Rys talks about GF100 in detail

Haha, so when his performance estimates are favorable and based on IRRELEVANT MADE UP CLOCKS it is a good article, but when vr-zone posts an article estimating the opposite based on REAL FACTUAL CLOCKS FROM AN ANNOUNCED PRODUCT it isn't. :rolleyes:

Wake me up when the third respin comes back from the Fab and clocks are finalized. Otherwise it could in theory be able to generate 1.21 gigawatts of energy to go back in time and it would mean nothing.

P.S: Are you enjoying bashing the people of this forum at beyond3d?
 
Until we see real benchmarks performance estimates mean nothing, absolutely nothing - an ati biased site could say the exact opposite of that estimation and you know what? means nothing.

I hope it's as faster than the 5870 and i also hope the shift towards gpgpu pays off in non gaming situations just as much, they're not on final silicon yet so it is always a guess and ati fanbois will point 1 way and nvidia the other so wait for benchmarks and then gloat.
 
All estimations aside, it was an informative article for someone like me who isn't up to date with the latest news going on like what Fermi really is. So thanks for sharing
 
You will be civil in this thread and keep comments directed at the article/OP itself or you will be banned.
 
If you've been following Fermi since it was announced, you'll know Nvidia didn't really talk about the specific graphics transistors in Fermi implementations. We're going to take a stab at that, though, using information gleaned from the whitepaper, bits teased from Nvidia engineers, and educated guesswork. Remember, however, that graphics transistor chatter does ultimately remain a guess until the real details are unveiled.
Click to expand...


Sigh, here I was excited.
 
martinmsj said:
Sigh, here I was excited.
Click to expand...

I clearly indicated that some portion of the article, contains "guesstimations", so that text portion shouldn't come as a surprise.

What's relevant is that Rys has lots of inside info and seeing him write this, surely is good indication of what Fermi will bring. Roughly twice the performance of a GTX 285, puts a single chip GeForce based on Fermi on the heals of the HD 5970 and sometimes beating it.
 
Silus said:
I clearly indicated that some portion of the article, contains "guesstimations", so that text portion shouldn't come as a surprise.

What's relevant is that Rys has lots of inside info and seeing him write this, surely is good indication of what Fermi will bring. Roughly twice the performance of a GTX 285, puts a single chip GeForce based on Fermi on the heals of the HD 5970 and sometimes beating it.
Click to expand...

i would love to know the price tag for this chip... if its close to 5970 in performance, we are looking at a $600 card..
 
cganesh75 said:
i would love to know the price tag for this chip... if its close to 5970 in performance, we are looking at a $600 card..
Click to expand...

Consider the estimated die size. GT200 was 576mm^2, and the GTX280 and was released with a $650MSRP. If GF100 is ~500mm^2, then the chip isn't going to be much less expensive based on die size alone. If you try to factor in TSMC's 40nm problems things look even worse for yields, but TSMC may have gotten that solved by now for all we know.

I think $600 is a fairly reasonable guess.
 
Wow I hope not. I wanna pick up 2 but god I don't feel like spending 1200 to do it. Was hoping for 400ish a piece. We'll see I guess.
 
Nvidia are dumb, but they aren't stupid. They will release this card based on price/preformance numbers. If it competes with the 5970, expect it to launch at 550 if not 500$. They know ATI will slash prices as soon as they launch so they have to set the bar low enough ATI will only match, not beat them in prices. However I think it will be much more along the lines of 10-20% faster than the 5870.

Assuming 10-20% faster than the 5870 and we set the stage for a 400$ GTX 380. That would push the price point of the 5870 to 350$ when avaliablility increases. There is no reason to lower the price if you can't keep units on the shelf. If the 380 breaches that magic +25% faster it is very possible we will see a 500$ price tag on the GTX 380 as it is the correct price/preformance vs a 400$ 5870.
 
The problem nvidia has is that because their card is so far behind the release of ati, it probably wont have to contend with just the 5870 - so even if it is faster (which i hope it is) it will be irrelevent.

Basically it will have to be quite a bit faster than the current 5870 to compete as there will no doubt be a refresh when fermi comes out.
 
A funny thing occurred to me reading the article this is the first time in a long while that nvidia does not have to worry about cutting into the price of their high end cards. The 295 cards are two 275 chips which could be dumped into lower priced cards and still sold. So they can set price high and not care about losing sales, since the high end belongs to AMD, anything 300 plus is fair game depending on the how fast it is.

There were two really interesting pieces in that article that may be really important. The logic units (SMD) are limited by how fast they can get new information to process. In the article they say nvidia told them they have two schedulers, which really would require they were linked in some way. Ryan said they don't believe they have two schedulers, kinda like people said that there could not be two schedulers, I mean memory controllers on the nforce chips. The things is what is the difference between a chip with two chips on it as each has it's own logic unit and a integrated unit with two virtual logic units? Because that is what two schedulers would look like from the outside. So chip would have to know what SMD were in use by each which could simply be written to shared memory.

From there some of the info he has just does not add up since the 768 shared memory suggests that they expect no more than six smd to be generating memory swaps or completed data refreshes per clock.

Some of the resulting numbers in the table don't make sense based on the info in the article. Then again it really is all speculation until we have a card in our hands :)
 
Buyers consider price/performance numbers. Manufacturers have to consider an additional factor: cost. If something is expensive to make, it will come with a high price tag regardless of performance.

I think this is what Kaldskryke is trying to say with the die size comparisons.
 
chklin said:
Buyers consider price/performance numbers. Manufacturers have to consider an additional factor: cost. If something is expensive to make, it will come with a high price tag regardless of performance.

I think this is what Kaldskryke is trying to say with the die size comparisons.
Click to expand...

It's a fallacy. Cost on a single product, especially the high end is irrelevant, maximizing profit is what happens. Nvidia can, and will sell the high end parts for break even or even a loss so long as it increase the sells of their lower end, OEM, and professional cards.

Even so, they can make a kit for under 100$ even with a 600mm^2 die. The die it's self is under 50$. Add 100$ for the partner and 100$ for the reseller you are talking a 300$ with ease.
 
Don't forget perceived value, if it costs more people think it has more value. Which has got to be the most idiotic thing I've ever heard but it is true for the prevailing masses.
 
I honestly could care less what that article says along with any other one.....nothing is going to be known until actual cards are out for review.......
 
Silus said:
I clearly indicated that some portion of the article, contains "guesstimations", so that text portion shouldn't come as a surprise.

What's relevant is that Rys has lots of inside info and seeing him write this, surely is good indication of what Fermi will bring. Roughly twice the performance of a GTX 285, puts a single chip GeForce based on Fermi on the heals of the HD 5970 and sometimes beating it.
Click to expand...

Nowhere did he say twice the performance. It's just often twice the raw stats.

But RV870 was twice the stats of RV770, yet was often only 30-40% faster for whatever reason.

I fully expect Fermi to be single chip faster than 5870...Nvidia's problems as usual deal with die size and cost for that performance. Basically it looks like a repeat of last generation.
 
vengence said:
It's a fallacy. Cost on a single product, especially the high end is irrelevant, maximizing profit is what happens. Nvidia can, and will sell the high end parts for break even or even a loss so long as it increase the sells of their lower end, OEM, and professional cards.

Even so, they can make a kit for under 100$ even with a 600mm^2 die. The die it's self is under 50$. Add 100$ for the partner and 100$ for the reseller you are talking a 300$ with ease.
Click to expand...

I dont know where you get you're figures but I'm sure they're all wildly wrong...

Die size is a very major part of costs and any reputable article on CPU/GPU pricing will tell you that...


I also severely doubt the partner and much less the reseller are getting anything close to $100 profit...

I mean I know in the case of a Xbox 360 or PS3, the retailer gets about $12-$18 profit on the 299 machine..

I think generally speaking dies cost around $100, but the actual costs would be incredibly complex and industry secrets. But the costs do scale with die size, it's that simple. You pay by the wafer, so the more chips you can get out of each wafer (aka, the smaller they are, removing yields from the equation) the less each chip costs. If I pay $1000 per wafer and get 100 chips from it, each chip cost $10, but if I only get 50 from it because they are twice as big, now each cost $20.
 
drakken said:
Don't forget perceived value, if it costs more people think it has more value. Which has got to be the most idiotic thing I've ever heard but it is true for the prevailing masses.
Click to expand...

Sure, if you ignore the fact that it's true in the vast majority of cases. Price is one of the most accurate indicators of quality.
 
Shark974 said:
Nowhere did he say twice the performance. It's just often twice the raw stats.

But RV870 was twice the stats of RV770, yet was often only 30-40% faster for whatever reason.

I fully expect Fermi to be single chip faster than 5870...Nvidia's problems as usual deal with die size and cost for that performance. Basically it looks like a repeat of last generation.
Click to expand...

He does actually:

Rys said:
The GF100's architecture means the SKU we've described (the GeForce GTX 380, possibly) comfortably outruns the GeForce GTX 285 in every way, to the point that (and we really generalize here, sorry) it should usually be at least twice as fast. Of course, you can engineer situations, usually in the middle of a frame, where the GF100 won't outpace the GT200 all that much, but in the main, it should be a solid improvement. The GF100 will outpace the Radeon HD 5870 as the top single-chip graphics product of all time, assuming AMD doesn't release anything else in the interim, between now and January. Look for the margins there to be a bit more slender, and we refer you to our Radeon HD 5870 review for the figures that'll let you imagine performance versus AMD's product.
Click to expand...

As for your RV870 vs RV770 comparison, NVIDIA's ALU increase scaling has been better than ATI's, so the assumption is not impossible.
 
Won't it be bandwidth limited? From what I can see, the memory bandwidth increase is only about 20% over the gt200.
 
Raap said:
Won't it be bandwidth limited? From what I can see, the memory bandwidth increase is only about 20% over the gt200.
Click to expand...

A bit more than that, but yes between 20-25%.

It's hard to say at this point. We really need benchmarks with actual game tests to know for sure and do some comparisons.
I see no indication of the GTX 285 ever being bandwidth starved, so at this point, I can only assume GF100 won't be either, but then again, these are two different architectures, so comparisons of this sort need to be done very vaguely.
 
You must log in or register to reply here.
Back
Top