I’m always typing tips :)
This GTX 650 Ti should be the last card of 1st 28nm generation.
Performance index must be like GTX 560 Ti.

Is this the longest desktop GPU generational roll-out from NV? It’s interesting that NV is finally completing the GTX650/650Ti/660/660Ti series roll-out; and with these cards they are targeting 7-9 months old HD7750/HD7770/7850/7870 GPUs.

What happens if HD8000 series launches in a quarter or so? AMD would obsolete the entire GTX600 series line. Either NV has spies that are aware about a major delay for HD8000 series (or its underwhelming performance??), or NV is about to walk into a gun fight with a knife. It sounds nice on paper but the clock is ticking since HD7970/7950 went on sale in January 2012. With the way NV is so casually releasing the rest of Kepler SKUs, it seems odd to me that they aren’t at all worried that in 4 months this whole GTX650/650Ti/660/660Ti business is going to be outdated.

nvidia fanboy here we go…..LOL

Well not the longest. GTX 580 – November 2010, GTX 560 SE March 2012 :)

i know it hurt u what GTX 670 did to your big ass hd 7970 and 7950 FLOP EDITIONS

if i recall correctly 7970 ghz its a killer for gtx 680 at higher resolution and the most games support nvidia special codes…that’s why nvidia won….and 7950 its a killer too but the problem was only the price at launch and you know why is that cause amd was the first to pull out his nex gen cards and i am no fanboy here i use nvidia too but nvidia dont decrease his price as amd does amd is cheaper and powerful..but has drivers issues and dont have a large support as nvidia does in games and FYI amd dont release their cards every 6 months they release after 11 months or 1 year

“AMD made theirs own trap. Theirs cards are made for middle class segment and that’s where they concentrated theirs effort.”

^ Not sure what you even mean by that. AMD has products from $95-$450 on the desktop with HD7970 GE being the fastest single-GPU this generation. HD7750-7870 had no competition from NV until this August. You may say that NV simply used 40nm Fermi chips but they lost market share doing that:
http://www.techpowerup.com/170575/Graphics-Shipments-in-Q2-Increased-2.5-Over-Last-Quarter-and-5.5-Over-Last-Year.html

Also, middle class segment is where the bulk of the $ is made. NV and AMD sell less than 10% of their SKU stack in the $400+ price segments. It’s not even about how AMD is doing vs. NV but what the gamers are getting. GTX500 line was terrible competition against the 28nm 7850/7870 cards. If someone bought a GTX560Ti / 448 / 570 instead of the HD7850/7870, they should be ready to throw those cards away soon or be prepared to seriously turn down settings as next generation games will use > 1.28GB of VRAM. You can overclock 7850 to GTX580 speeds (while still using way less power), but you cannot add extra VRAM to a GTX560 448 core / 570.

Finally, what you didn’t take into account is GTX650-660Ti line is competing against 1st generation 28nm parts from AMD. If AMD gets another head start with HD8000 series and delivers HD8850/8870 that are nearly as fast as GTX670, NV will need to drop prices and obviously there will be more advanced features on HD8000 cards as well. I am hoping NV launches very close to HD8000 series so we get good value and price pressure on both sides right away, not like this generation where NV was half asleep half the time.

I have no idea how GTX780 will perform vs. HD8970 but looking at how NV was unable to win this round, not sure where you are getting information that 8xxx will somehow be a disappointment (Unless you have some data/link/inside sources to back this up? or just your gut feeling)?

Some of us also play at 1600P and HD7970 @ 1050 mhz is already leading solidly in that resolution by 10-12% against the 680. That means NV needs to make up this 10-12% deficit and then match the % increase that 8970 has over 7970 GE and then add another 15-20% on top to really dominate the 8970. What this means in practice for 1440/1600P gamers is that IF for example HD8970 is 25% faster than HD7970 GE, GTX780 would need to be 55-58% faster than GTX680 to have a 15% lead over the 8970 at high resolutions. NV has to make a much larger performance improvement next round since they are behind in performance, especially at higher 1440/1600P resolutions.

His bias is blatantly obvious when he said NV makes more reliable cards. I mean after bumpgate with failed GF8/9 series and blown GTX570/590 cards from overclocking, even NV has issues with quality control from time to time. I’ve never had an AMD or NV card fail on me and I use them @ full load nearly 24/7 for the entire span of ownership but making a blanket statement like that without proof is just trolling.

U r right Nvidia do not cut prices on products but also all those products that has been declared a winner do not decrease their prices you can see it on other products…..i5 2500K is still sold for $200 AS IT WAS 2 YEARS BACK But AMD 8 core bulldozer starts at $300 and will end at $100… amd has reduced prices on GPUZ when nvidia have launched a better card at a lower price….if you r talking budget cards yes HD 7750 is awinner at that segment but $229 GTX 660 or $209 HD 7850 where GTX 660 is much better card in all aspects for just $20 more….and with all the features its up to you …..

AMD released theirs (really high overclocable) cards was 1.2 V. Some AMD cards have 1.3 V on load in 3D applications. Lets look at data I found during one of my researches:

Bad long term voltage for 45 nm CPU is 1.3 V. (Killing voltage should be 1.35-1.375 but it depends on EACH INDIVIDUAL CHIP.) Reasonable long term voltage for overclocking is 1.25 – 1.275. Voltage for reasonable normal use is 1.175. (But a lot of 45 nm chips can go down as low as 1.1.) Now the Gainward Phantom card I returned month ago, had 1.175 V at full power, 0.950 V in cinema mode. Basically AMD changed voltage three times after release, and a lot of AMD cards are distributed with voltage that’s lethal. NVidia isn’t much better in this area, however they try to force manufacturers to block voltage above certain level. Current cards are basically ticking bombs which would probably go for about 2 years, and then boom.

Then there is also this problem: “While I can’t provide a definitive explanation, I suspect that power consumption of AMD’s Pitcairn GPU varies wildly between individual batches.” http://www.techpowerup.com/reviews/MSI/HD_7870_HAWK/33.html

Well the current mess is a result of: 1. Intel using CPU boost, as if they didn’t know what it can cause in long term. 2. Manufacturers making OC, I heard some people who were claiming it’s great they can buy OC card and adding imagine these users (not them) would need to do it themselves. Thus AMD, and NVidia are releasing “OC” chips as well, automatic overclocking included. Obviously such chip don’t have ANY overclocking headroom and eat more electricity in normal situation. Nobody give a damn about GT 640/GTX 650, for this reason is GT 640 an incredibly power efficient chip.

Bit coin mining is also interesting problem. NVidia sucks in bitcoin mining, but AMD has only about 2MH/Watt on 7979, and funnily undervolted 5970 has 2MH/Watt, 7970 2.01 – 2.4 MH/Watt. That’s not that large improvement. There were specialized solutions with 10 MH/Watt, now they are preparing devices with efficiency 50-500 MH/Watt. I guess the era of using GFX cards as a specialized bitcoin mining equipment ended.

No HD7950/7970 card uses 1.3V at load. That’s just made up facts. You can go out right now and buy a Sapphire Dual-X 1Ghz 7970 card for $380 on Newegg and it will overclock to 1100mhz at 1.175V.

Also, AMD warranties GPU voltage up to 1.25V and you don’t need such high voltage to reach 1100-1150mhz clock speeds on HD7870/7950/7970 cards in the first place.

What you missed in your bitcoin mining calculation is that you can use the HD7000 series card for games and bitcoin mining at night. Thus, as long as bitcoin mining is profitable, it doesn’t matter at all if HD7970 is worse at it than HD5970 is. It just means you can now get a $800 of HD7970s and they pay for themselves every night you go to sleep. Next generation you have saved up enough $, sell the 7970s and get 8970s with the $ 7970s earned over those 12 months. It’s brilliant.

I skipped one paragraph with comparison of 28 nm voltages and calculation of safe voltage. Obviously 28 nm and 22 nm chips can’t take as much voltage as 45 nm, so when voltage for 45 nm overclocking is above 1.2 V, 28 nm should be at most 1.175 V, probably a bit less. (They’d need to do some tests, create curve, and calculate the approximation of voltage from a leakage estimation. Which would be valid until TMC would improve 28 nm process, then they would need to do the same research again.) How many AMD cards are running on 1.050 V?

AMD can claim warranties like that as they want, but the trouble is they don’t know consequences of high voltages on theirs cards because they didn’t did test for long enough, this’s the same problem as automatic overclocking on Intel’s chips. It’s working, so marketing department pushed for it… The trouble is with durability and some other problems it causes.

As for doing bitcoin mining overnight:
Well ASICs are arriving. WIth introduction of ASICs, the AMD position would be in comparison to ASICs similar to NVidia position vs AMD when compared in bitcoin mining. Of course this poor NVidia performance is inexcusable, if it’s caused by not having 50 million transistors on correct place of die, they should be spanked. But, profits from bitcoin mining would move elsewhere. Of course, ASICs entry is necessarily for bitcoin reliability, it solves problem of a government, like US, willing to attack bitcoin when theirs politicians would be inventing yet another Iraq like scam. ASICs would create natural technical barriers, and cause certain types of attacks to be really expensive even for governments.

Now a lot of people have AMD cards for bitcoin mining, they would want to sell them to get money for more efficient, and competitive devices. The result would be drop in 7970 prices. Can AMD afford that?