Preliminary ATI HD4870 benches

If it's the non-X2, maybe. This card isn't really meant to compete with the 8/9-series. Still, impressive if true.
 
Good stuff, though I think everyone wants to see how the HD 4870 does against the 9900 GTX.

Though if it retails at $299, that would be a really attractive price.
 
I honestly do not get these benchmarks. I have a similar system and disabling one of my 9800GX2's I score higher in every test then what is show here. Almost identical to the HD4870 except in the 3DMark score. But I can care less about a synthetic benchmark.
 
Yeah, grain of salt time. There are reasons to believe a site with pre-release scores, like having previous access to unreleased cards and/or at least showing a picture of the card (and driver model detection, GPU-Z, etc).

The source page has no proof they tested a HD 4870 card really.
 
These have already been posted on that HD4870-thread. Performance numbers are way upwards compared to specs:

In Crysis at 1680x1050 HD4870 wins 9800 GX2 by 29% which is pretty intresting number because:
1) If you'd put two HD3870's and assume you would get 100% efficiency from using those two cards you would still lose by 15% compared to single 9800 GX2
2) With those specs HD4870 would be less than two HD3870-cards

It should out perform HD3870 X2 though, but not that much.
 
Niceone said:
These have already been posted on that HD4870-thread. Performance numbers are way upwards compared to specs:

In Crysis at 1680x1050 HD4870 wins 9800 GX2 by 29% which is pretty intresting number because:
1) If you'd put two HD3870's and assume you would get 100% efficiency from using those two cards you would still lose by 15% compared to single 9800 GX2
2) With those specs HD4870 would be less than two HD3870-cards

It should out perform HD3870 X2 though, but not that much.
Click to expand...

assuming 100% efficiency is just wrong. At best you might get 75-80% efficiency.
 
Niceone said:
These have already been posted on that HD4870-thread. Performance numbers are way upwards compared to specs:

In Crysis at 1680x1050 HD4870 wins 9800 GX2 by 29% which is pretty intresting number because:
1) If you'd put two HD3870's and assume you would get 100% efficiency from using those two cards you would still lose by 15% compared to single 9800 GX2
2) With those specs HD4870 would be less than two HD3870-cards

It should out perform HD3870 X2 though, but not that much.
Click to expand...

You're forgetting a few upgrades to this new series. The core speed has been increased. The SPs speed now have a even more of a increase thanks to a multiplier similar to what Nvidia has done (I think ATI even managed to make their design more efficient). The TMUs have been increased from 16 to 32, allowing more of a fill rate (which was a major weakness in the R600 design).

All that combined possibly could out do the 3870 X2. SPs aren't everything.
 
I think he meant best-case scenario when he said 100% efficiency. He's saying that even at 100%, 2 x 3870's still can not beat a GX2, and since 2 x 3870s have combined higher specs than then suppose 4870, there's no way the 4870 could be that good.
 
Niceone said:
These have already been posted on that HD4870-thread. Performance numbers are way upwards compared to specs:

In Crysis at 1680x1050 HD4870 wins 9800 GX2 by 29% which is pretty intresting number because:
1) If you'd put two HD3870's and assume you would get 100% efficiency from using those two cards you would still lose by 15% compared to single 9800 GX2
2) With those specs HD4870 would be less than two HD3870-cards

It should out perform HD3870 X2 though, but not that much.
Click to expand...

how do you come to that conclusion? A single HD4870 has 2X the numbers (2.5X Texture, 2X shader, 1.9X mem bandwidth) of a single HD3870 across the board. Thats not including the problems we know the R600 family has with rasterization.

http://www.gpureview.com/show_cards.php?card1=564&card2=547
 
xBanzai89 said:
You're forgetting a few upgrades to this new series. The core speed has been increased. The SPs speed now have a even more of a increase thanks to a multiplier similar to what Nvidia has done (I think ATI even managed to make their design more efficient). The TMUs have been increased from 16 to 32, allowing more of a fill rate (which was a major weakness in the R600 design).

All that combined possibly could out do the 3870 X2. SPs aren't everything.
Click to expand...
SP's aren't everything, but this has slightly less theoretical TMU-performance than HD3870 X2 , much less ROP-performance, a lot less SP performance. Of course putting all that in to one chip helps, but there just doesn't seem to be anything that would make it that much faster than 9800 GX2. I mean 9800 GX2 out performs HD3870 X2 by 45-55% and now we should believe that HD4870 out performs 9800 GX2 by 30%? If HD3870 X2's performance would be 100, then 9800 GX2's would be 145-155 and HD4870's 189-202.

Basically HD4870 should be 90-100% faster than HD3870 X2? It's not like HD3870 X2 would be at HD3870 level in Crysis..

..and we aren't talking about new architecture here that could explain this.
 
MrWizard6600 said:
how do you come to that conclusion? A single HD4870 has 2X the numbers (2.5X Texture, 2X shader, 1.9X mem bandwidth) of a single HD3870 across the board. Thats not including the problems we know the R600 family has with rasterization.

http://www.gpureview.com/show_cards.php?card1=564&card2=547
Click to expand...
Well in that case 9800 GX2 should still win by 15% since 2xHD3870 isn't enough..but according to that article HD4870 would win by 29%?

Other funny thing:
That article which we are talking about here has different specs for this card than gpureview

gpureview:
Core: 850MHz
Shaders: 1050MHz
Memory: 3870MHz

Article:
Core: 800MHz
Memory: 3400MHz

Besides RV770 is still R600-architechture so it's unlikely that it would have different clocks for shaders.
-------

Besides are these numbers what? Max, min or average fps? I mean in Crysis 9800 GTX wins HD3870 X2 by huge margin in minimum fps-numbers, but it's tie in average and maximum.
 
GT200 should be basically 9800 GX2 on single chip
RV770 should be basically HD3870 X2 on single chip

GT200 would be made with 65nm as were G92 and would be based on G92
RV770 would be made with 55nm as were RV670 and would be based on RV670

..but what about rumours about following issues

Power consumption:
GT200: 250W [9800 GX2's TDP is less than 200W]
RV770: 150W [HD3870 X2's TDP is more than 200W]

Die-size:
GT200 should be 85% larger than G92
RV770 should be 30% larger than RV670
-----

Intresting at least Nvidia would go backwards with their power consumption efficiency by a great deal when AMD would take huge step forward despite the fact that there aren't that much changed in architecture

And that die size..If Nvidia 'doubles' things it means that die size almost doubles, but that wouldn't be an issue with AMD?
 
Higher core clock = Higher SP clock. 4000 series won't have the same SP clock as the 3000 series. Its still a R600 type of VGA but there has been tweaks to make it better. Such as from what I've heard, ATI managed to fix the AA issue they had with the R600.
 
On paper the R600 was awesome. There's no telling what a few of the right modifications could do.
 
I guess well see. From reading this stuff, I'm temped to get another 9800 GX2, for some reason I'm starting to have quad-SLI envy. This is what is really tough about being an enthusiast without unlimited funds. You want to buy the best, so you wait and see, only to be disappointed. Then when you go buy, BAM, had you waited another 2 weeks you'd have done a lot better. I actually might go eVGA for my next 9800 GX2 so I can step up. I've being buying XFX for the warranty but I'd like a little insurance this time.
 
heatlesssun said:
I guess well see. From reading this stuff, I'm temped to get another 9800 GX2, for some reason I'm starting to have quad-SLI envy. This is what is really tough about being an enthusiast without unlimited funds. You want to buy the best, so you wait and see, only to be disappointed. Then when you go buy, BAM, had you waited another 2 weeks you'd have done a lot better. I actually might go eVGA for my next 9800 GX2 so I can step up. I've being buying XFX for the warranty but I'd like a little insurance this time.
Click to expand...

It took you 2.5 years to realize this? Welcome to the club. To everyone asking what they should get as their new rig, I always offer the same piece of advice: In this business, another $2 and another 2 mins gets you 20X performance. Wait as long as you possibly can and prepare to go $100 over your budget (but don't include that $100 in your budget, you have to scam yourself into the right state of mind). Its just the way things go down. You have to draw the line somewhere, and be absolute in your choice.

This forum has people (including myself) who stick their heads up into the rumor stream and I think It puts us a few weeks ahead of the game, but then its always a judgment call. Sure, Fudzillas estimates on such and such a chip seem reasonable, and there's nothing to really suggest otherwise, but it is still fud, and making choices based on that can still burn ya. On a side note, Fud seems to have really toned it down, alot of what he said about the RV670 when it was still TBA was quite accurate (relative to his "ZOMG 30,000 3D MARKS w/ R600 AND PHENOM ZOMGZOMGZOMG" rants) and there was nothing so horribly exaggerated to speak of. He seems to be keeping his bais in check. If his guesses for RV770 turn out to be accurate I might actually consider him a decent source.

I think the RV770 is going to be a winner. There are my chips, its in, I'm in for RV770 being a successful chip. ATI gambled with an 80nm process and it went south. ATI gambled with a 55nm process and it seems to have paid off. I think they've fixed whatever rasterization (or otherwise) problem that was causing them so much AA and general ROP problems, I think the new 55nm chip, being ATIs 2nd crack at it, is as sucessful as fud thinks it is, and I think 4GHz GDDR5 isn't unreasonable considering samsung demod it at 6GHz just a few months back.

Denab on the other hand, now theres a failure. Nehalem is going to slaughter it.

As for the GT200, I donno, everyone seems to have their own seperate opinion of what it is. Many speculate its some form of new dual GPU chip, perhalps two dies on a single substraight? Fuds claiming its 65nm. VRzone confirms fuds estimates on 250W, and it would seem that with 1 Trillion transistors, a 65nm process would produce 250W. Maybe its two G92b chips? When Nvidia started the 9 series on such a POS, I figured they were clearning the decks for something new, like perhalps a new change to the nomenclature so as to release the GT200 as like the Geforce 1080, or perhalps it would be the end to Geforce? Now rumors are abound that the GT200 will be named the Geforce 9900s, which is a dissapointment, as it suggest Nvidia hasn't made any major strides over their current swath of G92 based products.

ANYWAYS, wait and see, there you go. I just don't want to wait, I want to see.
 
xBanzai89 said:
Higher core clock = Higher SP clock. 4000 series won't have the same SP clock as the 3000 series. Its still a R600 type of VGA but there has been tweaks to make it better. Such as from what I've heard, ATI managed to fix the AA issue they had with the R600.
Click to expand...
Well as long as they are based on R600 they'll have and according to that 'review' it will have 800MHz for core clocks
 
What is this ... I just bought my HD3870 and it's happily humming in my ShuttlePC. FawK! -it never ends-
 
Obi_Kwiet said:
On paper the R600 was awesome. There's no telling what a few of the right modifications could do.
Click to expand...

That's the MAIN problem I have with AMD atm. Everything they put on paper sounds great. The problem we have as consumers is what we read & see is what we expect to get. AMD, on both the GPU & CPU end has been a let down over the past year & a half. Well, I can't really say the 3870 was a letdown. It just didn't perform to some peoples expectations & I still believe the low priced point they set for it reflected that. I do want to see a good card come out of AMD once again. We can only hope something good comes out of the 4870 line. If it's one thing I can say about Nivida. They are not the greatest puncher but they have one hell of a counter punch (refresh) that usually hurts AMD bad.
 
DarkLegacy said:
That's the MAIN problem I have with AMD atm. Everything they put on paper sounds great. The problem we have as consumers is what we read & see is what we expect to get. AMD, on both the GPU & CPU end has been a let down over the past year & a half. Well, I can't really say the 3870 was a letdown. It just didn't perform to some peoples expectations & I still believe the low priced point they set for it reflected that. I do want to see a good card come out of AMD once again. We can only hope something good comes out of the 4870 line. If it's one thing I can say about Nivida. They are not the greatest puncher but they have one hell of a counter punch (refresh) that usually hurts AMD bad.
Click to expand...

personally I think that the 8800GTX was one hell of a right hook :p
 
I don't believe the review *BUT* I will say that the specs on RV770 might not seem impressive in comparison to the RV670, but that's not how microarchitecture works.

First of all, if R600 was truly bottlenecked, then it stands to reason that fixing that bottleneck can unleash what was being held back. Some say it was ROPS, but AA was already done by shaders so whether it stays at 16 or goes to 32 doesn't seem to be the problem. Some say it was TMUs - which according to gpureview and beyond3d was the problem, as it was combining AA and AF performance that hindered its performance (and lets face it, who doesnt turn on AF these days?). Doubling TMUs will certainly help, and if the 32 TMU / 480 SP rumor is true, then perhaps TMU's were the big bottleneck as there wouold be no reason to increase SP's if 32 TMU is not enough to utilize it.

And that's not even going into the possibility of a divided core/shader clocks (so 850 core / 1050 shader would certainly be impressive). GDDR5 seems to still be a month off from mass production (latest rumors say 4850 w/ GDDR3 first followed by 4870 and 4850's w/ GDDR5 - and yes, the 3.7+ GHz effective frequencies do exist for GDDR5, its a huge jump compared to GDDR3->GDDR4). Then again, memory bandwidth never seemed to be the problem with R600 (512-bit vs. 256 bit didn't make a difference, whereas it seemed to affect the G80->G92's a lot more).

So as of right now, performance is up in the air as to how good this thing can be since we don't even know if any of those values are correct.

However, do understand that the R600 architecture itself isn't bad. The original design was meant to be very expandable - that is, you can add blocks of it together and grow it with each new generation and manufacturing process. What killed R600 was that the 80nm process was a disaster - high leakage current and so on. Going to 55nm for RV670 certainly brought the card much more respect, but of course it was too late. However, 55nm has treated ATI very well recently and if all those A12 revision rumors of RV670 (as in core clocks increased towards the 800-900 range) carry over to the RV770 process, then there certainly can be room for an impressive card.
 
You must log in or register to reply here.
Back
Top