We'll see what shakes out at Computex and E3. I'm waiting on a 6700k to replace my 2500k, so I have some time. Right now the pricing on the 980Ti makes it tempting, but not enticing enough to replace 290 CF. Maybe AMD can change my mind.
- Status
1) How do you know that AMD was aiming at the $1000 price point?
2) Nvidia never leaves money on the table. They are pricing 980Ti to compete.
I am predicting an 8 HBM Fiji model as it is not impossible technically. Everyone keeps saying 4GB HBM and I don't know the reasons. I believe there will be a 4GB version also and that is why Macri was saying 4GB can be enough. He said you can have less than and more than 4 stacks HBM on the interposer. The microbump pitch is 55um = 380/mm2.
I wrote this in another post to answer certain concerns.
An HBM stack is 5x7 mm and handles 1024 memory lanes. These lanes take at most 1/3 of this area [actually less] which is roughly 12mm2. Thefore 4096 lanes = 48mm2 and 8192 lanes = 96mm2. Remember, the whole base of the GPU is now available for I/O not just the perimeter. Even 8192 lanes should only be 1/6 the die area of Fiji [Reputed to be 550-600 mm2]. From an areal perspective, lack of space is not a problem.
For the huge interposer costs criticisms:
If you read the HBM papers, the interposer can be as large as the wafer [not talking costs here]. Interposers can also be as low as $2/200mm2. Remember, the steps to produce an interposer is not advanced lithography. Here is a 2012 item.
http://electroiq.com/blog/2012/12/lifting-the-veil-on-silicon-interposer-pricing/
At the recent Georgia Tech-hosted International Interposer Conference, Matt Nowak of Qualcomm and Nagesh Vordharalli of Altera both pointed to the necessity for interposer costs to reach 1$ per 100mm2 for them to see wide acceptance in the high-volume mobile arena. For Nowak, the standard interposer would be something like ~200mm2 and cost $2. The question that was posed but unanswered was: "Who will make such a $2 interposer?"
Less than a month later, this question began to be answered as several speakers at the year-ending RTI ASIP conference (Architectures for Semiconductor Integration and Packaging) began to lift the veil on silicon interposer pricing.
Sesh Ramaswami, managing director at Applied Materials, showed a cost analysis which resulted in 300mm interposer wafer costs of $500-$650 / wafer. His cost analysis showed the major cost contributors are damascene processing (22%), front pad and backside bumping (20%), and TSV creation (14%).
Ramaswami noted that the dual damascene costs have been optimized for front-end processing, so there is little chance of cost reduction there; whereas cost of backside bump could be lowered by replacing polymer dielectric with oxide, and the cost of TSV formation can be addressed by increasing etch rate, ECD (plating) rate, and increasing PVD step coverage.
Since one can produce ~286 200mm2 die on a 300mm wafer, at $575 (his midpoint cost) per wafer, this results in a $2 200mm2 silicon interposer.
It's not just tech forum discussions, AMD and NV share production facilities for their GPUs and they also share several AIBs who distribute both of products.
Once AIBs get their briefing and ES, both companies are fully aware of what the other is planning.
NV would NOT lower the price on the 970 & 980 without compelling competition because these SKUs actually have been selling amazingly well, with articles claiming even NV is surprised by how good they are selling!
During the 2x AAA bundle for 970/980, this is basically what I had predicted: they are trying to sell as many as possible at current prices, because come near AMD's launch, they will have to do a big price drop, especially for the 980.
happy medium
Lifer
- Jun 8, 2003
- 14,387
- 480
- 126
The aftermarket overclocking gtx980ti beasts will destroy 106% without water cooling.
crashtech
Lifer
- Jan 4, 2013
- 10,695
- 2,294
- 146
Maybe, but the price pressure is not on yet. We'll see how the math shakes out once Fiji is released.
crashtech
Lifer
- Jan 4, 2013
- 10,695
- 2,294
- 146
Computex is on this week, if they don't give up a few details in the coming days, I'd say you have a point.
Azix
Golden Member
- Apr 18, 2014
- 1,438
- 67
- 91
For the past few years, AMD has always made a policy of extreme secrecy before releases. They never talk about anything concrete; what little is released ahead of time is nothing more than marketing fluff.
Partly this is corporate culture. I think it may also be in part due to the backlash against John Fruehe's inaccurate and unethical comments prior to the Bulldozer release. To avoid another Fruehe situation, they decided the best solution was to just not allow anyone to say anything specific.
Partly this is corporate culture. I think it may also be in part due to the backlash against John Fruehe's inaccurate and unethical comments prior to the Bulldozer release. To avoid another Fruehe situation, they decided the best solution was to just not allow anyone to say anything specific.
Unless Fiji is in the new Mac Pro and AMD signed a "no leaks" contract with Apple.
Nvidia would have never wanted 980 Ti performance results to leak before launch, as it would have eaten into potential Titan X sales... AMD has every reason to leak Fiji performance results at this point...
They should have released it together with this movie:
https://www.youtube.com/watch?v=bS5P_LAqiVg
Movie is awesome btw.
https://www.youtube.com/watch?v=bS5P_LAqiVg
Movie is awesome btw.
Incidentally, with the release of the GTX 980 Ti and accompanying benchmarks, it looks like we can toss these charts in the trash can. These ChipHell rumor charts, which many people considered likely to be reliable, show the Titan X with a 10%-12% advantage over "GM200 cut"; in reality, Titan X only came out ahead by about 4%. It also attributes a much larger difference in power consumption between the two cards than is in fact the case.
Of course, it's possible that these charts were accurate at the time they were posted, and Nvidia adjusted the clock speeds at the last minute to match or beat Fiji. But we can't take anything in here at face value since what we can now check has been proven incorrect.
Of course, it's possible that these charts were accurate at the time they were posted, and Nvidia adjusted the clock speeds at the last minute to match or beat Fiji. But we can't take anything in here at face value since what we can now check has been proven incorrect.
Azix
Golden Member
- Apr 18, 2014
- 1,438
- 67
- 91
They should have released it together with this movie:
https://www.youtube.com/watch?v=bS5P_LAqiVg
Movie is awesome btw.
wow
They have their old cards to think about. including 295x2 cards.
@JDG1980
Those charts were from LAST year. As we've seen on the chip itself, it was produced in Q3 2014. Been sitting there because no competition from AMD, NV can milk 980s all the way to the bank.
Those charts were from LAST year. As we've seen on the chip itself, it was produced in Q3 2014. Been sitting there because no competition from AMD, NV can milk 980s all the way to the bank.
Azix
Golden Member
- Apr 18, 2014
- 1,438
- 67
- 91
honestly though, they could have just had that out at a higher price from back then if it was to milk the 980.
I'd understand keeping it back to launch and milk a titan
LTC8K6
Lifer
- Mar 10, 2004
- 28,520
- 1,576
- 126
Then AMD is really stupid, and I'd wonder about buying anything of theirs in the future.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
Reference 980TI boosts to 1202mhz per AT. Overclocking to 1477mhz resulted in "The gains from this overclock are a very consistent across all 5 of our sample games at 4K, with the average performance increase being 20%."
That means 980TI 100% -> 120% with 1477mhz overclock.
If Fiji XT is 106% stock and overclocks 15% -> 120-122%.
See, your comment that after-market 980Ti cards will destroy Fiji XT OC is not certain even if Fiji XT has 106% stock performance.
Where NV is very competitive are 2 key points:
1) 6GB of VRAM. Similar performance and similar overclocking and prices, I'd take a 6GB card.
2) Power usage in overclocked states is excellent for the 980Ti since it more or less overclocks on stock dynamic voltage. I just don't see AMD's Fiji XT overclocking 15-20% with such a minimal increase in power usage. NV's Maxwell is a remarkable overclocking architecture since it overclocks as well as the 7970 but with very little increase in power usage due to keeping voltages more or less stock (1.23V and lower).
As far as water cooling goes, to me that's a + not a -.
There are other factors of course such as more and more GW games. If we start bringing GW into the equation, 980Ti could easily be 25-50% faster than a Fiji XT.
But it's also hard to say how well Maxwell will age. WIll it be a repeat of Kepler, cuz that would be a disaster!
1.5 years from the time 780Ti launched, 980Ti is beating it by 91% in The Witcher 3, one of the biggest games of 2015. R9 290 > 780Ti. Yuck.
Kepler is also getting owned in GTA V. 980Ti beats 780Ti by 65%, R9 290 reference > 780Ti. Yuck.
That's 1 thing that would worry me about 980Ti vs. Fiji XT: Will NV stop optimizing the drivers for Maxwell in 18 months when Pascal comes out? I know for 100% fact AMD won't since AMD has HD7000-Fiji or 4.5 years of GCN architecture so they can't stop optimizing for a long time.
AMD never released performance numbers prematurely for HD4870, 4890, 6970, HD7970/7970Ghz or R9 290X and neither for any of their dual chip cards like 5970/6990/7990. Most of those cards were giving NV a run for the $.
Do you not remember 9700pro? It came out of nowhere and left NV for dead with 2 generations of failures - GeForce 4 and 5 got destroyed by 9700 and 9800 series. While I don't expect 9700Pro repeat again as that's once in a 20 year situation, just because AMD is quiet, doesn't mean their product stinks. It could or could not be the case.
Is this your first time experiencing a staggered GPU launch from 2 competitors?
How do you not remember that X800XT Platinum Edition beat the 6800Ultra despite launching later and ATI not making a peep about its performance prior to launch?
"With ATI's performance on par in older games and slightly ahead in newer games, the beefy power supply requirement, two slot solution, and sheer heat generated by NV40 may be too much for most people to take the NVIDIA plunge." ~ AT I guess you don't remember cuz let me guess you bought GeForce 6.
Also, after Samsung's S6 mopped the floor with iPhone 6 in almost all CPU/GPU and flash memory benchmarks, and smoked the iPhone 6 in the camera and battery departments, should Apple start releasing had data on iPhone 6S? I guess Apple has nothing to show since the are eerily quiet....
They should have released it together with this movie:
https://www.youtube.com/watch?v=bS5P_LAqiVg
Movie is awesome btw.
Epic 80s! They should have gotten David to release ads where he stars himself as Kung Fury to promote the Fiji FURY launch! Hitler would be wearing green clothes haha.
Ya and how many people online at TPU and other forums kept trolling how there is no way ChipHell would have all of these GPUs in their hands when NV had the entire Maxwell stack ready during 36-37th week of 2014, every card in fact from Titan X to a cut-down GM200.
I disagree. Those charts seem extremely accurate for early sample cards on early drivers. Also, we don't know how cut down that GM200 6GB was in their posession. Was it a 2688 shader card or the 2816 shader one? How do you know that NV didn't have a cut down GM200 and as yields improved they decided to go ahead and release a 2816 shader one? That little 4-5% extra increase gets us exactly to 4% of the Titan X from the 132.7% 980TI had. Think about it 2816 shader card and 980TI at the clocks below would end up at 139% on those ChipHell 4K charts which is just 4.5% behind the Titan X!
ChipHell's 980Ti boosted to 1192mhz and the retail version per AT review boosts to 1202mhz
ChipHell's Titan X boosted to 1179mhz and the retail version per AT review boosted to 1215mhz
As you can tell in different reviews the boost clock is slightly different. Sure, the 980Ti seems underperforming a little bit but it could have been a case of the drivers not optimized yet for a cut-down part, or the part was slightly more cut down. Either way, NV had 2816 shader 176 TMU, 96 ROP GM200 6GB chips 36th week of 2014 and KNOWING this released a 980 for $550. :sneaky: They essentially wanted to release a 980Ti as a "successor" to a 2-year-old 780 if you will and saw that AMD had nothing based on industry rumours and just went in pricing GTX560Ti successor at $550 following the already successful launch of $499 GTX680.
ChipHell charts are pretty close to reality
vs. reality
That's damn close considering these leaked like 6 months ago. The outliers are 960/285 but once we saw newer games come out that use . 2GB of VRAM at 4K, it wouldn't be surprising if their performance bombed more with games like Shadow of Mordor, GTA V, DAI, etc.
ChipHell had GTX970 SLI 8% faster than Titan X. With latest drivers, and 980Ti coming in 2-4% within Titan X, TPU concludes:
"NVIDIA set an MSRP of $650 for the GTX 980 Ti, which is reasonable and falls in line with the company's previous pricing strategy. There are other options at that price level, too. First, there is the GTX 970 SLI at $640 or so. It should offer around 8% more performance, but requires two PCIe slots and depends on multi-GPU driver support for optimum scaling, so I think I'll take the GTX 980 Ti over that."
ChipHell might have pawned every rumour site in the world. If AMD upped the clocks to 1.05Ghz and improved drivers since last year, I wouldn't be surprised if that 149% moves to 153-155% against the R9 290X at 4K. Of course all of this rests on the assumption a 4096 shader, 64 ROP, 256 TMU Fiji XT actually lives. I guess we will find out soon.
Last edited:
If AMD leaks performance numbers, the naysayers will claim they are desperate to rain on NV's parade.
sandorski
No Lifer
- Oct 10, 1999
- 70,874
- 6,411
- 126
AMD has usually been secretive about new releases. Way back before the release of the original Athlon little was known about it. When it finally released i was quite the shock.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
A lot of people on this forum have extremely short memories or they just act that way. For example, some of them claim that AMD should have launched Fiji cards a long time ago at prices as high as possible and if NV's 980Ti/Titan X beat those cards, well not a problem, AMD could have lowered the prices when NV launched. The irony of this argument is that AMD did exactly that with a 7970 $550 but when $499 680 came out, AMD dropped prices about $50-75 2.5 months after its original launch. This very forum and many others online wents nuts, absolutely nuts and hysterical how AMD flopped, how they overpriced the flagship part milking early adopters, blah blah blah.
Yet today, we hardly see the same reaction from the same people wrt to Titan X vs. 980Ti, yet NV just wiped out $350 of value from the Titan X in 2.5 months. Can you imagine if AMD did that? We wouldn't hear the end of it on these forums if AMD launched Fiji XT January 1, 2015 for $850 and then 980Ti tied it at $650 and AMD had to lower prices $200-250. :whiste:
With AMD having all the prices of NV's product stack, and possibly the rumoured GTX960Ti that might launch soon at $249, AMD is in a pole position to dictate price/perfomance at launch and do to Maxwell what NV did with Kepler's 660Ti/670/680 cards. AMD can undercut every single NV card by $50 or offer 5-10% more performance for the same price and NV's entire stack looks overpriced.
Prime targets will be 750Ti/960/980 because each of those cards is garbage when it comes to price/performance. With the right marketing launch strategy, those 3 cards can be made entirely irrelevant by new AMD cards. The question is does AMD have a new marketing plan in place to talk about how people buy Gaming graphics cards for games? If AMD delivers full HVEC 4K decoding/encoding hardware support in the budget range, 750/750Ti are seriously jeopardized. 960 is such a poor value, it should be easy to dethrone it with an unlocked 285. 980, not much needs to be said about it. A $499 3500 shader Fiji or a 10% faster R9 390X for $399 should make a $499 980 a bad buy.
Last edited:
AMD would be stupid to turn down contracts with Apple. It's free marketing (and some of the best marketing in the world) for AMD, not to mention guaranteed money. So what, AMD turns down a contract with Applejust so they can leak specs and possibly sway a handful of on-the-fence tech nerds?
Give me a break.- Status
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter