Geforce GTX 295 previews

[chizow]

Originally posted by: SickBeast

Originally posted by: Creig
When you stop to think about it, that's 1.93 billion transistors for the 4870X2 vs 2.8 billion transistors for the GTX295. That's a difference of very nearly one entire RV770 GPU! :Q

In other words, AMD's GPUs are still the most efficient. 3 4870s in crossfire would beat out this new NV card and anything else they have to offer.

You're not considering clockspeed and power draw, so no you can't make any direct claims about efficiency. The 4870s are clocked 20-25% higher than GT200 which results in higher relative temps and much higher TDP per transistor. The end result is RV770 is a poor overclocker (a 4870 is arguably an overclocked 4850) and even when overclocked, results in much smaller gain as its fewer "turns of the wheel" with less transistors.

 

[Keysplayr]

Originally posted by: SickBeast

Originally posted by: Wreckage

Originally posted by: SickBeast

It's disappointing because AMD was not even trying to produce a high-end GPU, and yet the 4870x2 is still within an earshot of a re-spin of NV's monolithic GPU.

The 4870x2 is not a high end GPU, it's 2 GPUs. 😕

Yeah, two midrange GPUs, and they're still within an earshot of two of NV's massive, re-spun, and monolithic chips.

You believe R770 was intended to be a midrange GPU only because "we" were told they were by AMD's engineers, according to Anand's most recent meeting with them. For all you know, AMD could have been trying like hell to beat out Nvidia's top end, and what they finally ended up with could only be deemed "mainstream/high end" as opposed to enthusiast. Then, it would be easy to say, "We never intended to compete for the top spot". This may or may not be true. But don't take it as gospel my friend. IMHO.

If AMD was not interested in enthusiast segment as they alluded to in that interview, they would never have released a 4870X2. Apparently, they did want to.

 

[chizow]

Originally posted by: keysplayr2003
If AMD was not interested in enthusiast segment as they alluded to in that interview, they would never have released a 4870X2. Apparently, they did want to.

Yep, beginning with the "little core" strategy and RV670, they've said their answer for the high-end would be multi-core. This is why Anand, Derek and various tech sites and forums have been getting on their case so heavily about promised features on the 4870X2 (shared memory, CF sideport) that never materialized and more recently, their horrible CF driver support.

 

[thilanliyan]

Originally posted by: chizow
You're not considering clockspeed and power draw, so no you can't make any direct claims about efficiency. The 4870s are clocked 20-25% higher than GT200 which results in higher relative temps and much higher TDP per transistor. The end result is RV770 is a poor overclocker (a 4870 is arguably an overclocked 4850) and even when overclocked, results in much smaller gain as its fewer "turns of the wheel" with less transistors.

I think by efficiency he means performance per transistor not performance per watt (ie. I think AT (correct me if I'm wrong) in one of their first RV770 reviews did a performance per transistor take and AMD was ahead.)

 

[thilanliyan]

Originally posted by: keysplayr2003
Do we have to upgrade? No. Is it prudent to upgrade to accurately show performance? I would say so. So, you might go a bit easier on us when you refer to "free" in the context of everything being "free" for us, when it clearly isn't.

It's still $500+ (and more than likely more if you get GTX 280 SLI or a similar config) that you don't have to spend.

 

[OCGuy]

Originally posted by: thilan29

Originally posted by: chizow
You're not considering clockspeed and power draw, so no you can't make any direct claims about efficiency. The 4870s are clocked 20-25% higher than GT200 which results in higher relative temps and much higher TDP per transistor. The end result is RV770 is a poor overclocker (a 4870 is arguably an overclocked 4850) and even when overclocked, results in much smaller gain as its fewer "turns of the wheel" with less transistors.

I think by efficiency he means performance per transistor not performance per watt (ie. I think AT (correct me if I'm wrong) in one of their first RV770 reviews did a performance per transistor take and AMD was ahead.)

Which .000001% of the world cares about. If it is a better card and can be priced for around the same as the competition's flagship, it is a win. However:

nV is going to have to do some major work looking ahead to the next round. Yes they made a better card with the shrink and sandwich, but they need to take an ATi like approach with more efficient chips. I cannot see this approach working 2 cycles from now for some reason. Then again, noone thought that GT200GX2 was possible, so we will see.

 

[chizow]

Originally posted by: thilan29

Originally posted by: chizow
You're not considering clockspeed and power draw, so no you can't make any direct claims about efficiency. The 4870s are clocked 20-25% higher than GT200 which results in higher relative temps and much higher TDP per transistor. The end result is RV770 is a poor overclocker (a 4870 is arguably an overclocked 4850) and even when overclocked, results in much smaller gain as its fewer "turns of the wheel" with less transistors.

I think by efficiency he means performance per transistor not performance per watt (ie. I think AT (correct me if I'm wrong) in one of their first RV770 reviews did a performance per transistor take and AMD was ahead.)

Yes they did, but they also failed to take clockspeed and power draw into consideration. 🙂

For example, if Nvidia cranked up voltages and clockspeeds as high as they could (max overclocked versions you see out there) and then you compared them, you'd see their performance and efficiency per transistor was very much the same.

With the 55nm die shrink its even more clear 4870 isn't as efficient as you'd think as the 285 is expected to come in at 189W, the GTX 260 presumably less than that (both less than 4870) and the 295 requiring less power than the 4870X2. All this despite the monolithic core size!

 

[nosfe]

why are you talking about watts when the others are talking about transistors? also, the P4 would also be as good as an Athlon, it just needs to crank up the frequency as high as it can.... see the problem?

 

[chizow]

Originally posted by: nosfe
why are you talking about watts when the others are talking about transistors? also, the P4 would also be as good as an Athlon, it just needs to crank up the frequency as high as it can.... see the problem?

Yes, that's exactly the point, but I'll give you a hint, the 4870 would be the P4. 😉

 

[Creig]

Originally posted by: chizow
For example, if Nvidia cranked up voltages and clockspeeds as high as they could (max overclocked versions you see out there) and then you compared them, you'd see their performance and efficiency per transistor was very much the same.

There's your problem right there. "IF" If is not reality. Reality is that the design and complexity of the GT200 obviously does not allow it to be clocked as high as the RV770, otherwise Nvidia would have already done so.

 

[Creig]

Originally posted by: chizow
Yes, that's exactly the point, but I'll give you a hint, the 4870 would be the P4. 😉

:roll:

 

[nosfe]

when talking about power consumption, yes, it's weird in a way, the 4870 with its smaller die and manufacturing process eats more power than the gtx 260. The thing is, with a smaller die comes lower price so it evens out a bit, you pay less upfront but a bit more afterwards because of the higher power consumtion. i don't think that nvidia would have asked so much for the gtx 260/280 if the die would have been smaller and they're paying because of it right now by loosing market share

 

[chizow]

Originally posted by: Creig
There's your problem right there. "IF" If is not reality. Reality is that the design and complexity of the GT200 obviously does not allow it to be clocked as high as the RV770, otherwise Nvidia would have already done so.

Reality?

No ifs needed....

Now compare those to ATI's highest "overclocked" models:

+50MHz on water........

+25MHz on "Golden Sample"...more like "Brown Sample"

Nvidia has always built OC'ing headroom into their parts, whether to extend the life of their parts or to allow card partners to produce significantly OC'd parts. ATI on the other hand has always released parts stretched to their limit at stock. So if you have an inclination to overclock or if you're interested in significantly overclocked factory parts, you would benefit from Nvidia's conservative stock clock speeds.

 

[nosfe]

yep, ati partners are quite conservative, i don't know why, i managed to overclock my 3850 from 670 to 730 with the nonstock cooler that it came with(and i was looking for the cheapest i could find), maybe i just got lucky

 

[Creig]

Originally posted by: chizow
Nvidia has always built OC'ing headroom into their parts, whether to extend the life of their parts or to allow card partners to produce significantly OC'd parts. ATI on the other hand has always released parts stretched to their limit at stock. So if you have an inclination to overclock or if you're interested in significantly overclocked factory parts, you would benefit from Nvidia's conservative stock clock speeds.

What, do you think AMD cards don't have any overclocking headroom? There are many people overclocking their stock 4850s to 700MHz, 800MHz and beyond. Some are hitting even higher than that, but I believe they involve volt mods. Just because current AMD partners don't sell highly overclocked cards doesn't mean they're not capable of it. Perhaps XFX can show the other AMD vendors how it's done now that they're going to be producing 48X0 series cards as well.

 

[chizow]

Originally posted by: Creig
What, do you think AMD cards don't have any overclocking headroom? There are many people overclocking their stock 4850s to 700MHz, 800MHz and beyond. Some are hitting even higher than that, but I believe they involve volt mods. Just because current AMD partners don't sell highly overclocked cards doesn't mean they're not capable of it. Perhaps XFX can show the other AMD vendors how it's done now that they're going to be producing 48X0 series cards as well.

I didn't say they didn't have any, I'm just saying they're historically not very good at it. Not only are they not very good at it, their board partners don't seem comfortable releasing any significantly overclocked variants or warrantying overclocking by the end-user. So again, while overclocking is not important to everyone, these are certainly luxuries provided by Nvidia partners and parts for those that are interested in OC'ing.

And again, voltmodding, Nvidia parts overclock better without any voltmods and they also realize additional gains with voltmods. We already know a 4850 can be voltmodded to hit higher clocks, its called a 4870. Anyways, we'll see how things pan out with XFX, hopefully they don't offer a different warranty for ATI parts as that would kinda negate the significance of their offerings.....

 

[Creig]

dp

 

[AmberClad]

Comments that have nothing to do with the topic at hand (particularly personal attacks and other comments sidetracking the discussion) have been deleted...

- AmberClad

 

[Keysplayr]

Originally posted by: Creig


"What, do you think AMD cards don't have any overclocking headroom? There are many people overclocking their stock 4850s to 700MHz, 800MHz and beyond. Some are hitting even higher than that, but I believe they involve volt mods. Just because current AMD partners don't sell highly overclocked cards doesn't mean they're not capable of it. Perhaps XFX can show the other AMD vendors how it's done now that they're going to be producing 48X0 series cards as well."

"There's your problem right there. "IF" If is not reality. Reality is that the design and complexity of the GT200 obviously does not allow it to be clocked as high as the RV770, otherwise Nvidia would have already done so."

I'm a little confused. If AMD partners were capable of highly overclocked cards, why wouldn't they sell them? I could understand the part about GT200 chips not being able to clock to R770, because GT200 has nothing in common with R770. Two different architectures and really shouldn't be compared clock to clock. Ever. I think they should only be compared by what each GPU can do in the end and that's pretty much it. Clocks should only be considered when comparing like architectures. 4850 -> 4870 would make sense.

 

[Creig]

Originally posted by: keysplayr2003
I'm a little confused. If AMD partners were capable of highly overclocked cards, why wouldn't they sell them? I could understand the part about GT200 chips not being able to clock to R770, because GT200 has nothing in common with R770. Two different architectures and really shouldn't be compared clock to clock. Ever. I think they should only be compared by what each GPU can do in the end and that's pretty much it. Clocks should only be considered when comparing like architectures. 4850 -> 4870 would make sense.

That's a good point. I'm not sure why AMD partners don't sell their cards as highly overclocked as some Nvidia AIBs do. There is definitely the potential there given how easily end users seem to get 700+ out of them. All I can think of is that the manufacturers currently putting out AMD cards have never really gotten into overclocking and don't wish to run the risk of increased returns. As I said, perhaps XFX can show the rest of the AMD manufacturers how to put out a well overclocked card. If enough enthusiasts turn to XFX and recommend them to friends/family/customers, perhaps the other AMD companies will be forced to take notice and develop their own overclocker friendly cards/warranties.

 

[ArchAngel777]

Originally posted by: Creig

Originally posted by: keysplayr2003
I'm a little confused. If AMD partners were capable of highly overclocked cards, why wouldn't they sell them? I could understand the part about GT200 chips not being able to clock to R770, because GT200 has nothing in common with R770. Two different architectures and really shouldn't be compared clock to clock. Ever. I think they should only be compared by what each GPU can do in the end and that's pretty much it. Clocks should only be considered when comparing like architectures. 4850 -> 4870 would make sense.

That's a good point. I'm not sure why AMD partners don't sell their cards as highly overclocked as some Nvidia AIBs do. There is definitely the potential there given how easily end users seem to get 700+ out of them. All I can think of is that the manufacturers currently putting out AMD cards have never really gotten into overclocking and don't wish to run the risk of increased returns. As I said, perhaps XFX can show the rest of the AMD manufacturers how to put out a well overclocked card. If enough enthusiasts turn to XFX and recommend them to friends/family/customers, perhaps the other AMD companies will be forced to take notice and develop their own overclocker friendly cards/warranties.

This is a plausable explanation. However, we can take MSI, a partner of both nVidia and AMD/ATI and this might give us a slight clue. The MSI 280GTX runs at 650 instead of 602. The MSI 4870 is at 780 instead of 750. This, in my opinion, is evidence that AMD/ATI do not have the headroom that nVidia currently does. I could be wrong, though. Perhaps AMD has policies in place with their partners to prevent a certain clock rate being exceeded... Be nice to have a good answer on this. However, if I recall correctly based on forum posts from users, I don't think the AMD cards can exceed 830ish without major volt mods and I believe there are many users surpassing 700 core with 280GTX

 

[Wreckage]

Originally posted by: Creig

There's your problem right there. "IF" If is not reality. Reality is that the design and complexity of the GT200 obviously does not allow it to be clocked as high as the RV770, otherwise Nvidia would have already done so.

Then it would be all loud and hot like the 48xx.

The GT200's have plenty of room for overclocking and based on reviews they are more than fast enough.

 

[evolucion8]

Originally posted by: Wreckage

Originally posted by: Creig

There's your problem right there. "IF" If is not reality. Reality is that the design and complexity of the GT200 obviously does not allow it to be clocked as high as the RV770, otherwise Nvidia would have already done so.

Then it would be all loud and hot like the 48xx.

The GT200's have plenty of room for overclocking and based on reviews they are more than fast enough.

You talk like if the RV770 power consumption were miles ahead of the GT200, the difference is insignificant, nVidia consumes less power when idle, and consumes slighly more in full load. Like someone stated, in a per transistor basis, the ATi architecture is more efficient because it has much less transistors and the performance difference is small, but in a power efficience matter, nVidia is doing great because it consumes slighly less power and it has a bigger die with an older manufacturing process with more transistors, even though the transition to 55n didn't brought too many differences, it still appreciable specially with the Heat Dissipation.

 

[ronnn]

Originally posted by: keysplayr2003

Originally posted by: ronnn
I agree with the cookie. If in a few handpicked games it can't win at high res\aa, claiming the performance crown will be tough. Likely things will change when the vapour turns into actual product.

Quote from final page of Guru3d's preview:

"Also despite the fact we were limited to testing a handful of games, we internally of course did run the majority of benchmarks with other games already. And the performance widespread is consistent and the card worked with any game we threw at it."

I don't think it'll be tough at all.

At what resolution and aa? I don't expect nvidia pr to find it tough, just disappointing after 8 months to even need pr.



When I post a notice that I've had to remove posts that stray into the area of personal jabs, that's a hint for everyone participating in the thread to not continue doing it. In case I wasn't explicit enough:

Lay off the personal attacks and keep the debate/discussion on the hardware.

- AmberClad (Video Moderator)

 

[nRollo]

Originally posted by: ronnn

Originally posted by: keysplayr2003

Originally posted by: ronnn
I agree with the cookie. If in a few handpicked games it can't win at high res\aa, claiming the performance crown will be tough. Likely things will change when the vapour turns into actual product.

Quote from final page of Guru3d's preview:

"Also despite the fact we were limited to testing a handful of games, we internally of course did run the majority of benchmarks with other games already. And the performance widespread is consistent and the card worked with any game we threw at it."

I don't think it'll be tough at all.

At what resolution and aa? I don't expect nvidia pr to find it tough, just disappointing after 8 months to even need pr.

QFP and reported.



In the future, do not contact individual moderators (i.e. me) if you think something needs moderator attention -- this is clearly stated in the forum guidelines:

12) Do not contact individual moderators. Please direct violations of these guidelines to moderator@anandtech.com, AnandTech Moderator by PM or post in Personal Forum Issues to request attention for something. Please be patient with moderation requests and inquiries. Users found in violation of this directive will be given only one warning and pointed in the right direction before further action is taken.

- AmberClad (Video Moderator)