3 week's there will actually be a thread about this card :o

In the meantime, another post! YAY! :rolleyes:
Posted on Reply
#17
radarblade
Considering the massive improvements Nvidia brought with the Fermi GPU's after the 2XX series, I'm pretty confident that it's more than just the 17% increment with Kepler against Tahiti in terms of performance. Notably too, AMD pretty much is late in terms of driver optimization which always gives Nvidia a head start in the market. AMD would simply knock down it's prices when Kepler launches thinking they would sell more which is an obvious fail. :D
Posted on Reply
#18
Aquinus
Resident Wat-man
radarblade said:
Considering the massive improvements Nvidia brought with the Fermi GPU's after the 2XX series, I'm pretty confident that it's more than just the 17% increment with Kepler against Tahiti in terms of performance. Notably too, AMD pretty much is late in terms of driver optimization which always gives Nvidia a head start in the market. AMD would simply knock down it's prices when Kepler launches thinking they would sell more which is an obvious fail. :D
AMD doesn't care, they're selling their cards now because Kepler is taking so long. By the time AMD has to drop prices, the people who were going to buy high end 7000-series chip most likely already have and if Kepler takes much longer it will have to compete with the 8000-series. :banghead:
Posted on Reply
#19
USA1fan
I'm new on these forums (lurk occasionally though). This situation sounds suspiciously like a period of time years ago when AMD released the 9700Pro cards while NV was still hanging on the GeForce 4 Ti series cards. At the time NV kept spouting upcoming specs, and promising that their new, leading-edge-performance product was right around the corner. I had a Ti4600 that I held onto for over 9 months while several of my buddies enjoyed their 9700Pro cards, running circles around my video performance in games. When the GeForce 5 (aka GeForce FX 5x00) cards came out, not only did they require the dustbuster coolers, they were still pretty much inferior to the ATI hardware, even with NV using driver hacks to (behind the scenes) reduce actual graphics capabilities during gameplay (to pump up overall apparent performance).

In the end, I joined my buddies with the 9700Pro, after being an NV loyalist ever since the GF2 series. I was happy for another year or two on, before NV finally got their overall package right, with the FX 6x00 series (still debatable then). Eventually, ATI was caught doing some of the same 'dynamic optimization' of their own, leveling the playing field with future hardware, and NV even released some actual class-leading product again. So I'm not saying don't buy nVidia. What I AM saying is if you need good performance now, buy the best you can afford from what is CURRENTLY available. If you hold your breath, you may pass out before the thing you think you're waiting for actually arrives. Specs are just vapor until the hardware's available for you to buy. Don't get caught by the hype. ;)

(and right now, regardless of rumors, AMD / ATI rules the roost again)
Posted on Reply
#20
Benetanegia
Texturing power is +50% higher than GTX580 despite being smaller than Tahiti. Fermi already had some very heavy optimizations on texturing.

GTX580 = 49 GTexel/s

HD6970 = 84 GT/s

HD7970 = 118 GT/s

So you can assume that both compute and texturing power are way above GTX580 (and by extension Tahiti to a lesser extent).

Only ROP power would be lower. But I don't really believe the specs in which the 3dcenter calculations are based on, because I don't think that Nvidia would increase compute and texturing power so much, and put a ROP power that is actually smaller than on a GTX560. Unless it has 64 or 48 (decoupled) ROPs, in wich case it would be higher than HD7970.

And to me, 700 Mhz is just not very believable for a 320mm^2 chip on 28 nm. Going backwards by so much? Naaaah. Smoke screen.
Posted on Reply
#21
Nihilus
So the GTX 680 is basically an HD 7950 with higher Rechenseistung? In Nvidia's defense I'm sure it will have higher power consumption and cost too! :laugh:
Posted on Reply
#22
Casecutter
This is interesting but compute has been in Nvidia favor so not surprised. Then I agree that if this GK104 (undoubtedly the Uber/FTW version clocks) is at 700MHz, what’s the lowest versions get for clocks?

I have a hard time believing Nvidia won’t set an MSRP for a card that could spar with a 7970 for any less than $480. Even with a more cost effective die and 256-Bit vs. 348-Bit, I just don’t envisaging Nvidia under cutting price more than say 12% at the start (not part of their DNA).

I won’t go as far as saying, AMD artificially jacked-up their price, but seized ~10% premium because at the beginning of January they probably didn’t figure Nvidia would play it this way, it is uncharacteristic of them. AMD probably erred to normal track and figure if the GK104 if used to combat the 7970 they still have wiggle room to find traction in the appropriate $/performance to new competition. At least this isn’t a GTX 260/4870 type of "uh-Oh" moment… AMD has probably had 4-5 weeks to deliberate how this might develop, and probably made contingencies. So now if we find rebates flowing next week(s) that may be a “tale-tell” of what’s coming.

Finally, who know if Charlie has this right... but what if something close to this has occurred, could it have an effect to this short term? Pent-up demand and low stock if close to a 7970, there's still guy's that would dole-out top dollar. :ohwell:
http://semiaccurate.com/2012/03/07/tsmc-suddenly-halts-28nm-production/
Posted on Reply
#23
Boomstick777
jmcslob said:
So what does that mean when the two are compared all around?
EDIT

Get the DCUII GTX 670, that is a good GPU...
Posted on Reply
Add your own comment