New Drivers Cause HD3870 to Beat 8800GT with High RES & AA

[Soul Rider]

For those of you who doubt the validity of the german benchmarks and the performance of the 8800GT 512mb with 8xAA, look at this..

Quoted from Bit-Tech review, although their tests only go upto 1920x1200 -

In Quake Wars, there are two different stories to tell because, depending on the anti-aliasing settings, there are two quite different performance characteristics to take in. The first is a disappointing one, if we’re honest, because the factory overclocked Asus EAH3870 TOP is quite a bit slower than the similarly-priced Nvidia GeForce 8800 GT when 4xAA is applied.

It’s as if Nvidia has been hard at work optimising its performance with 4xAA and the trend isn’t helped by the fact that RV670 doesn’t feature a hardware-based MSAA resolve unit, which means the final resolve stage of the anti-aliasing process has to be handled by the shader. Of course, this gives much tighter control over the quality of the pixel sub samples, as you move from fixed function to floating point calculations for the resolve stage.

What’s interesting is that when you move to 8xAA, things change quite drastically and the Asus EAH3870 TOP actually manages to outperform most cards. This is because Nvidia’s G92-based products appear to suffer under the heavy workload because of the 256-bit memory interface – RV670’s ring bus memory controller is likely to be the reason for this. At the same time, Nvidia products with larger memory bus widths (and sufficient framebuffer sizes) show more respectable performance in comparison to the otherwise faster G92-based products.

Link to article this was taken from.

It's a well known that the nVidia 8800GT Excels at 4xAA, but when 8xAA is enable it kinda just dies.. at least, I actually read all the reviews, so I knew about it, seems some of you are a little unbelieving to this possibility..

Will try and find some other reviews about it's poor 8xAA performance if I can get a chance, but I'm a bit busy in work at the moment..

This post has taken over an 1 1/2hrs to write!!

 

[gurusan]

Well, how about some g80/g92 people run the Crysis benchmark @ 1680x1050+8xAA and some 38xx people do the same....

 

[Soul Rider]

Ahh, some investigation has provided some interesting results and possible reason for it...

The 8800GT 512 has a memory bandwidth of only 57GB/s, whereas the 3870 has a memory bandwidth of 72GB/s.

When 8xAA is enabled, it becomes too much for the bandwidth of the GT and it dies. This is confirmed by comparing the other nVidia scores around it:

The performance at high res & 8xAA of the 8800GT 512 is comparable with the 8800GTS320MB, but the 8800GTS 640 and the ultra's etc are not affected. On checking the figures, the 8800GTS 320 has very similar bandwidth figures to the 8800GT 512.

They both fail to handle 8xAA.

The other nVidia cards have a much higher bandwidth, and there is no significant failure at 8xAA.

Do people in these forums actually do any research themselves, or do they just make things up in an attempt to start an argument and disagree with someone who has done research.

TBH I always thought the OCUK forums were better than this, guess I was wrong..

 

[Soul Rider]

that is tosh, all of it. either you are saying the gefore 8 series drivers were perfect all along or they have been consistently crap. either way you are totally incorrect. the g80 came on leaps and bounds from when the cards were first released, thanks to drivers, so that blows your theory right out of the water. And id refrain from the 'only ati owners' comments as well, becuase that does nothing but stir up emotions. that is a discussion baord, not a playground.

Not what I said at all, so stop flaming and start READING.

I said that

We've all been with ATI before, we know, once a product beds in, the drivers make tremendous improvements. Something nVidia have never really been able to do. ATI seem to know how to continually tweak their hardware for better performance.

Now I shall break that down as you are unable to understand it.

Once a product beds in ATI are able to make very large consistent improvements over time to their products. While nVidia too make improvements with their driver releases, the improvements they make from drivers over the life of their cards is generally less than that of ATI. ATI seem to know how to continually tweak their hardware for better performance.

I can't believe I had to explain that? Now refute it with a valid sensible argument thankyou.

 

[oweneades]

Do people in these forums actually do any research themselves, or do they just make things up in an attempt to start an argument and disagree with someone who has done research.

TBH I always thought the OCUK forums were better than this, guess I was wrong..

No the problem here is that we here DO research into these cards and thats why we know these benchmarks are a pile of horse excrement.

There is no way that a HD3870 can suddenly be faster than a 8800 Ultra at high resolutions when 8 x AA is applied. A 8800 Ultra comes into its own at high resolutions with high levels of AA.

What makes the review even more smelly is that the HD2900XT a card which has proven in most reviews to still be quicker than the HD3870 thanks to its higher memeory bandwidth doesn't gain the same performance advantage.

 

[Soul Rider]

But the whole 2XXX series was dropped because of an inherent problem with it's ability to deal with AA. Thus the design of the board was changed.

Also you seem to be confused....

The Ultra beats the 3870 time and again at 8xAA. May I suggest you read the review and this whole thread. I am talking about the limitation in bandwidth on the 8800GT 512.

The Ultra is far superior in every result except the one on the first page of the review, so my guess is someone else doesn't know how to use the next button...

I wish people would actually look and read, then combine this information with what they already know, then look into it.

You obviously havent done the research you claim, nor have most of the other posters in the thread.

 

[willhub]

Yep Soul Rider these benchmarks cannot be trusted they aint even using the latest NV drivers listin to james.miller he speakers the truth and if 8xAA just kills the GT then I dont know what card I have in my PC then.....

Soul Rider the 2*** series were dropped because ATi needed a refresh and they aint exactly going to use the 2*** name again as people would not buy them half as much as they buy the 3870, the AA problems are still there with the 3870 the only difference is, is that they've done a refresh, lower process so it runs cooler, added GDDR4, DX10.1, PCI-E 2.0, SM4.0 and 256bit bus. If you look at benchmarks you can see the 2900XT even beats the 3870XT in a few benchmarks and they aint far at all, no drivers are going to give the 3870 aint dramatic boosts and leave the 2900XT were it was go figure.

I'd compare those results from Bit-Tech myself if I had the games they use with 4xAA.

 

[oweneades]

But the whole 2XXX series was dropped because of an inherent problem with it's ability to deal with AA. Thus the design of the board was changed.



You obviously havent done the research you claim, nor have most of the other posters in the thread.

If you had bothered to do any research yourself then you would know the design of the HD3870 is simply just a die shrink of the HD2900XT. The way it handles AA i no different than the HD2900XT, it relies on its shaders as it doesn't have a hardware MSAA unit unlike the 8800 Series. The HD2900XT should actually have an advantage at higher levels of AA thanks to its higher memory bandwidth but according to this review it shows no advante what so ever.

But of course I don't know what I am talking about

 

[willhub]

If you had bothered to do any research yourself then you would know the design of the HD3870 is simply just a die shrink of the HD2900XT. The way it handles AA i no different than the HD2900XT, it relies on its shaders as it doesn't have a hardware MSAA unit unlike the 8800 Series. The HD2900XT should actually have an advantage at higher levels of AA thanks to its higher memory bandwidth but according to this review it shows no advante what so ever.

But of course I don't know what I am talking about

He's only doing it because he has a 3870, I've had a 2900XT and 8800GT so I know which one is the best performer and the 8800GT was not a sidestep but was a large upgrade in terms of performance.

 

[J.D]

No the problem here is that we here DO research into these cards and thats why we know these benchmarks are a pile of horse excrement.

There is no way that a HD3870 can suddenly be faster than a 8800 Ultra at high resolutions when 8 x AA is applied. A 8800 Ultra comes into its own at high resolutions with high levels of AA.

What makes the review even more smelly is that the HD2900XT a card which has proven in most reviews to still be quicker than the HD3870 thanks to its higher memeory bandwidth doesn't gain the same performance advantage.

Maybe when it comes to the Ultra it seems a little bit wonky but I believe it about the 8800GT when 8xAA is applied. Read the article from Bit-Tech as well.

For you or ANYONE to doubt this without the hardware in front of you to test it is a great example that you don't know for sure but are just taking an educated guess from previous results. An educated guess will just not cut it here so unless you have the proof in the pudding then you calling the reviews BS is just how your post shall be taken. Sorry but how am I meant to believe you, or your thinking?. You're not alone here. Willhub (usual doomsday ATI carp from day one so has to be disregarded in any ATI/AMD talks and always will by me and a lot of people until he starts talking sense), RavenXXX who used to say AMD/ATI are the bees knees until he bought an Nvidia card, so that's another person who I wouldn't trust).

So if anyone who isn't unbiased and wants to test this out then let's put this to the test. I'm at a friends house just now but I have a 2900Pro and a 8800GTX at home and can try both but my resolution only goes to 2048x1536, plus I have no 8800GT which seems to be the card in question here.

I don't mean it in a nasty way Willhub and Raven but I've been reading your posts for a while now and everything I've said about you can be found in just about every AMD thread that's been since the 2900 release. Willhub = fair play, he had troubles from the start and a lot of the time wasn't getting the performances of other 2900's. Raven however, biggest turncoat I've seen. Even Tom|NBK who switched to the dark side didn't down ATI/AMD in any way except mark out the truth which is respectful and is why I will read his posts. Willhub had a bad time and only speaks of that even though other owners don't show the same attitude and this is why I'll never listen to anything AMD related when Will is involved. Raven was hilarious , read his posts about AMD and then see him turn that around quicker than lightning when he bought Nvidia so another persons posts to disregard.

 

[willhub]

I'm not downing ATi the 3870XT is a better card than the 2900XT and its what the 2900XT ahould of being but fact stands it still has AA problems. I can try Bioshock with AA if its works now and some other games like UT3, TDU, SEGA RALLY.

SEGA RALLY anyone?

Of course I need to change drivers first AA is borked. Whats best drivers ATM for 8800GT?

 

[J.D]

I'm not downing ATi the 3870XT is a better card than the 2900XT and its what the 2900XT ahould of being but fact stands it still has AA problems.

Yeah I know m8 but you always relate your experience in these threads and your performance was lacking compared to others when you had your 2900.

There is differences between the 3870 and 2900. Just look at the specs of both cards. It's just like the G92 is for the GT and new GTS. It's not just a die shrink it seems but I'm not saying that AA is totally changed but it might be a little bit better with this new board. It seems to be from the first reviews and not just the ones shown in this thread. If you look at older reviews you see that the 3870 is a little bit faster in some titles where as the 2900 is a little bit faster too but that was down to memory bandwidth. Obviously AMD found out what was best to be cut and made the changes accordingly. That's is why it was cut down to 256bit on the ring-bus and changed elsewhere. I know AMD haven't had the best year but I doubt they are going to make bad changes on the already tarnished product (2900) as they had more than ample time to try and sort some of the problems out.

We all know that ATI/AMD have the better drivers/ drivers team so how is it so crazy that they have started to show the improvements we've been waiting on all this time?. With ATI taking the new approach with the shader resolve who are we to say that they can't or haven't improved it?.

 

[Soul Rider]

Look, you are totally missing the point of this thread.

Can I spell it out.

I AM NOT SAYING THE 3870 IS BETTER, OR EVEN AS AGOOD AS, THE 8800GT.

THE GT PEE'S FROM A GREAT HEIGHT ON THE 3870.

WHAT I AM SAYING IS AT 2500x1600 RES WITH 8xAA, THEY ARE FINALLY SORTING OUT THE AA ON THE ATI CHIPS. ATI DO THIS SORT OF THING REGUARLY THROUGH SOFTWARE UPDATES.

@ Oweneades - Obviously you don't.

Yes the 3870 is just a die shrink of the 2900. I never contested this point.

You will also notice that Clock for Clock, the 2900 ****** over the 3870, again, this is not in dispute.

What is being shown is that ATI have finally sorted out their AA issues. If my card was named the 2950 as originally intended, this driver release would still have fixed the AA issues.

Can you lot get over comparing it with it's older brother, I know it's the same apart from a die shrink.

Jeez, people cannot help flying in and starting arguments without actually understanding the points being made.

Do you have a mic? If so when I get home from work I am willing to sit down and explain to you the point of this review.

If you also did more research into the background of this review site you would discover the fact that the review was carried out on Germany's BIGGEST AND MOST TRUSTED review site. There is no comparable site in the UK, we do not have a site that is as big and widely trusted in the UK as this site is.

Millions of people worldwide know this site, you do not, so it's all rubbish. Yeah, remind me to believe your every word in future blindly.

Have you done this test yourself? Has any other site done this test yet? Any which you trust? No. This is all we have to go on, but the trend with other sites, as noted in the bit tech review, is consistent.

So take your head out of your *** and learn something, and that is:

You know nothing, I know nothing, no-one else knows anything, and everything you thought you knew was out of date by the time you learnt it.

Remember that, and you'll be ok.

Also note, and this is important, the only card tested with the new drivers in that review was the 3870. They did not retest the 2900 or the 8800GT. Use the new and old review (that's why I posted both) to see the improvements in performance. Also, make sure you are reading the correct HD3870 results from the stock card, and not the Factory Overclocked PowerColour version, but of course, you wouldn't make that mistake would you???

 

[willhub]

Yeah I know m8 but you always relate your experience in these threads and your performance was lacking compared to others when you had your 2900.

There is differences between the 3870 and 2900. Just look at the specs of both cards. It's just like the G92 is for the GT and new GTS. It's not just a die shrink it seems but I'm not saying that AA is totally changed but it might be a little bit better with this new board. It seems to be from the first reviews and not just the ones shown in this thread. If you look at older reviews you see that the 3870 is a little bit faster in some titles where as the 2900 is a little bit faster too but that was down to memory bandwidth. Obviously AMD found out what was best to be cut and made the changes accordingly. That's is why it was cut down to 256bit on the ring-bus and changed elsewhere. I know AMD haven't had the best year but I doubt they are going to make bad changes on the already tarnished product (2900) as they had more than ample time to try and sort some of the problems out.

We all know that ATI/AMD have the better drivers/ drivers team so how is it so crazy that they have started to show the improvements we've been waiting on all this time?. With ATI taking the new approach with the shader resolve who are we to say that they can't or haven't improved it?.

I will do benchmarks on XP. But the only differences are SM4.0, DX10.1, 256bit, die shrink.

 

[J.D]

Soul rider. If people are willing to comment without the hardware in front of them that can prove differently then why even bother with the comments. Be graceful, look at their posts and disregard them as they have you. At least you have something of substance when all they have is experience which would mean absolutely nothing if a new driver is used that MIGHT have changed a few things.

If we can stop all this arguing and get some benchmarking done to try and prove this then we might get somewhere. We have 2 sites that seem to be respectable against, well nothing really so unless there is something to prove otherwise then why are we arguing?.

I will do benchmarks on XP. But the only differences are SM4.0, DX10.1, 256bit, die shrink.

I don't know if 1680x1050 with 8xAA will produce the results needed but we can only try .

What about heat, quietness of the card too?. Just a die shrink would only consist of the die being shrunk so can we please drop that one as it's not true although close to it.

 

[willhub]

heat and quietness are results of the die shrink.

 

[J.D]

heat and quietness are results of the die shrink.

Quietness LMFAO - NAH. Revised m8. Heat of course.

SM 4.1, 256 bit. C'mon .

Oh the G92 is a die shrink of the G80 then .

 

[Soul Rider]

I guess people don't realise that there is more to a graphics card than the GPU. The actual onboard bus architecture and memory do have a rather big say in the matter too, but hey...

 

[Cyber-Mav]

lets not forget the main reason for the die shrink, large reduction in power consumption.
the 2900 sucked up a lot more power than the gtx yet still performed a lot slower than the gtx.

the 3870 has a far more reasonable power consumption rate, its around equal to the 8800gt, but lets not forget that the 8800gt still manages to outperform the 3870 by around 15% while still sucking up the same amount of juice as the 3870. ati still haven;t got it right when it comes to performance per watt.

but this is not new to me at all, the last guy used to do exactly what soul rider is doing and try and find reviews and other material that showed the 2900 to be quicker than the 8800gtx. and that guy eventually ended up looking the fool of the forums when he switched over to a gtx.

guess its true what they say, nvidia users are looking forwards to new games, ati users are looking forwards to the next catalyst driver release......

 

[willhub]

Yes J.D the 8800GT is just an G80 die shrink, but I thought they added some extra units to give the performance lead over the GTS 640/320.

I know that the GDDR4 is something else new on the 3870XT and in some cases gives a boost in performance but the 3870 still has a lower bus I'm sure the 2900XT had a 512bit bus and that was also using some ring bus design?

Anyway I think those reviews are BS at this time especially that german one, I'd rather see results from people on this forum which means much more than some review site to me tbh.

guess its true what they say, nvidia users are looking forwards to new games, ati users are looking forwards to the next catalyst driver release......

That is true but at the moment NV users are looking forward to new driver releases since AA is borked in the latest ones.