NVIDIA Engaged in PsyOps Against HD 7970: Chinese Forum

[cadaveca]

Originally Posted by claylomax

Who's Dave?

What rock did you crawl out from? Seriously though, I've been TPU's motherboard reviewer(and now ram) for a little over a year now.



<--------


Nice to see my name color matches my Avatar, too.

[ensabrenoir]

Originally Posted by cadaveca

What rock did you crawl out from? Seriously though, I've been TPU's motherboard reviewer(and now ram) for a little over a year now

<--------
Alw
Nice to see my name color matches my Avatar, too.

Always wondered how u pronounce that..... I see it as ca dave ca

Ca ca is well u know....and Dave is in the middle of it
Dont mean any disrespect or anything....just always wondered

[Benetanegia]

Originally Posted by kajson

The Intel comparison only works if you're saying Nvidia has a totally superior product, and the profit Nvidia is turning over compared to Intel isn't really in the same ballpark healthwise.

Most rumors do say that Kepler will be better than AMD's offerings. And no I'm not talking about rumors coming from Chiphell. The general idea is that performance class Kepler matches/beats HD7970 and a mid-range/performance class chip would never sell for more than $300. $300 is already a stretch considering the card that will replace is now selling for $200.

My own crystal ball says so, based on Fermi vs GCN, and considering the posibility that Nvidia improved on what Fermi was a "failure", instead of just simply assuming Kepler will be even worse.

I stand by my point anyway that it wouldn't be the first time that Nvidia does something like that. They did it with the 8800 GT and it worked wonders for them and they did it with GTX460 and it worked very well too. Both were arguably bigger chips than GK104 will (or could) be, and unlike now (potentially) also much bigger than AMD counterparts, so the profit margin is potentially much bigger this time around. They will also have the high-end chip to enjoy the great high-end margins, while forcing AMD to very low margins, by fighting their high-end with their mid-range/performance chip, that is potentially smaller and maybe 256 bits. Time will tell of course, but one thing is 100% sure, the concept that Nvidia cannot release a better value card is simply wrong. They they did it plenty of times before and they could do it now. In fact, in 5 years Nvidia never had a better oportunity than now. AMD finally entering the GPGPU scene after 5 years of Nvidia doing so could only have this consequence. I don't know if this advantage will materialize or not, but common sense says that it will.

[thebluebumblebee]

I find this intriguing, because most companies today put their employees through some form of "Service Excellence" and one of the basic things that you are taught is to "under promise and over deliver". But, for some reason, those who sell CPU's and GPU's have been allowed to do just the opposite and we allow them to get away with it. Intel is saying that Ivy Bridge is supposed to be 100% faster then Sandy bridge. Anyone believe them? Wasn't SB's performance also lower than we were told? (was it supposed to be a 50% increase but ended up more like barely 20%?) Don't even need to mention AMD's flub with Phenom.

[HumanSmoke]

Originally Posted by thebluebumblebee

Intel is saying that Ivy Bridge is supposed to be 100% faster then Sandy bridge.

Sorry...
That's bullshit.
You won't find any Intel literature claiming any such thing (although I'm open to being proven wrong). As far as I'm aware this is Intel's comparison from the Ivy Bridge roadmap (with iGP performance percentage significantly higher...but then if you're starting from such a low baseline numbers...)

Originally Posted by thebluebumblebee

Anyone believe them?

Believe them? I'm having a hard enough job believing you.

Originally Posted by thebluebumblebee

Wasn't SB's performance also lower than we were told? (was it supposed to be a 50% increase but ended up more like barely 20%?)

Nope....But yeah, it's a real pig. Performance like that should be damn easy to beat

Originally Posted by thebluebumblebee

Don't even need to mention AMD's flub with Phenom.

You probably don't need to go that far back if you're looking for AMD CPU hyperbole

[cadaveca]

Originally Posted by ensabrenoir

Always wondered how u pronounce that..... I see it as ca dave ca

Ca ca is well u know....and Dave is in the middle of it
Dont mean any disrespect or anything....just always wondered

Yes, Dave in the middle of some caca.

CA = Canada though. There is a story behind that, of course, but you'll ahve to pop into the TPU BF3 Teamspeak to get that one outta me

Originally Posted by thebluebumblebee

I find this intriguing, because most companies today put their employees through some form of "Service Excellence" and one of the basic things that you are taught is to "under promise and over deliver". But, for some reason, those who sell CPU's and GPU's have been allowed to do just the opposite and we allow them to get away with it. Intel is saying that Ivy Bridge is supposed to be 100% faster then Sandy bridge. Anyone believe them? Wasn't SB's performance also lower than we were told? (was it supposed to be a 50% increase but ended up more like barely 20%?) Don't even need to mention AMD's flub with Phenom.

So, of course, I've been really busy the past couple of weeks. So to come to the forum and fimd this topic is surprising..I'm kinda outta the loop right now.

Anyway, my first thought was exactly what you just posted. AMD spins some stuff about BD, and everyone forgot?

I kinda see it as part of my job here @ TPU to dispell any rumours or such, by providing reviews that show the truth, with results that anyone can duplicate. I'm lucky in having the oppotunity to investigate these things myself in this way, but hte one thing I have learned in the past year is that this is marketing, and marketing may start off based on projected targets, and those targets get missed. BD, for example. While it was poorly executed, in my opinion, the marketing staff @ AMD had some fantastic ideas...just didn't execute them well.

If I was Nvidia, I'd want to hop on that, and push AMD's buttons, so to speak, reminding potential customers that AMD has in the recent past, failed to deliver.

So now, I see we ahve a thread here about how people feel about AMD delivery of the 7970, and similar topics on other forums...wouldn't surprise me one bit if it was all done at Nvidia's behest, and I find no fault in them doing so, either. It's marketing, and anything you can do when providing marketing to spin stuff to your favor is gonna happen.

[seronx]

Pift 7970, it doesn't have a chance

[DigitalUK]

a card that comes out at least 6 months after the competition had better be faster.

[buggalugs]

Im not sure if its true or not, it wouldnt surprise me, I know Apple did stuff like that for years.

Whenever there was a negative news article about Microsoft, say the latest virus that was causing havoc, the first comments in the comment section were Apple guys saying how good Apple products are and that people should buy a Mac.

Apple are the masters of stealth advertising

[Nihilus]

Originally Posted by thebluebumblebee

I find this intriguing, because most companies today put their employees through some form of "Service Excellence" and one of the basic things that you are taught is to "under promise and over deliver". But, for some reason, those who sell CPU's and GPU's have been allowed to do just the opposite and we allow them to get away with it. Intel is saying that Ivy Bridge is supposed to be 100% faster then Sandy bridge. Anyone believe them? Wasn't SB's performance also lower than we were told? (was it supposed to be a 50% increase but ended up more like barely 20%?) Don't even need to mention AMD's flub with Phenom.

Very well said, bumblebee. I think the other major culprits are cell phone companies and internet service providers. These "digital age" companies provide nice products but customer service went to crap. Oh, and everyone needs to stop putting rumors on a pedestal. That's everyone's fault. Absolutely zero validity in today's sources. COLD HARD facts are the only thing that matter now. You don't see Dodge or Chevy "leaking rumors" about an "F-150 killer". If they do nobody cares, and they will wait for road tests.

[Nihilus]

Originally Posted by seronx

Pift 7970, it doesn't have a chance

Originally Posted by DigitalUK

a card that comes out at least 6 months after the competition had better be faster.

What does this have to do with marketing? TPU has many hardware stories for this kind of flag-waving. You two "bros" remind me of when Sony had hired people for "Guerrilla Marketing" saying how AWESOME the PSP was.

[THE_EGG]

Originally Posted by R_1

This round in GPU wars ain't cheap for end-users. AMD pricing is a lot like Nvidia-s, even higher. That is not MSRP $249-$349 for HD4870/4850 , or $259-$379 for HD5850/5870, or $299-$369 for HD6970/6950. We are facing $???-$550 MSRP for HD7950/7970, even worse $650-$750 in Europe. That is over 200% higher price for Europe, compared to previous AMD generation and lot worse, as we accounted $199 HD4850 vs. $650 HD7950. So let's see what GK104 can do. Patience can pay off this time.

lol its $750-$850 in Australia while the gtx 580 ranges between $480-$600. GTX 580's at my local comp stores are nearly always sold out since the 7970 came out.

[seronx]

Originally Posted by Nihilus

You two "bros" remind me of when Sony had hired people for "Guerrilla Marketing" saying how AWESOME the PSP was.

Any case the
512 CUDA part will be clocked beyond 900MHz
1024 CUDA part will be around 500-700MHz

Both will kill the 7970, I has done the maths

[THE_EGG]

Originally Posted by seronx

Any case the
512 CUDA part will be clocked beyond 900MHz
1024 CUDA part will be around 500-700MHz

Both will kill the 7970, I has done the maths

hmm that 512 CUDA part sounds disappointing considering I've overclocked my 580 to 905 (925 for benching). All I can say is that it had better be an awesome overclocker like the 7970.

[qubit]

Originally Posted by seronx

Any case the
512 CUDA part will be clocked beyond 900MHz
1024 CUDA part will be around 500-700MHz

Both will kill the 7970, I has done the maths

Yes, it might "kill" the 7970, but then again it might not. We just don't know until the official benchmarks. I love the certainty that you claim it will, by "doing the maths", lol.

[THE_EGG]

Originally Posted by qubit

Yes, it might "kill" the 7970, but then again it might not. We just don't know until the official benchmarks. I love the certainty that you claim it will, by "doing the maths", lol.

+1 here, BD certainly looked fantastic on paper but in real life it was an epic fail so there is no idea in knowing performance until official benchmarks and results are released.

[alwayssts]

Sure, it's rumor and speculation until we see real products and results, but when multiple independent sources whom have proven to be correct in the past all say the same thing it lends credence to the distinct possibility of there being some truth...especially when it makes sense given nVIDIA's regular formula and AMD made some odd decisions...which nvidia hinted at and I will point out. Does that make us trying to infer schills? Why? Because although nVIDIA has a pretty terrible engineering rep they may have made a good decision?

For instance:

nVIDIA's mid-range chip has been around the same performance/clock of AMD's high-end since G92. Given architectural differences, the main discernible difference in performance can be boiled down to clockspeed and market placement. There was never going to be a 2GB 880mhz SKU for GF114 because it would have been over the mid-range 225w and hurt sales of GF110. I'm sure there is math thrown into the mix figuring in yields with that high of a clock on that size chip and that played a role in their chip stack decisions. Perhaps that is why GF114 was abundant and 6970 not-so-much.

AMD generally clocks their parts very high compared to nvidia on a given process which allows this overlap to occur. On the 7970 SKU they were clearly conservative given it's high-end market placement yet small die size/minimalist design and that is SUPER FREAKING KEY TO THIS CONVERSATION. Look at power consumption in W1z's review and where it lands compared to nVIDIA's last gen (or even 6970). Between a 560ti and (6970/)570. That is telling, if not foreboding. Tahiti is a good chip; The 7970 would be a great high-end SKU if clocked higher on the core. As it sits it is meddling between where the mid-range and high-end should be, yet is built for enthusiasts. There's overclocking, but that is not how reviews compare products...they use stock skus.

Look at performance of 7970 at 1920x1200 or 1920x1080 versus a slew of other cards...or the leaked 7950 results. At those resolutions 7970 is not the most efficient, it is clearly aimed higher-up the foodchain...this is backed up by the 384-bit bus and 3GB of memory that do squat at 1920. 7950 is fairly efficient at 1920, but a little under-powered...yet is still built with above 1920 in mind (384-bit, 3gb)...WTF?

I simply ask myself:

Is it feasible nvidia will create a chip similar in size to Tahiti for the mid-range?
Yes.

Is it feasible it will be more efficient at 1920?
Yes, if between 7950-7970 architecturally. This makes sense as all signs point to nvidia releasing another chip (or even sku using the same chip) for higher resolutions.

Will a 256-bit bus/2gb hurt them in this market?
No, unless the mem clockspeed is super low in proportion to the core.

Is it likely slower at stock than 7970 and perhaps 7950?
Probably; if clockspeeds are low for one of the myriad of reasons (keep the price/power down, nvidia's memory controller still sucks and/or they use slower/low-power gddr5, it corresponds with where they want it placed in their own product stack or versus the competition etc). The point is that there is room for a better design than Tahiti aimed at the largest possible market. That, and there is always overclocking.

I personally don't write about these things to root for either camp, even if my heart as a geek lies with ATi. I just find it all interesting to try to figure out, and maybe some of my ramblings help someone else learn something or make a better decision for their needs. I think a lot of times people like me come across like we're making zealous assumptions to further a fanboy agenda (and some are, or just being hyperbolic) when in reality we're just like anyone else interested in this stuff, just a bit more long-winded or in some cases lacking the ability to communicate rationally. I think this may be the case at CHH...but who knows.

[seronx]

Originally Posted by THE_EGG

hmm that 512 CUDA part sounds disappointing considering I've overclocked my 580 to 905 (925 for benching). All I can say is that it had better be an awesome overclocker like the 7970.

Beyond 900MHz

It can be anywhere in 950MHz(200W TDP) to 1150MHz(250W TDP)

• 512 -> 950-1150MHz
• 768 -> 750-900MHz
• 1024 -> 550-700MHz

200W -> 250W on each

Originally Posted by qubit

Yes, it might "kill" the 7970, but then again it might not. We just don't know until the official benchmarks. I love the certainty that you claim it will, by "doing the maths", lol.

It will 7970 is a minor improvement on Fermi...so little Kepler will just roll right over it like a steamroller

[HalfAHertz]

Originally Posted by seronx

Beyond 900MHz

It can be anywhere in 950MHz(200W TDP) to 1150MHz(250W TDP)

• 512 -> 950-1150MHz
• 768 -> 750-900MHz
• 1024 -> 550-700MHz

200W -> 250W on each



It will 7970 is a minor improvement on Fermi...so little Kepler will just roll right over it like a steamroller

Here we go again. You did the exact same number pulling from one's behind before the bulldozer launch and we all know how that turned out. Please do us and yourself a favor and keep your crazy math skillZ to yourself.

[qubit]

Originally Posted by seronx

It will 7970 is a minor improvement on Fermi...so little Kepler will just roll right over it like a steamroller

Ya know, I have confidence that nvidia's next gen will be significantly faster than the 7970 and my GTX 580 is excellent. However, there's no way you can say with any certainty that this will be the case.

Get us some leaked benchmarks from a reputable source and we'll start to take you seriously. Yeah, this I'd really like to see.

[seronx]

Originally Posted by HalfAHertz

Here we go again. You did the exact same number pulling from one's behind before the bulldozer launch and we all know how that turned out. Please do us and yourself a favor and keep your crazy math skillZ to yourself.

And, I was right Bulldozer is the highest performing part on the market

Originally Posted by qubit

Ya know, I have confidence that nvidia's next gen will be significantly faster than the 7970 and my GTX 580 is excellent. However, there's no way you can say with any certainty that this will be the case.

ALUs x Ops x Clockrate

All you need to know

You make it sound like every architecture known to mankind is some magical unknown unfathomable device and I can tell you there isn’t anything fundamentally “magic” about processor design

[qubit]

Originally Posted by seronx

And, I was right Bulldozer is the highest performing part on the market



ALUs x Ops x Clockrate

All you need to know

You make it sound like every architecture known to mankind is some magical unknown unfathomable device and I can tell you there isn’t anything fundamentally “magic” about processor design

Bulldozer is indeed the highest performing part - in your mind.

That ALUs x Ops x Clockrate thing is BS. The efficiencies of the design make a huge difference. That's why it takes so much money, time and effort to bring a new design to market.

[Nihilus]

Originally Posted by seronx

And, I was right Bulldozer is the highest performing part on the market

ALUs x Ops x Clockrate

All you need to know

You make it sound like every architecture known to mankind is some magical unknown unfathomable device

Nothing is perfectly linear, there are always bottlenecks and unforeseen surprises - good or bad. You were kidding about BD, right? Please remember these tech companies are massive. So even 100 "independent sources" is just a small chunk of there marketing department. They are doing it with next-gen consoles too.

[seronx]

Originally Posted by qubit

Bulldozer is indeed the highest performing part - in your mind.

You just are proving you have no idea what the heck you are talking about

Originally Posted by qubit

That ALUs x Ops x Clockrate thing is BS.

It isn't BS.

Originally Posted by Nihilus

Nothing is perfectly linear, there are always bottlenecks and unforeseen surprises - good or bad. You were kidding about BD, right?

No surprises on this run and no I am not

People are referencing Bulldozer as a whole and Bulldozer as a whole is rather very competitive

[EarthDog]

Originally Posted by Damn_Smooth

They really need to put up or shut up. It's not like they're slowing sales of the constantly sold out 7970 anyway.

The card is great, but its extremely limited availability is a paper launch to me. Even though the 7950 was delayed for more availability, I would expect it to be limited as well. Not as bad as the 7970, but limited.




This place cracks me up.