ATI Radeon HD 4000 Series discussion

Quote:

---

GTX 280 = X4098
GTX 260 = X3782

---

if these are true then they might actually be worth the pricetag.

what is the source of the vantage scores?

Yeah I don't think people realize that the card is nearly 2x the score of the 8800 Ultra at those settings, where the 8800Ultra still has the edge on the G92 cards (besides the 9800GX2). And 3dMark always seems to be optimized well for multi-GPU setups, so 9800GX2 will score higher than real-world game performance where they're at the mercy of multi-GPU driver optimizations.

Quote:

---

Originally Posted by adamsleath

if these are true then they might actually be worth the pricetag.

what is the source of the vantage scores?

---

Problem is, the GTX 260, being only 0.8-0.85 of a 280, should not score that high. 3200-3400 at best I reckon.

There's some problems with the source! Ahem...

Quote:

---

Originally Posted by adamsleath

if these are true then they might actually be worth the pricetag.

what is the source of the vantage scores?

---

http://gathering.tweakers.net/forum/...ssage/30144631
CJ :D

sprechen zie deutsche?

Quote:

---

Originally Posted by adamsleath

sprechen zie deutsche?

---

Dutch man.

ok.

Quote:

---

Originally Posted by adamsleath

ok.

---

And ATI numbers are fake.
http://www.forumdeluxx.de/forum/show...postcount=1378
:(

Quote:

---

Originally Posted by adamsleath

ok.

---

yeah man, watch it buddy.

Quote:

---

Originally Posted by ownage

And ATI numbers are fake.
http://www.forumdeluxx.de/forum/show...postcount=1378
:(

---

Hmm... I hope the real ones are higher by a fair amount. If ATI fails me again I am just going to get a GTX280.

Quote:

---

Originally Posted by 003

Hmm... I hope the real ones are higher by a fair amount. If ATI fails me again I am just going to get a GTX280.

---

I doubt they will be any higher. And they don't need to be at half the price of GTX 280.

Quote:

---

Originally Posted by ownage

Why not? The HD4870 score seems more then logic. HD4870 seems to be 80% faster then the HD3870. It was rumoured the HD4870 would be 50% faster in overall performance, but the HD4870 has far better AA performance, so 80% doesn't seems wrong to me.

---

Yeah I realize that, but the chart initially showed the 3870=4850 and something like x1950 (which I quoted) and that's what I saying, way too low, should be 50%+, 80% sounds perfect to me actually

Quote:

---

Originally Posted by Helmore

Oh and AliG it says the 3870 scores 1440 while the 4870 does 2595.
Oh well, I won't make any conclusions about their performance until the cards are officially released.

---

As I said, look at what I quoted, the poster fixed it

Quote:

---

Originally Posted by Scimitar

Remember that the shader clock on the 4870 is rumored to be 1,050 MHz (200 MHz over the core clock). When you take this into account, the 4870 has (in theory) twice the shader power of the 3870, not 50% more.

---

Once again, I understand this

Quote:

---

Originally Posted by 003

Hmm... I hope the real ones are higher by a fair amount. If ATI fails me again I am just going to get a GTX280.

---

Ever heard about R700? :rolleyes:

Quote:

---

Originally Posted by Calmatory

Ever heard about R700? :rolleyes:

---

It seems more and more people are trying to compare 4870 to GTX280...It's like they never heard of dual GPU card called 3870X2 :shrug:.Similarly R700 will be introduced and I really expect it to perform on par or slightly worse than GTX280.

Quote:

---

Originally Posted by informal

It seems more and more people are trying to compare 4870 to GTX280...It's like they never heard of dual GPU card called 3870X2 :shrug:.Similarly R700 will be introduced and I really expect it to perform on par or slightly worse than GTX280.

---

Problem is (and I'm sure not everyone will agree) is that ATI is still a product cycle behind, all due to R600, and arguably, started earlier then that, with the 1800XT, and they haven't caught up yet. In August when R700 launches, nVidia will just release their dual chip card. If AMD would have a card that performed as well as the GT280, then they would sell it for a much higher price. Lower price just means that it can't compete performance wise.

Even with nvidia giving them time with the whole 9800GTX debacle, they were just laughing and perfecting G200.

When ATI's new architecture launches, nVidia will be there with their new card, or G200 on a smaller process, whatever is necessary to stay on top. Unless nVidia fails with a chip or process, like the FX5800, I think they'll stay on top, they've got the time and resources to do so. AMD doesn't have that necessarily. Or if AMD would drop a bombshell next year with a new architecture, the latter being more likely. But like we've seen before, because nVidia has got a cycle ahead, they can test, and use the smaller process on their midrange cards first, and then doing it to the big guns. Perfection.

Quote:

---

Originally Posted by Tim

In August when R700 launches, nVidia will just release their dual chip card.

---

GT200 will not be on a dual chip until a die shrink

They couldn't do it with G80 and G80 was less hot and less power-intensive as GT200 will be. What makes you think they will suddenly have a dual card :rolleyes:

Quote:

---

Originally Posted by Tim

Problem is (and I'm sure not everyone will agree) is that ATI is still a product cycle behind, all due to R600, and arguably, started earlier then that, with the 1800XT, and they haven't caught up yet. In August when R700 launches, nVidia will just release their dual chip card. If AMD would have a card that performed as well as the GT280, then they would sell it for a much higher price. Lower price just means that it can't compete performance wise.

Even with nvidia giving them time with the whole 9800GTX debacle, they were just laughing and perfecting G200.

When ATI's new architecture launches, nVidia will be there with their new card, or G200 on a smaller process, whatever is necessary to stay on top. Unless nVidia fails with a chip or process, like the FX5800, I think they'll stay on top, they've got the time and resources to do so. AMD doesn't have that necessarily. Or if AMD would drop a bombshell next year with a new architecture, the latter being more likely. But like we've seen before, because nVidia has got a cycle ahead, they can test, and use the smaller process on their midrange cards first, and then doing it to the big guns. Perfection.

---

Technically, Nvidia's 9900GTX WILL be a current gen card, seeing as it will not have a GTX 2xx based card in that range for a little time to come.

In addition, a single 4870X2 w/ GDDR5 should cost around $599 as well and should be provide a significant boost over the 3870X2. (960 Shaders and 64 TMUs anyone ;) )

Perkam

Quote:

---

Originally Posted by zerazax

GT200 will not be on a dual chip until a die shrink

They couldn't do it with G80 and G80 was less hot and less power-intensive as GT200 will be. What makes you think they will suddenly have a dual card :rolleyes:

---

I doubt that nVidia will make a dual-chip GT200 based card. Even with a die-shrink, a single die is still going to be a space heater.

IMO, with die-shirnk nVidia are going to raise the clocks of GTX 280, add 16 more SPs and they'll do some simple improvements.

How did cj of daamit get the new cards I wonder?

Looks like the HD48xx series will be beaten by the GT2xx series again. But I don't mind, as long as the lower the HD3870 to about $100 when the HD48xx series releases I'm happy.

Quote:

---

Originally Posted by gOJDO

I doubt that nVidia will make a dual-chip GT200 based card. Even with a die-shrink, a single die is still going to be a space heater.

IMO, with die-shirnk nVidia are going to raise the clocks of GTX 280, add 16 more SPs and they'll do some simple improvements.

---

The GT200 seems to have only 240 SPs instead of the popular consensus of 256 + 96TMU. No redundancy that is, like the 8800GTX at launch.

They can't really shrink to much efficiency, they have to get a balance between the bus width (too big) and ROP count (too much). 384bit + GDDR5 in the future seems to be the best balance, otherwise they'd still have to use a 400+mm^2 die in 40nm!

Why are people disregarding a chance for ATI to make the same RV770, just with double the specs, and still a small chip compared to the GT200 later on (units are small compared to memory bus), I have no idea. That chip is apparently, called the R(V)870 or what you may say otherwise.


Heard GT200 info from a buddy who has looked at the core die itself, so I'd let you guys go on from there. ;)

Quote:

---

Originally Posted by awdrifter

Looks like the HD48xx series will be beaten by the GT2xx series again. But I don't mind, as long as the lower the HD3870 to about $100 when the HD48xx series releases I'm happy.

---

The HD4k series was never ment to compete w/ the GTX2xx series.

Quote:

---

Originally Posted by w0mbat

The HD4k series was never ment to compete w/ the GTX2xx series.

---

That point will never get across m8, its useless to try.

Heck, Nvidia has their GTX 280 going up against the 3870 on their editor's day last week :rofl: , so there's no way people will suddenly realize that a $350 card was never meant to compete with a $600 card. :p:

Over time members will see the performance and price and decide for themselves for what they "NEED" for their gaming needs :cool:

Perkam

Quote:

---

Originally Posted by Macadamia

They can't really shrink to much efficiency, they have to get a balance between the bus width (too big) and ROP count (too much). 384bit + GDDR5 in the future seems to be the best balance, otherwise they'd still have to use a 400+mm^2 die in 40nm!

---

40nm should allow them to shrink the core to ~220mm2, so +400mm2 for a dual GPU card.

Quote:

---

Originally Posted by LordEC911

40nm should allow them to shrink the core to ~220mm2, so +400mm2 for a dual GPU card.

---

It doesn't work like that. Look at G80->G92. It went from 90 to 65 (same steps as the GT200 shrink) BUT was still crazy big (320-330mm2) and that was with 2 64-bit mem controllers taken.


GT200 won't shrink well. At least the memory bus won't. Even with that out of the way (256 bit), they will still be seeing 350+mm^2 per chip (not the big deal) and 120-150W per chip (BIG deal)