The ATI Radeon 5XXX Thread

[Morais]

High Quality Pics of PCB
and here

Fixed

[tdream]

Nobody posted this.

http://forum.beyond3d.com/showpost.p...postcount=2969

AMD's slides.

http://www.xtremesystems.org/forums/...&postcount=820

2 pages back

[LaMpiR]

High Quality Pics of PCB:

1, 2, 3, 4, 5, 6, 7, 8, 9

Doesn't work for me.

[Origin_Unknown]

arg - stop posting the same stuff!

[skugpezz]

The fact that those percentages means the game will STILL not be playable fully maxed out at 60fps high resolution. Something most of us have wanted to do for two years and this makes it FOUR gpu line-ups later(8800/2900, 9800/3870, GTX280/4870, and now Rv870/GTX380). That's if you can count the 9800/3870 as another line up, but ATi/NVidia sure did.

So yes, we're definitely a bit disappointed here.



Easy, driver bugs. There's a lot of cases in games where certain AA/AF combinations trigger issues, like huge loss of performance(sometimes AA works on one card, but chokes the one above it thanks to driver issues), or in the case of the GTX 295 SLi breaks(Wolfenstien is an example of this). That's why the best we can do is wait for the full reviews, as they are sure to cite any issues or show anything unsual about numbers acquired.... I really hope those crysis numbers are wrong.



Crysis isn't the only benchmark, but how many other games right now are still unplayable maxed out even current hardware? Most people still rocking the old 8800GTX's haven't had the need to upgrade because they still play everything fine! Nothing else has looked as good as crysis, and no other game has brought hardware down like crysis... Thus, yes our eyes are on it's performance.

Also, it has been stated time and time again, the problem IS the hardware and not the engine. No other game has that many shaders, that far of a draw distance, that caliber of special effects, that many polygons, that much effected by the physics, AND that high resolution of textures at the same time. It was made for future hardware, but sadly that deal was sealed right before the sudden slow-down in hardware acceleration, which began when ATi realized they couldn't continue to compete in the high-end after the R600 costed them way more than they ever made back off of it.

In short, Crysis is the only bench that gives us a glimpse of what future games will really push these cards to, and as such it's the numbers most of us care most about. IF any other game tried to push that level of detail, you wouldn't be able to max that out either...

so you expected the 5870 to be 4x 4870...LOL

[Jamesrt2004]

Seriously: what are you guy smoking there?

Do you really think that after 1.5 years without ANY product release ATI will introduce something that's only 10% faster than its competitor's fastest card, spending zillions on R&D, manufacturing, marketing etc while knowing the counter-strike (GT300) just couple of months (Jan-Feb) away?

Seriously, whatever you are smoking there that makes you think this I'd be afraid to try it.

well when your dissapointed cos your overhyping a great product dont blame me, even if im right ill be happy cos its great, if not then ill be even happier... win/win....

your over hyping STILL even with benches ... at minimum it seems 5~% above 295 which is awesome and i stand by what I said.... at $100 cheaper then the 295 its a great deal and price


and even then thats comparing, lowered pricings.... against launch prices so meh...

[jaredpace]

I call bull

The numbers for GTX 285 are about right, between 50-60fps at 1920x1200 4xAA/16xAF; numbers for HD 4870 X2 are too high, should be lower, also 55-65fps. Single HD 4890 sucks in HAWX... 28fps, in that chart it's faster than GTX 285...
source: http://www.xbitlabs.com/articles/vid..._13.html#sect2

here another source with HAWX 1920x1200 4xAA/16xAF
http://www.tomshardware.com/charts/g....W.X,1337.html
GTX 285: 58.90
HD 4890: 37.90
HD 4870X2: 63.90

bottom line: GTX 285 is faster than HD 4890 in HAWX.

jmke, it's with or without AF

Ati 4890
1920 * 1200 4AA/16AF 37fps
1920 * 1200 4AA/1AF 67fps
http://74.125.47.132/translate_c?hl=...j7m3vvvL7csRug

[Sh1tyMcGee]

Nobody posted this.

http://forum.beyond3d.com/showpost.p...postcount=2969

AMD's slides.

I wish they would announce a release date already with some concrete facts, all this teasing is getting a bit annoying.

[Farinorco]

If these charts are more or less accurate, HD5870 is more or less what I expected when specs were finally leaked. I've been trying to pack some info from that mess of bars and numbers, and more or less:

HD5870 vs GTX285: Average +53%, with 40% of the tests being >+50% difference, with huge variations due to the different configurations (specially AA).

Taking into account only 4xAA mode except in the few cases where it's not possible (STALKER CS and BattleForge) for consistency, I got: Average +46%, with ~33% of the titles being >+50%, and ~66% of the titles being >+33%. Those results are much more consistent between them.

Great improvements since RV770, I think this is going to make hard for NVIDIA to change the last round situation (little to no performance advantage depending on the models for much bigger and higher end focused chips = losses and difficulties to compete on other market segments) when they release their GT300 parts, but let's hope the Green Goblin can make a miracle (like ATi did with RV770) and balance the market again...

[XKaan]

This is more or less exactly what I expected as well, and having said that I'm pleased.

Currently I am running a 4870X2, so I will be waiting until the 5870X2 comes out to upgrade. The performance increase will be worth the investment, especially since I am sure to get some decent money out of my 4870X2 anyways.

[Olivon]

die shot, dual core?

Fake done by a canadian fantroll, Wirmish ... Shame on him

[jaredpace]

thank you, we suspected fake

[tajoh111]

Oh, I know full well crysis isn't poorly optimized. The engine was written for future cards, but sadly the following generations never delivered like CryTek had hoped for. I mean, at the rate hardware was going when crysis was being worked on, it would've hit the point required by now if it kept at that pace, ignoring dual-gpu cards.

9800xt to x800xt around a 100% increase, an even higher gain going with NVidia from fx5950 to 6800Ultra
6800Ultra to 7800GTX, again, roughly double, x1850xt to x1900xtx(I think we can all say the x1800 was to short lived to count, so we'll skip to the x1900xtx.)
7900GTX to 8800GTX... Do I even have to mention this one? ATi didn't make nearly as big of a leap on this one.
8800GTX to 9800GTX... Was this even an upgrade? I know in some cases the 8800GTX out-performed thanks to the higher rop and memory bus.
9800GTX to GTX 280? nowhere NEAR the same leap...

Basically, if we had stayed on the same course we were on, the GTX 280/4870's would've been 2 to 4x what we saw(making the dual-cards insanely fast), which would've put crysis playable at even 1920x1080p WITH 8xAA. Sadly, now we probably won't see that for another generation, unless NVidia has a serious trump card up their sleeves, which I wouldn't be surprised if they do. Many forget, during the "road to G80" not much was said other than it would be a DX10 part. Infact, NVidia just sat silent while sites like xbit labs claimed it would be a 48 PS/24 VS card. Silence means nothing from this company.

That said, I'm not placing any bets at all on this round, but the company that wins in crysis is likely to win the whole thing. Pretty much everything else is already playable maxed out anyway on current hardware... The fact that this card loses to the GTX 295(which granted, that card IS a beast and is dual-gpus) in crysis, and barely wins in crysis war head makes me sad though, very very sad.

This is the worst part. These are AMD slides, so these are canned benchmarks which show best case scenario, and probably alot slower in alot of situations.

I am surprised alot more people didn't question the first crysis benchmarks someone posted without even a screenshot and consisted of just forum text. Similarly people aren't even pointing out how bad these slides are and that they are even AMD slides. Just shows how biased his forums has become lately.

Looks like I get performance like this or better with my trifire setup. This consume alot less power obviously, but I think I will wait to see what NV comes up with.

If NV is able to cure its AA8x problems, AMD is in for a beating.

[Final8ty]

This is the worst part. These are AMD slides, so these are canned benchmarks which show best case scenario, and probably alot slower in alot of situations.

I am surprised alot more people didn't question the first crysis benchmarks someone posted without even a screenshot and consisted of just forum text. Similarly people aren't even pointing out how bad these slides are and that they are even AMD slides. Just shows how biased his forums has become lately.

Looks like I get performance like this or better with my trifire setup. This consume alot less power obviously, but I think I will wait to see what NV comes up with.

If NV is able to cure its AA8x problems, AMD is in for a beating.

Maybe because they are AMD slides & people couldn't careless about them why people are not commenting on them.
But parts of your post surely shows your bias, so why complain about bias.

[jmke]

jmke, it's with or without AF

Ati 4890
1920 * 1200 4AA/16AF 37fps
1920 * 1200 4AA/1AF 67fps
http://74.125.47.132/translate_c?hl=...j7m3vvvL7csRug

ah, good catch
I'd rather have 2xAA/8xAF then 4xAA/1xAF though; for most recent cards 4xAF comes free of performance hit...

[Calmatory]

Calmatory,
I'd say...
IF R600 was made on 65nm like the rest of HD2k series, it could have had 16RBE/32TMUs/640SPs and it would've easily competed/beaten G80. nV would have been forced to make the successor of G80 (that is, 65nm "G90") a real upgrade with more SPs than on G80, instead of a tweaked shrink of G80, aka. G92. Then both would've shrinked to 55nm just to decrease die sizes. Then a new round; R600's architectural successor could have been a 55nm 24ROP/56TMU/1120SP chip (320SP RV670 would never had existed!) to battle nV's 55nm 32ROP/64TMU/256SP "G100". Then shrink to 40nm -> ad nauseam.

That would have only made the last gen faster. The new stuff would be around just as fast.

Unless AMD would go with HUGE dies again. Perf/mm² would remain about the same. Same with the price and wattage etc.

[Olivon]

Another leakage from CJ Forum :

Beyond3D

~P17.000
~X8200
Tested on a Core i7 965 system with 6GB memory.

HD5850 vs GTX285:



Big UP to the source !

[Helloworld_98]

Another leakage from CJ Forum :

that Crysis warhead benchmarks for 2560x1600 looks a bit off on ATi's side, 9fps is quite a lot in that range and in the other benchmarks the 5870 comes damn close to the 295 at 2560x1600.

I also wonder why they didn't include a 2560x1600 with AA benchmark

[T2k]

@t2k, it's 10% faster than the dual chip GTX295. The 5870 1gb is not the fastest card, there is still the 2gb 5870 six trillian, and the 5870x2, and possibly whatever this chip is.... hahahah

I have no idea what are you talking about. Each card will compete with its similarly-priced NV counterpart which to me still means 5870x2 will go against 295... if so then I think it's more than ridiculous to think it will only be 10% faster.

Any other way of comparing cards is utterly illogical idiocy, period.

[T2k]

well when your dissapointed cos your overhyping a great product dont blame me, even if im right ill be happy cos its great, if not then ill be even happier... win/win....

your over hyping STILL even with benches ... at minimum it seems 5~% above 295 which is awesome and i stand by what I said.... at $100 cheaper then the 295 its a great deal and price


and even then thats comparing, lowered pricings.... against launch prices so meh...

Put down that crackpipe, dude - you're not making sense, read my post again, it's all there.

PS: GTX295 is around $480 and I expect 5870x2 to be available around $500 ergo it is the only valid comparison for me.

[jaredpace]

ah, good catch
I'd rather have 2xAA/8xAF then 4xAA/1xAF though; for most recent cards 4xAF comes free of performance hit...

Yes same here. Ati's AF methods have never been as effective as NV's. Hopefully this changes with cat 9.10 and 5800 series.



From chiphell on new angle independent AF & SS AA:

The following is the structure of Everygreen profile:

20 SIMD Cores
- Each SIMD Cores contains a 5D ALUs, a total of 1600 ALUs

80 Texture Unit
- Each SIMD Cores are four TU

Has two Rasterizer unit, Hierarchial-Z units are double (two), but the Geometry / Vertex Assembler is still one each for

- The use of SIMD Core implementation of the triangle interpolation, no longer using a separate hardware unit to perform this operation, SIMD Core this added interpolation instructions (lrp?)

Improved the performance of Constant Buffer

Faster Geometry Shader Performance

OpenGL context, achieved a 12-bit sub-pixel accuracy, raising the Line rendering performance and speed of operation clippiing

Double-precision performance of 544GFLOPS.

IPC Enhancements
- More flexible implementation of dot product instructions
- To achieve single-cycle MUL, dependent ADD instruction Co-issue
- Support the provision of hardware implementation of the SAD instruction can be achieved through the OpenCL

Aspects of texture units and cache
- Complete the second element 68B bilinear texture filtering
- Completion of 272B 32-bit per second texture fetch
- L1 tex cache bandwidth 1TB / s
- L1 to L2 bandwidth of 435 GB / s
- Each memory controller with 128KB L2 cache
- Support for texture 16k * 16k
- Support for DX11 in BC6 / 7 Compressed Texture
- Perfect angle dependent anisotropic texture filtering, the performance of decay rate consistent with previous AF algorithm
- The details of the ideal degree of adjustment ensures a higher quality of texture mapping

AA, it will basically single-cycle performance of the overall fold up, but the texture units are now available from compressed AA color buffer to read data, and will provide Supersampling AA (previously only in the RADEON 8500 the public before)

Stream Computing aspects
- Follow the IEEE 754-2008 precision
- Provide NVIDIA GT200 introduced Coalesces Memory Write
- Scatter operation can be done single-cycle 64 32-bit values, the generation is 32, but ...
- Support for 32-bit atomic operations
- Flexibility to configure the 32KB LDS (each SIMD core of a LDS, each 32kB)
- Full-chip has a GDS (Global Data Share), size of 64KB
- To achieve global synchronization
--Append/Consume Buffer

[jaredpace]

I have no idea what are you talking about. Each card will compete with its similarly-priced NV counterpart which to me still means 5870x2 will go against 295... if so then I think it's more than ridiculous to think it will only be 10% faster.

Any other way of comparing cards is utterly illogical idiocy, period.

okay, then as you say the 5870x2 DOES go against the gtx295. current 295 price is $499, 5870x2 price set to be $599 at end of october.

But already the $399 5870 1gb single gpu card is destroying the $499 gtx295...

So what are you saying? What is rediculous about what is happening?

[jaredpace]

SWEET















http://gathering.tweakers.net/forum/...ges/1368223/10

[Olivon]

SWEET

^Right, Really Powerfull

[Farinorco]

This is the worst part. These are AMD slides, so these are canned benchmarks which show best case scenario, and probably alot slower in alot of situations.

I am surprised alot more people didn't question the first crysis benchmarks someone posted without even a screenshot and consisted of just forum text. Similarly people aren't even pointing out how bad these slides are and that they are even AMD slides. Just shows how biased his forums has become lately.

Looks like I get performance like this or better with my trifire setup. This consume alot less power obviously, but I think I will wait to see what NV comes up with.

If NV is able to cure its AA8x problems, AMD is in for a beating.

We can't know if those AMD papers are similar or different to what we will see when reviews come, but I don't think the numbers they are giving are bad at all. What did you expected?

GTX 285 was already ~25% more performant than HD4870 1GB at 4xAA configuration. Those papers show HD5870 like ~45% better than GTX285 at 4xAA. So both combined make a ~80% improvement over HD4870 1GB.

That's a wonderful real world improvement for a +120% theoretical computational power increase (10% higher clocks and 2x processing units) and a 33% memory bandwidth increase, given that those real world tests with games don't always take full advantage of the architectures (you know, there are titles more or less optimized for every architecture so not every title takes full advantage of it) and that GPU's are not the only variable that affects framerates.

I don't think NVIDIA has an easy path to "beat" this, depending on what you call "beat". Of course, if beating is having a marginal win in raw performance with a twice as big and expensive chip compared to the competitor... but this is exactly what happened last time, and I don't think NVIDIA considered that like if they had won anything...