Radeon HD4870 scores 12600 in 3Dmark

[hayder.master]

Radeon HD4870 scores 12600 in 3Dmark
i don't know but the site say this, what you think

http://www.fudzilla.com/index.php?option=com_content&task=view&id=7973&Itemid=1

[wolf]

lol, lower scores with the 4870 than my 9800GTX with the quad at stock :P

not to mention nvidia is on a roll making drivers that boost performance on G80/92 considerably.

http://forums.techpowerup.com/showthread.php?t=62762

[Kasparz]

Radeon HD4870 scores 12600 in 3Dmark
i don't know but the site say this, what you think

http://www.fudzilla.com/index.php?option=com_content&task=view&id=7973&Itemid=1
If it scores 12.6k on Q6600, it will score 15k with yorkfield at 5GHz. Or overclocked it will break 20k barrier with single card.

[wolf]

not many people have a yorkfield at 5ghz man.... most have a core 2 duo/quad in the 3-4ghz range, and AMD users will have even lower cpu scores.

and using the 3dmark 06 score calculator, my 9800GTX on a 5ghz quad core would pull ~19,000 points.....doesn't really seem next gen to me, no wonder they need to go the dual-gpu-in-one-solution route to come out on top

[tkpenalty]

This proves AMD hasnt optimized the card for 3D Mark 06. 3D Mark does not prove actual performance anyway. Nothing surprising, the GTX280 also scores "low", around a 9800GTX.

Some of the mods hate that site with a passion... I'd wouldnt, as they usually are right...

BTW. To debunk the 480SP RV770: http://www.fudzilla.com/index.php?option=com_content&task=view&id=7955&Itemid=65. Ask w1zz yourself too...

[sefu]

That score is embarasing :laugh: 17179 3D06 with my 9800GTX, and can do more.

[wolf]

hey sefu, grab the 177.35 drivers from in the thread i linked above, and see if your score goes up, i got ~750 points for free

[sefu]

Tks y will try later today.

[wolf]

drivers from post number 121 and modded inf's from post 114 :)

[DanishDevil]

I think it's crap. Did they use the GDDR3 version? I just about break 15k with my E8500 and 3870x2 @ stock...

[wolf]

iunno man it seems plausible enough

[panchoman]

its fud guys, dont get too worked up on it

[oli_ramsay]

I'm gonna raise the bull shit flag
http://ezuploads.com/309/1386309/10.jpg

that's a disgracefull score if it's true but considering AMD are claiming 4850 has 75% performance of GTX 280, then I doubt its real.

http://forums.vr-zone.com/showthread.php?t=290466

[DanishDevil]

I concur. It's crap.

[batmang]

I'm gonna wait for REAL reviews. None of this speculation CRAP.

[HTC]

I'm gonna raise the bull shit flag
http://ezuploads.com/309/1386309/10.jpg

that's a disgracefull score if it's true but considering AMD are claiming 4850 has 75% performance of GTX 280, then I doubt its real.

http://forums.vr-zone.com/showthread.php?t=290466

Nice ... flag ...

[Megasty]

Fud was raised on BS, so it can can only produce the like...

[wolf]

remember leaked GTX280/260 reviews were pretty spot on......

[ghost101]

When will people learn. You CANNOT use 3dmark06 for these new cards. You need to crank up the settings or move onto 3dmark vantage. We've already seen this with the HD4850 and GT200 cards. Cant believe people still use such an old benchmark.

As an example, try and get the 9k the hd4870 got with a 9800gtx @ stock with a quad at 2.4ghz @ 1920*1200 with 8x AA. Wouldn't get close.

http://www.overclock3d.net/gfx/articles/2008/06/15174157251l.png

This is with a quad @ 3.6ghz and AA @ 4x

[Megasty]

When will people learn. You CANNOT use 3dmark06 for these new cards. You need to crank up the settings or move onto 3dmark vantage. We've already seen this with the HD4850 and GT200 cards. Cant believe people still use such an old benchmark.

As an example, try and get the 9k the hd4870 got with a 9800gtx @ stock with a quad at 2.4ghz @ 1920*1200 with 8x AA. Wouldn't get close.

http://www.overclock3d.net/gfx/articles/2008/06/15174157251l.png

This is with a quad @ 3.6ghz and AA @ 4x

Atleast someone knows what is going on. The bottleneck for 3dmk06 is pretty darn apparent :respect:

[zanex]

Definitely need to mvoe to Vantage or use higher settings, 3dMark06 default scores are bottlenecked. Not to mention that you can't compare scores across different rigs.

Besides, 2 x 4850's in CF scored > 13k at 1920 x 1200 + AA + AF and X4800 to X5500 or so last I saw in Vantage.

[hayder.master]

i think this is something wrong in test of 4850, cuz i see it in other test's it run well , more than that score ,

[wiak]

so many nvidia fanboys bitching in a ati forum..
given the fact one 4870 scores around X4600 in 3dmark vantage against a GTX 280 that scores X5200, should say alot, but well everything comes in the price point, and the 4870 looks atlot good for its 299 price point, and 2x 4870 should be around %50 faster than GTX 280, and costs less at the same time ;)

[TonyStark]

Most likely CPU bound. My card gets around 11,4xx and the HD4870 is supposedly twice as fast.

[wolf]

Lol

[hayder.master]

right we need's one of us test it by himself

[dark2099]

Personally I think that is a little low considering my 4850 will do 12.2k (I think would have to check the screen shot) with my E8500 at 4.5ghz.

[wolf]

this just solidifies my point of how much ATi shaders suck compared to Nvidias.

now its a whopping 800, and nvidias 128 SP cards fare VERY well against them.

LOL

and don't go telling me that nvidias are clocked faster, i can do the math on this for anyone who wants to know it.

[erocker]

I'm still not convinced there are 800sp's on ATi cards. If you look around at some of the press releases for the 48xx series, manufacturers are posting 480sp's.

[btarunr]

I'm still not convinced there are 800sp's on ATi cards. If you look around at some of the press releases for the 48xx series, manufacturers are posting 480sp's.

I don't think 480 SP's would have been able to beat 640 SP's (In Crossfire HD3850). The HD4850 beats it.

[wolf]

mmmmmmm, in any case id say its odd that 240, 128 and even 64 nvidian SP's compete so well, what are ATi missing?

[btarunr]

mmmmmmm, in any case id say its odd that 240, 128 and even 64 nvidian SP's compete so well, what are ATi missing?

And you think NV and ATI use the same exact approach to the shader?

NV's shaders differ in many ways from ATI. Google out.

In turn, there's a demerit in NV's design. A 64 SP-laden 9600GT is only 5~7% slower than a 112 SP-laden 8800 GT. Increase in number of shaders isn't exactly exponentially increasing performance, while the jump from 320 to 640 (in HD3870 X2) is almost doubling up performance, better example is between 320 and 800...the performance is more than 2x (comparing HD3850 and HD4850).

Each company has design flaws, NV included.

[wolf]

your point is well made, however each shader on an ATI GPU can do MADD, and at best an nvidian can do MADD + MUL, it just doesnt add up, especially this "A 64 SP-laden 9600GT is only 5~7% slower than a 112 SP-laden 8800 GT"

and i mean, a different approach is one thing, but 800 vs 128 ? thats a helluva stack.

[btarunr]

One in every 5 SP's in ATI's architecture performs MADD + MUL. Effectively you have 64 of these in a R600, 160 in RV770. Add to that, NV's shaders get greater throughput out of higher clock-speed that's async to the core clock.

[wolf]

yes but 800 SP's at 800MHZ, and 128 SP's at 2000 MHZ still does not add up. (not stock clocks either, but serves my point of comparison)

like i said, i can do the math, and post it here, it's just the results dont make sense to me, the difference is too great to put down to architectural differences.

[btarunr]

One in every 5 SP's in ATI's architecture performs MADD + MUL. Effectively you have 64 of these in a R600, 160 in RV770. Add to that, NV's shaders get greater throughput out of higher clock-speed that's async to the core clock.

Learn how ATI's shaders work first. ATI believes in working out "wonderful numbers" to make the product specs look bright. Originally when G80 and R600 were the only "high-performers", NVidia regarded their shader architecture to be composed of 128 "Unified Shaders" (where "unified" = pixel + vertex + anything you program it to), while ATI said theirs had 320 "Stream processing units". What's a "stream"? A set of linear calculations with little or no "branched calculations" (or MADD + MUL). So they have 120/320/800/<insert wonderful number here> stream processors. Eventually the industry adopted "SP" as a unit, for detemining the quantity of "whatever" handles shaders in a DX 10+ GPU. And hence even NVidia uses "SP" because it's their way of laughing at ATI saying "we did it with just 128 SP's while they struggle with 320 of them" so that the average consumer (who doesn't actually understand how this thing works) holds NVidia on a higher ground.

yes but 800 SP's at 800MHZ, and 128 SP's at 2000 MHZ still does not add up. (not stock clocks either, but serves my point of comparison)
like i said, i can do the math, and post it here, it's just the results dont make sense to me, the difference is too great to put down to architectural differences.

How do you think those game developers that kiss NV's arse (TWIMTBP) "optimise" their games to NV? Simple. They know that in a 8800 GTX, there are 128 shader unts that can handle complex/branched calculations (MADD + MUL) so they simply snip their code to send the shaders complex calculations (and end up minimising the code), while in a HD2900 XT, effectively there are only 64 such complex units, so when it's bombarded by a process that sends these complex calculations, effectively the simple shader units are left under-utilised. In a game that's optimised for R600 (if there ever is), the programmers keep their code un-complicated so the calculations are more linear and hence the quantity of calculations are made up for by the quantity of calculators for these broadly linear calculations.

Now, an RV770 has 160 of these "complex SP's" and so, even with a game like Crysis (that's a full-on NV-arsekiss), it performs well. The G80 still ends up getting a little ahead because of its higher throughput owing to a higher shader clock.

[wolf]

given your points, wouldn't it be a more wise decision to go with the hardware that has games optimized for it?

[DanishDevil]

It's up to personal preference. Besides, those of us that don't have nVidia boards and like the Intel chipsets really have no choice in the matter if we want to consider multi-card setups. When it comes down to choosing ATi or nVidia, one has to look at more than just the card's performance.

Besides, I personally think that nVidia is falling very far behind with Hybrid SLI, and I like the option of being able to mix and match the 3000 series with the 4000 series.

nVidia may be ahead in performance at times, but since ATi begun its new Crossfire connectors with the x1950Pro, it has been leaping past nVidia with compatibility and options in multi-card setups.

And you have to admit, that if you want a nice OCing nVidia SLI board that doesn't have some huge design flaw like RAID corruption, you're going to be paying out a LOT of money. If you want a Crossfire board that will do the same, you have many many more options.

[btarunr]

given your points, wouldn't it be a more wise decision to go with the hardware that has games optimized for it?

Optimised games are a minority. Funny but true. Neutral games perform equally in both, additional memory bandwidth (in some ATI cards using GDDR4/5) and the new AA components the RV770 feature could just give you the edge over G92.

[yogurt_21]

160(complex)x800MHZ vs 128(complex capable)x2000MHZ would give you a better look at why ati's shaders arne't showing that much gain 640 of those units are not quite in use yet but will likely make a difference in physics. so in actuality if you look at complex svs complex ati's shaders seem to be able to do the sameb work faster at a lower clock speed making them more efficient than nvidias, you just have to look past the 800unit bs.
the rv770xt will have a 1050MHZ shader clock giving it more of a boost so we'll see how that and ddr5 do against the gtx260. you've got to remmeber that each company does things differently so just lookign at 800vs128 isn't going to do you any good. you have to compare complex shaders to complex shaders and then run the clock speeds. and clock speed is where nvidia is winning, and yet the 9800gtx fails to dominate 4850 despite having twice the complex shading power. so who's architecture seems to be inefficient?

[wolf]

in my opinion, its still ATI's.

[WarEagleAU]

Wow informative and yet confusing :D

[trt740]

Radeon HD4870 scores 12600 in 3Dmark
i don't know but the site say this, what you think

http://www.fudzilla.com/index.php?option=com_content&task=view&id=7973&Itemid=1

I think thats thats miore about drivers than speed at this point.

[trt740]

this just solidifies my point of how much ATi shaders suck compared to Nvidias.

now its a whopping 800, and nvidias 128 SP cards fare VERY well against them.

LOL

and don't go telling me that nvidias are clocked faster, i can do the math on this for anyone who wants to know it.

then why does a 3870x2 run it so well?

[wolf]

because it has twice the ROPS and close to twice the pixel fill rate that a 48xx card has, this accounts for alot of performance.

also twice the ROPS and pixel fill rate of a 9800GTX....

do you want me to do the figures?