so I spent a little time with the reference 980 today...

[Silverforce11]

and for those that think blower cards dont really make a difference to case temps I did a little test in Crysis 3.

with my 780 at stock 1019 boost the I would hit 65 C

with my 970 oced to 1458 boost I would hit 64 C

with the reference 980 oced to 1465 I only hit max temp of 57 C on cpu

now of course that can vary with ambient temp and I know the day I tested the 970 it was a couple degrees warmer than today. when I tested the 780 it was actually cooler than today. so if anything my cpu temps would be another 1 or 2 degrees cooler if a tested on the same day as 780.

That is the main advantage of blower cards, that heat gets out the case instead of dumping 200W in such a small space, as well as infrared radiating on components nearby.

Open air cards need very good system ventilation and in CF/SLI, defeats their advantage of quiet operation when you need a lot of case fans to keep them from overheating.

 

[toyota]

1503 on stock volts of just 1.187. so much for ASIC score meaning much as its only 66%.

 

[Subyman]

The power usage is nice. 980 overclocked is still more power efficient compared to a stock 780. 50% more clocks while being on parity with energy usage. That's a big jump.

 

[Majcric]

No denying the 980 is great card, but on the other hand, it's a shame Nvidia gimped Ti with such low VRAM vs its performance. I'm pretty Nvidia never released Ti 6gb to avoid comprising the sales of the 980, well my opinion anyway.

In fact, I'd love to see Ti in certain performance scenarios with more VRAM.

 

[lavaheadache]

1503 on stock volts of just 1.187. so much for ASIC score meaning much as its only 66%.

mine is 67%

 

[toyota]

No denying the 980 is great card, but on the other hand, it's a shame Nvidia gimped Ti with such low VRAM vs its performance. I'm pretty Nvidia never released Ti 6gb to avoid comprising the sales of the 980, well my opinion anyway.

In fact, I'd love to see Ti in certain performance scenarios with more VRAM.

they were going to have 6gb on the 780 and 780 ti but only a couple 780 models got the 6gb and no Ti ever came with it. its clear now they told card makers not to make a 6gb Ti with the 980 release looming.

 

[toyota]

mine is 67%

what about your voltage?

 

[tviceman]

So all in all, are you glad you went with a reference 980 over, say, the MSI gaming 980?

 

[toyota]

So all in all, are you glad you went with a reference 980 over, say, the MSI gaming 980?

overall yes

 

[Z15CAM]

Get a Koolance RP401X2 , a Coolgate Rad and a couple Nouctia Fans with a PMW400 pump and shoot for the STARS with either an EK or XSPC RAZOR GPU Water Block.

 

[ams23]

right but that is based on the TDP they gave the 980.

One would think it is based on actual average power consumed by the GPU in each specific game that they benchmarked. That said, if you look at the chart NVIDIA released, it is clear that perf. per watt differences between GTX 980 and GTX 680 are very much game-dependent (in some cases the improvement is >2x but in most cases the improvement is somewhere between 1.7-1.9x in the sample of games that they provided):

With GTX 980, keep in mind that average power consumption can be less than GTX 680 when gaming, while at the same time peak power consumption can be more than GTX 680 when gaming. For instance, see the Metro LL average and peak consumption graphs here: http://www.techpowerup.com/mobile/reviews/NVIDIA/GeForce_GTX_980/24.html

When it comes to total energy consumption, the more important result is average power consumption.

 

[Rvenger]

mine is 67%

65.8%... and mine buzzes.

 

[caswow]

What they said is 2x improvement in perf. per watt of GTX 980 reference vs. GTX 680 reference, which appears to be pretty accurate (note also that their comparison is GPU-specific power consumption, and not system power consumption, because they are trying to isolate the GPU perf. per watt).

its strange nobody is talking about the die size of the new maxwell "midrange" cards.

gk104 is 294mm² and gm204 is 398mm². this means theoratically their new midrange is way bigger and way pricier to produce. 36% bigger die its almost as big as hawaii. either big maxwell wont be much faster than its "midrange" brothers or it will be freaking huge because maxwell is all in all bigger than its kepler cores.

 

[ShintaiDK]

its strange nobody is talking about the die size of the new maxwell "midrange" cards.

gk104 is 294mm² and gm204 is 398mm². this means theoratically their new midrange is way bigger and way pricier to produce. 36% bigger die its almost as big as hawaii. either big maxwell wont be much faster than its "midrange" brothers or it will be freaking huge because maxwell is all in all bigger than its kepler cores.

Price on 28nm have decreased quite alot since. Yield matured as well.

I would even guess that GM204 today is cheaper than GK104 was at its introduction.

 

[RussianSensation]

its strange nobody is talking about the die size of the new maxwell "midrange" cards.

gk104 is 294mm² and gm204 is 398mm². this means theoratically their new midrange is way bigger and way pricier to produce. 36% bigger die its almost as big as hawaii. either big maxwell wont be much faster than its "midrange" brothers or it will be freaking huge because maxwell is all in all bigger than its kepler cores.

Well, even a 550mm2 28mm Big Maxwell would be a huge step up in speed over the 980. Don't forget it'll probably come with at least 6GB of VRAM too. With no response from AMD in the near future, I would have liked for NV to wait until April-May 2015 and instead release a 20nm GM200. I think they are likely to do a 28nm version though and maybe leave 20nm for the refresh.

Toyota, congrats on the 980. With your 780 overclocking so poorly and your 980 a solid overclocker, looks like you are getting a 40% increase in performance in demanding titles. Pretty good as 780 @ 1150 was a pretty fast card.

Btw, what CPU cooler do you have? I think the more powerful the CPU cooler is, the less the impact of after market air cooled cards dumping air into the case. And frankly your mid-60s temperatures were nothing to worry about. I mean it would have mattered if your CPU ran at 80C and the card forced instability on your CPU overclock. From what I've read on Titan Black and 780Ti reference overlocking, the reference cooler loses badly in noise and temperatures to after-market solutions such as MSI Gaming or Windforce. I think one of the main reasons the reference 980 cards use so much less power at load is because they had power load balancing, likely missing on custom PCB after-market cards, and missing on all 970 cards.

P.S. It still pisses me off that NV released a 4GB 980 $550 card and yet it ships 6GB 970M and 8GB 980M for mostly 1080p laptops. Clearly NV did this on purpose to leave room for 970/980 4GB price drops next year. Once AMD responds, NV will release 8GB versions and lower prices.

 

[lavaheadache]

It's called sandbagging.

 

[swilli89]

Price on 28nm have decreased quite alot since. Yield matured as well.

I would even guess that GM204 today is cheaper than GK104 was at its introduction.

That's a pretty naive guess, absolutely no offense intended. With a die area that is about 33% larger you're guessing that the already mature 28nm process jumped more than 33% after the process was in mainstream production. I really doubt this is the case for any major production process no matter the length of time it "matures".

 

[ShintaiDK]

That's a pretty naive guess, absolutely no offense intended. With a die area that is about 33% larger you're guessing that the already mature 28nm process jumped more than 33% after the process was in mainstream production. I really doubt this is the case for any major production process no matter the length of time it "matures".

Are you saying 28nm gate cost and yield is static over time at foundries?

 

[swilli89]

Are you saying 28nm gate cost and yield is static over time at foundries?

No I didn't even imply that. While none of us non-insiders know the actual average yield for these designs; lets call them 90% right now. Yields would have had to previously been around 68% to make your assumption somewhere close to being correct. I know enough from the articles that I have read that no one produces at 68% yields.

I'm sure there has been efficiency and yield gains. Just saying I doubt that a chip 33% larger on the same process is somehow cheaper to make now due to process maturity.

 

[tviceman]

Well I got my MSI Gaming 980 today. It's got two 8 pin connectors. Some of you might be happy about that, but I'm running on a 450watt PSU that doesn't have two 8-pin connectors so I had to use an adapter. Not an issue, but case clutter is a pet peeve of mine. On top of that, after reading this thread (Toyota's experience) with a reference 980, I'm having slight buyers remorse now. Hahaha. I should get a kill-o-watt so I can see what my power consumption is at. I think the open-air cooler is probably the right card for me once I move to the RVZ01 case, as it puts the GPU in it's own compartment and stops the spread of heat for the most part.

 

[ShintaiDK]

No I didn't even imply that. While none of us non-insiders know the actual average yield for these designs; lets call them 90% right now. Yields would have had to previously been around 68% to make your assumption somewhere close to being correct. I know enough from the articles that I have read that no one produces at 68% yields.

I'm sure there has been efficiency and yield gains. Just saying I doubt that a chip 33% larger on the same process is somehow cheaper to make now due to process maturity.

Yield is only 1 part. Wafer cost have gone down as well.

From my calculations, TSMC's average 28nm wafer price by quarter has been as follows:

Q4'11 = $7,000 (USD)
Q1'12 = $5,000
Q2'12 = $4,000
Q3'12 = $3,000
Q4'12 = $2,500

(note these are not "guesses", these numbers drop out of TSMC's published financials)

http://forums.anandtech.com/showpost.php?p=34512855&postcount=1040

 

[tviceman]

No I didn't even imply that. While none of us non-insiders know the actual average yield for these designs; lets call them 90% right now. Yields would have had to previously been around 68% to make your assumption somewhere close to being correct. I know enough from the articles that I have read that no one produces at 68% yields.

I'm sure there has been efficiency and yield gains. Just saying I doubt that a chip 33% larger on the same process is somehow cheaper to make now due to process maturity.

Cost goes down over time as yields improve and wafers get cheaper. Otherwise there would be no way GK104 cards would be selling for $279 max at any sort of profit when they debuted at $499. GK104 would have been EOL'd way before GM204's arrival if Nvidia had to eat a loss on each GK104 based card still in inventory. But how much does cost go down over time? I have no idea.

 

[rtsurfer]

yes the reference 980 cards seem pretty efficient. mine will run at 1503 on stock voltage which is pretty good. my stock voltage seems to be 1.187. it will always start out at 1.206 then drop down to 1.187 at the same time it drops the clocks 13 mhz. after that it will never go back up or drop if under full load. that drop seems odd but reviews have seen it too plus its consistent in every bench.

Nice card but that stock voltage is High.

1.187V after the drop is what I used to get on my 780 after doing +30mv(the max) in Precision X. Dont remember how much it was on stock & I don't have the card anymore.

 

[toyota]

what I dont get is why the reference cards can oc as good or better than the fancy non reference cards with all their improved cooling and components and higher voltage. this card can do 1503 at 1.187 which seem crazy. it will also do that at much less power consumption than the non reference cards at factory clocks.

I have not even seen what this card can really oc too as I tried too much voltage at 1522 and made it throttle a few times. I think it will do at least 1520 to 1530 with only a bump or two in voltage.

 

[swilli89]

Otherwise there would be no way GK104 cards would be selling for $279 max at any sort of profit when they debuted at $499.

Yes costs come down over time in pretty much every non supply-limited industry on the planet. But without knowing their original profit margin, merely knowing they dropped the price 40% doesn't tell you anything. They could have gone from a 60% profit margin to a 20% for all we know. Their pricing implies zero information about their cost structure.