PDA

View Full Version : 5870/5970 vs. GF100 (Fermi)


Pages : 1 2 [3] 4 5 6 7 8 9 10 11

shadow001
01-20-10, 04:45 PM
Well, we play different games and at different resolutions so as much as you may be maxing out your cards, I'm not sure I'll be so lucky, but we will see soon enough.

I can assure you that my mainstay game can cripple 3 x 280 @ 2560 x 1920 and equally the 5970 at the same settings on a 4GHz i7 so that's why I stand by my earlier comments.

I don't really want to get into a game specific discussion here as it is totally off topic.

Enjoy :)


Hence why i used the Crysis reference....I don't think there's anything harder to run at the moment,and i'm also using an i7 at 4 Ghz.


So you can imagine the scenario if Fermi's are even faster still at the hardware level.

Skinner
01-21-10, 06:38 AM
Going from one HD5970 to a second one seems to scale very nicely actually,at least onn average and maximum frame rates,not so much for the minimum though.

Got some strange results too
I went from two HD5870's in CF running on 950/1300 to two HD5970's in quad CF running on 785/1200 and my framerates more then doubled in certain apps (scene's with sparsed (so beautyfull) grid super sampling aa in Oblivion for example and certain place in Stalker clear Sky.)
In Farcry2 DX10 I get 130 fps with every thing as high as possible and 16xAF 8x (ms)aa

Skinner
01-21-10, 06:46 AM
I've ran benchmarks in Crysis at absolute max settings at 4X antialiasing and got the same result when benching with 8X antialiasing,using supersampled AA,not the lesser quality multisampling method.....The cards simply don't care,and they are running at stock clocks,imagine when i overclock them.


.

You know you have to run Crysis in dx9 too enable SSAA and set the AA tab in CCC to app pref and set ingame AA?

hell_of_doom227
01-21-10, 08:42 AM
Got some strange results too
I went from two HD5870's in CF running on 950/1300 to two HD5970's in quad CF running on 785/1200 and my framerates more then doubled in certain apps (scene's with sparsed (so beautyfull) grid super sampling aa in Oblivion for example and certain place in Stalker clear Sky.)
In Farcry2 DX10 I get 130 fps with every thing as high as possible and 16xAF 8x (ms)aa

Does ATI supports tri-Crossfire. How would 3xHD5870 scale?

Johnny C
01-21-10, 09:07 AM
Does ATI supports tri-Crossfire. How would 3xHD5870 scale?

gerardfraser has a 5970 and a 5870, which would effectively be tri fire. They were posting all kinds of benchies in the rumor mill gtx380 thread. Seems like tri-fire with 5870 speeds was the gravy zone in that heaven benchmark it was ridiculous the framerates they were gettings.

Skinner
01-21-10, 01:05 PM
Does ATI supports tri-Crossfire. How would 3xHD5870 scale?

http://www.benchmarkextreme.com/Articles/HD%205870%20TriFire/P2.html

But you won't get this kind of scaling in every app. some are crazy good, others are less impressive.

JoeJonnyBoy75
01-21-10, 03:39 PM
I'm not sure if anyone else mentioned this but the test results we've gotten from nVidia were manually overridden to force 16AF. This setting cannot be controlled by the benchmark.

I'd be interested to see the benchmark results on a 5970 running at 16AF. At least we'd have a more accurate comparison.

hell_of_doom227
01-22-10, 08:13 AM
http://www.benchmarkextreme.com/Articles/HD%205870%20TriFire/P2.html

But you won't get this kind of scaling in every app. some are crazy good, others are less impressive.

That was interesting, think it's worth adding another HD5870 :)

Johnny C
01-22-10, 09:57 AM
I'm not sure if anyone else mentioned this but the test results we've gotten from nVidia were manually overridden to force 16AF. This setting cannot be controlled by the benchmark.

I'd be interested to see the benchmark results on a 5970 running at 16AF. At least we'd have a more accurate comparison.

I don't know if most people do this or not. But the first thing I do after a new driver install is force 16X AF thru CCC or control panel....been doing it for years.

I don't even wanna look at a game that doesn't at least have 2XAA not to mention less than 16XAF would look like garbage.

Skinner
01-22-10, 01:02 PM
AF16x don't stress moderne GPU anymore as much. Btw, I mostly have it forced.

Johnny C
01-22-10, 04:56 PM
I'm not sure if anyone else mentioned this but the test results we've gotten from nVidia were manually overridden to force 16AF. This setting cannot be controlled by the benchmark.

I'd be interested to see the benchmark results on a 5970 running at 16AF. At least we'd have a more accurate comparison.

In FC2 or in Heaven?

Xion X2
01-22-10, 10:00 PM
GTX 285 SLI isn't as fast as 5970. One review (Guru3D) doesn't make a trend. If any of you can pull other reviews that show this, I'd like to see them.

FiringSquad, for example, clearly shows in their review that a 5970 is often twice as fast (in some cases THREE times as fast) as a single GTX285. Even if SLI scaled 100% on GTX285 it would often get beat.

On Stalker: Call of Pripyat, a single 5970 gets 94.6 vs 37.6 for a single 285 at 1920x:

http://img207.imageshack.us/img207/2638/fall1920.gif

That's well over twice the performance of a single 285, (~2.5x) so tell me how SLI will beat a 5970?

When you increase resolution, it gets even worse. 66fps for 5970 vs. 22 for 285, 3x the performance.

http://img96.imageshack.us/img96/8981/fall2560.gif


And now, Crysis. At 1920x, 5970, again, is more than twice as fast as 285, and I can tell you from experience that neither SLI nor Crossfire scale anything close to 100%. More like 50%.

http://img195.imageshack.us/img195/3872/cry1920.gif

Again, at a higher resolution the gap only widens. 29fps for 5970 vs 13fps for 285:

http://img32.imageshack.us/img32/3375/cry2560.gif


But that's just one site, so let's look at another.

Anandtech, testing Far Cry 2, got 75fps for the 5970 vs. 62 for 285 SLI:

http://img253.imageshack.us/img253/4233/20708.png


Everyone is also overlooking that 5970 is purposely undervolted and underclocked from the factory just so that it slips in under the power spec. Do a search and you will see that it's fact. 5970 uses higher binned chips than 5870 does which means they will clock just as well if not better as long as your cooling is sufficient.


Essentially AMD took the board they had designed for the high-end dual-GPU offering and lowered the clocks to meet the <300 watts requirement needed by many OEMs and consumer power supplies. But with the binned GPUs, higher speed memory and improved cooling solution AMD has provided a design with a LOT of overclocking headroom! Even just looking at one spec, the memory modules, shows the picture: though the memory is running at 1.0 GHz it is rated at 1.2 GHz.

...The cooler that AMD has included on the Radeon HD 5970 is actually a vapor chamber design that is rated at 400 watts providing a 33% gap between the rated TDP at stock clocks and where you can safely push your card. The Overvolt utility does exactly what you think it does – it increases the voltages going to the GPU and to the memory on the HD 5970. The software is incredibly basic – a slider bar with only two positions – and is essentially an on/off switch.



http://www.pcper.com/article.php?aid=820&type=expert&pid=3

All you have to do to leave GTX285 SLI in the dust even further is to click a single button in the Catalyst to set the 5970 to its recommended voltage and clocks.

Some of you have green-tinted glasses.

Hari Seldon
01-23-10, 07:14 AM
Just my take on the Fermi.
It's an impressive architecture, however I think they are less interesting for myself as a gamer.
Perhaps if I was using my computer for GPGPU'ing I'd consider buying it.
I've been buying nvidia card for a long time. My first nvidia card was the Riva TNT a long while ago. Only bought 2 non nvidia card since then, the R300 and now the Radeon 5870.
However, I think I prefer ATI's new design, a little smaller chip, cheaper to make and should draw less power

jolle
01-24-10, 07:09 AM
Just my take on the Fermi.
It's an impressive architecture, however I think they are less interesting for myself as a gamer.
Perhaps if I was using my computer for GPGPU'ing I'd consider buying it.
Every transistor in there that handles compute and GPGPU is part of the shader core that runs the shader code when gaming, so it´s not really a waste if you dont pull any GPGPU applications.
To add to that, GPGPU and compute does include running physics on the GPU in games, be it via PhysX, Bullet or Havok.. CUDA, OpenCL or DX Compute.
Postprocessing done via DX Compute seems like something we´re gonna see more of aswell, this is also a type of GPGPU usage.

Now, how ATi and NV compare in these situations, one GPU, rendering, doing compute based postprocessing and perhaps OpenCL based physics at the same time, we don´t know yet, both should have really fast context switching going between these different tasks atleast.

Im not really in a rush to upgrade yet which is nice, so I´m just gonna sit back and see how this all plays out.
It is comforting to know that the ATi offering is as solid as it is though, should NVs turn out less then awesome.

Slytat
01-24-10, 10:47 AM
GTX 285 SLI isn't as fast as 5970. One review (Guru3D) doesn't make a trend. If any of you can pull other reviews that show this, I'd like to see them.

<snip>

Some of you have green-tinted glasses.

Serious case of Pot, Kettle, black there.

I don't want to rekindle this argument again but in my case at least, I wasn't arguing that 285's (SLI) were faster, merely that they aren't that far behind.

As you seem so intent on reminding us about the 5970's voltage and clocks, I'm sure you'll concede that 285 SLI is Nvidia's LAST generation and is almost 2 years old. It's fairly impressive that it keeps up with ATI's brand new flagship card.

Another thing, if you are going to do comparisons, at least use charts that have 285 SLI results in them and stop using theoreticals.

http://www.guru3d.com/article/radeon-hd-5970-review-test/14

I know, you don't like G3D. The thing is that A : He does actually test 285 SLI and B : His test results are very similar to mine so I'm going to go with G3D and my own hands on experience. As you can clearly see in that test, the 5970 is faster, but not by much.

HAWX is the only game where I saw the 5970 over 15% faster in my own testing and I suspect that is driver related.

In conclusion, it is faster but for a new card, a ~10% performance increase is nothing to write home about and does not "leave GTX285 SLI in the dust even further".

I think it is you who has the red tinted glasses on :rolleyes:

PS : You can overclock 285s as well, the fact of the matter is, the 5970 will be faster, just not by much.

Johnny C
01-24-10, 12:03 PM
Serious case of Pot, Kettle, black there.

I don't want to rekindle this argument again but in my case at least, I wasn't arguing that 285's (SLI) were faster, merely that they aren't that far behind.

As you seem so intent on reminding us about the 5970's voltage and clocks, I'm sure you'll concede that 285 SLI is Nvidia's LAST generation and is almost 2 years old. It's fairly impressive that it keeps up with ATI's brand new flagship card.

Another thing, if you are going to do comparisons, at least use charts that have 285 SLI results in them and stop using theoreticals.

http://www.guru3d.com/article/radeon-hd-5970-review-test/14

I know, you don't like G3D. The thing is that A : He does actually test 285 SLI and B: His test results are very similar to mine so I'm going to go with G3D and my own hands on experience. As you can clearly see in that test, the 5970 is faster, but not by much.

HAWX is the only game where I saw the 5970 over 15% faster in my own testing and I suspect that is driver related.

In conclusion, it is faster but for a new card, a ~10% performance increase is nothing to write home about and does not "leave GTX285 SLI in the dust even further".

I think it is you who has the red tinted glasses on :rolleyes:

PS : You can overclock 285s as well, the fact of the matter is, the 5970 will be faster, just not by much.

It doesn't change the fact that the direct competition for a single 285 is a 5850....not the 5870 at all, not only that....the 5850 beats the 285 in just about every game out there...while costing less money.....and using less power. Big one here "costs less money" than nV's last generation, that they EOL'd instead of dropping prices to compete(because there just wasn't enough margin to drop pricing any further.) Hell the 4890 trades blows with the 285 and I've seen them for $200 bucks.

You wanna talk about Sli...talk about 5850 X-fire as competition, again...it will cost less, and use less power while returning better framerates. It will also support Eyefinity at extreme resolutions and DX11. I know you'll argue that the 5970 is essentially a 5850/5870 crossfire setup on a single card, which it is, and it's even less expensive than a 285 Sli setup making it an even better value.

If you're talking about people who already, at this point, own 285's...yeah they are still fast. I mean hell even an 8800GTX is gonna throw up decent numbers in most games with good quality setting, just not on the extremes.

If someone currently has a 285 SLi setup...then I see no need to upgrade to anything either nV or ATi at this point...and I wouldn't recommend it until such time as DX11 is more in the mainstream swing of things.

I jumped on the bandwagon upgrading from a 4870 to a 5850 and honestly....it was money well wasted. If I hadn't had a sale for my entire other system, I'd have been pretty damn disappointed because my 18 month old 4870 1gb played every game that I have.....on high settings with decent frames. Hell I even played thru Crysis DX10 with 2xAA....

Slytat
01-24-10, 12:06 PM
It doesn't change the fact that the direct competition for a single 285 is a 5850....not the 5870 at all, not only that....the 5850 beats the 285 in just about every game out there...while costing less money.....and using less power. Big one here "costs less money" than nV's last generation, that they EOL'd instead of dropping prices to compete(because there just wasn't enough margin to drop pricing any further.) Hell the 4890 trades blows with the 285 and I've seen them for $200 bucks.

You wanna talk about Sli...talk about 5850 X-fire as competition, again...it will cost less, and use less power while returning better framerates. It will also support Eyefinity at extreme resolutions and DX11. I know you'll argue that the 5970 is essentially a 5850/5870 crossfire setup on a single card, which it is, and it's even less expensive than a 285 Sli setup making it an even better value.

If you're talking about people who already, at this point, own 285's...yeah they are still fast. I mean hell even an 8800GTX is gonna throw up decent numbers in most games with good quality setting, just not on the extremes.

If someone currently has a 285 SLi setup...then I see no need to upgrade to anything either nV or ATi at this point...and I wouldn't recommend it until such time as DX11 is more in the mainstream swing of things.

I'm not talking about cost, power consumption or anything else. I was talking specifically about the 5970 vs 285 SLI in terms of speed.

Johnny C
01-24-10, 12:13 PM
I'm not talking about cost, power consumption or anything else. I was talking specifically about the 5970 vs 285 SLI in terms of speed.

Like I said....if you already own them....yes the 285's are still a great setup an extremely fast setup.

Buying today..Jan 24/10..you'd have to be daft to buy anything from nV. When GF100 is out the door, and we see some actual performance numbers and pricing, things might change. Until then anyone saying buy nV....is wearing green tinted glasses.

Slytat
01-24-10, 12:25 PM
Like I said....if you already own them....yes the 285's are still a great setup an extremely fast setup.

Buying today..Jan 24/10..you'd have to be daft to buy anything from nV. When GF100 is out the door, and we see some actual performance numbers and pricing, things might change. Until then anyone saying buy nV....is wearing green tinted glasses.

I own ATI and Nvidia and buying Nvidia now would be incredibly stupid. The only reason to buy Nvidia now would be if you were adding a second (or third) card to an existing setup :)

It's either ATI or wait for Fermi.

You are a little off topic tbh, I wasn't giving anyone advice on what to buy, just making an observation about the speed of the respective cards.

Johnny C
01-24-10, 12:32 PM
I own ATI and Nvidia and buying Nvidia now would be incredibly stupid. The only reason to buy Nvidia now would be if you were adding a second (or third) card to an existing setup :)

It's either ATI or wait for Fermi.

You are a little off topic tbh, I wasn't giving anyone advice on what to buy, just making an observation about the speed of the respective cards.

You're right...but since we have no Fermi numbers the whole thread is a little off topic.

But yes adding an additional card would be a good reason...hadn't thought of that.

shadow001
01-24-10, 12:34 PM
I'm not talking about cost, power consumption or anything else. I was talking specifically about the 5970 vs 285 SLI in terms of speed.


so keep in mind that even a single HD5970 hits CPU limitations far more often than an GTX285 SLI would under the same game at the same settings,so that also limits the gap between both,but it's not the cards fault if it can't be fed fast enough with the data it requires to perform at it's peak.


I'm sure Fermi will also suffer the sharper edge of the CPU limitation far more often than the GTX285 cards,and having said that,a GTX 285 SLI setup is still an awsome setup for pretty much any game on the market right now.


Here's the sick part.....We'll probably have cards out by either late this year or early next that will be close to twice as fast as an HD5970 X2,so what kind of CPU will be needed to feed that monster is what i'm wondering,especially in triple or Quad GPU setups....It's crazy basically.

Xion X2
01-24-10, 01:02 PM
Serious case of Pot, Kettle, black there.

How does stating the facts give me red-tinted glasses? You and Sowk said that 285 SLI was comparable to a 5970; I simply showed that it isn't. In some cases, a 5970 is 3x as fast as a 285. In most, it's at least twice as fast, and you will hardly ever see SLI on the 285 scale to 100%.

Simple deductive logic. Has nothing to do with being biased toward one company or the other.

I don't want to rekindle this argument again but in my case at least, I wasn't arguing that 285's (SLI) were faster, merely that they aren't that far behind.

You don't want to "rekindle" anything, yet you repeatedly ignore facts that are right in front of you. In most cases, a 5970 is at least twice as strong as a GTX285, and in some cases it is three times as strong. Given that SLI will almost never scale better than 80% in most titles, how does that equate to "aren't that far behind?"

As you seem so intent on reminding us about the 5970's voltage and clocks, I'm sure you'll concede that 285 SLI is Nvidia's LAST generation and is almost 2 years old. It's fairly impressive that it keeps up with ATI's brand new flagship card.

And here we go again. Are you really serious? "It keeps up?" Where, exactly? In one benchmark test? (Guru3D)?

Where are the rest of the tests that show this? Again, one case doesn't make a trend, especially when that review appears to be the outlier among other reviews that can be accessed across the web.

I've not seen another review yet that repeats Guru3Ds findings. If you can find one, then I would really like to see it. Honestly. But until I do, I will keep firm with my stance.

Another thing, if you are going to do comparisons, at least use charts that have 285 SLI results in them and stop using theoreticals.

No, because it's not necessary. You'd like me to, because then you could use your one little benchmark over at Guru3D to prove your point since there are hardly any 285 SLI benches out there that do a direct comparison to 5970. But it makes no sense to deduce that you can't accurately gauge how an SLI setup would run by simply doubling the performance of a 285 (SLI = 2x,) or is that math too complicated for you?

I think it is you who has the red tinted glasses on :rolleyes:

Yes, because guys with red-tinted glasses often spend 1200$ on Nvidia cards and 300$ on waterblocks to fit Nvidia cards to build watercooled SLI systems:

http://www.nvnews.net/vbulletin/showthread.php?t=85643&highlight=project+poseidon

And they often praise the architecture of their upcoming competitor:

But for a single GPU to come within 20% of that [5970].. think about it. It's pretty impressive.

And they often rip on their own drivers while praising their competition's drivers:

Nvidia drivers aren't "perfect," but they're better than ATI's.. especially when it comes to multi-GPU.

http://www.nvnews.net/vbulletin/showpost.php?p=2025008&postcount=9

Yes, I'm clearly biased. :rolleyes:

]PS : You can overclock 285s as well, the fact of the matter is, the 5970 will be faster, just not by much.

It's not overclocking. Read the PCPer article again:

though the memory is running at 1.0 GHz it is rated at 1.2 GHz. [/B]

They're not "overclocking" the card; they're setting it at its factory-rated speeds. The card is downclocked on both the memory and the GPU to slide in under power specs. This was not the case with GTX285; it was already running at its factory rated speeds.

hell_of_doom227
01-24-10, 02:24 PM
I was surprised by Crossfire setup. In my experience crossfire scales really well, i will probably add one more HD5870 card.(nana2)

Slytat
01-24-10, 03:26 PM
How does stating the facts give me red-tinted glasses? You and Sowk said that 285 SLI was comparable to a 5970; I simply showed that it isn't. In some cases, a 5970 is 3x as fast as a 285. In most, it's at least twice as fast, and you will hardly ever see SLI on the 285 scale to 100%.

Simple deductive logic. Has nothing to do with being biased toward one company or the other.



You don't want to "rekindle" anything, yet you repeatedly ignore facts that are right in front of you. In most cases, a 5970 is at least twice as strong as a GTX285, and in some cases it is three times as strong. Given that SLI will almost never scale better than 80% in most titles, how does that equate to "aren't that far behind?"



And here we go again. Are you really serious? "It keeps up?" Where, exactly? In one benchmark test? (Guru3D)?

Where are the rest of the tests that show this? Again, one case doesn't make a trend, especially when that review appears to be the outlier among other reviews that can be accessed across the web.

I've not seen another review yet that repeats Guru3Ds findings. If you can find one, then I would really like to see it. Honestly. But until I do, I will keep firm with my stance.



No, because it's not necessary. You'd like me to, because then you could use your one little benchmark over at Guru3D to prove your point since there are hardly any 285 SLI benches out there that do a direct comparison to 5970. But it makes no sense to deduce that you can't accurately gauge how an SLI setup would run by simply doubling the performance of a 285 (SLI = 2x,) or is that math too complicated for you?



Yes, because guys with red-tinted glasses often spend 1200$ on Nvidia cards and 300$ on waterblocks to fit Nvidia cards to build watercooled SLI systems:

http://www.nvnews.net/vbulletin/showthread.php?t=85643&highlight=project+poseidon

And they often praise the architecture of their upcoming competitor:



And they often rip on their own drivers while praising their competition's drivers:



http://www.nvnews.net/vbulletin/showpost.php?p=2025008&postcount=9

Yes, I'm clearly biased. :rolleyes:



It's not overclocking. Read the PCPer article again:



They're not "overclocking" the card; they're setting it at its factory-rated speeds. The card is downclocked on both the memory and the GPU to slide in under power specs. This was not the case with GTX285; it was already running at its factory rated speeds.


You can pick and choose the "facts" that support your argument but your opinions really don't matter to me and I couldn't care less whether you've said "nice things" about Nvidia.

What does that have to do with anything?

As I've repeatedly told you, I've done my own testing and it's in line with G3Ds. The 5970 just isn't that much faster (but it is slightly faster). 10-15% faster against 18 month old technology might get you excited but it doesn't do much for me.

As I said the last time, we'll just have to agree to disagree.

Thanks

:)

PS : Maybe you should write to Hilbert (the owner of G3D) and tell him he doesn't know how to do benchmarks properly :rolleyes:

Woodelf
01-24-10, 04:12 PM
This may cause some to not include the "5970 are meant to run that fast" argument when basing their purchase decisions with dx11 cards:


"AMD's product warranty does not cover damages caused by overclocking, even when overclocking is enabled via AMD OverDrive™ or ATI OverDrive™ software"

http://www.amd.com/US/PRODUCTS/DESKTOP/GRAPHICS/ATI-RADEON-HD-5000/HD-5970/Pages/ati-radeon-hd-5970-overview.aspx