Better 3870/3850/2900xt/8800gt reviews @ VR-zone

[Tatty_Two]

I think you guys are forgetting two important things.

First, there has been no driver release since the 3870 was released. The 7.10's only have basic support for the 3870, and it WILL get better with the next few sets of driver releases. It's always the same with ATi.

Second, can any of you actually get hold of an 8800GT? Nope...didn't think so.

I'm not an ATi fanboi, but the 3870 is cheaper, and from one review I read (posted on the front page) it has more consistent FPS than the 8800GT and it has better IQ IMO.

I can't wait to see the 512Mb 8800GT, but if they again limit stocks then how exactly are you going to get hold of one hmm?

Lol, half the people down my road have 8800GT's, and apart from the initial driver release there has been no more (less beta) releases for the 8800GT, dont forget the highly anticipated 2900pro came out not so long ago and apart from an initial and somewhat small release in the UK it was also damn hard to get your hands on one of them for a time.....Ohhhhh and the GT is only a bit more expensive because of retailers not having enough of them, before Christmas I would guess that the will drop £25 in the UK which almost puts the 2 cards side by side. The point is here that unlike the HD2900 that was released 6 months after the G80 which pretty much had developed as much as possible by then so only the 2900 moved forwards and after initial dissapointment did brilliantly through good driver releases. On this occasion both cards are pretty much brand new, both have the opportunity to compete in driver development but the 8800GT has in places a 15-20% headstart, at least in performance terms, if ATi catch up and overtake them and in 6 months time the 3870 is the faster of the 2 cards (which I suspect is unlikely looking at the differences in Architecture) then it really does go to show that ATi's driver development is wayyyyyyy better than NVidia's!

 

[DarkMatter]

Lol, half the people down my road have 8800GT's, and apart from the initial driver release there has been no more (less beta) releases for the 8800GT, dont forget the highly anticipated 2900pro came out not so long ago and apart from an initial and somewhat small release in the UK it was also damn hard to get your hands on one of them for a time.....Ohhhhh and the GT is only a bit more expensive because of retailers not having enough of them, before Christmas I would guess that the will drop £25 in the UK which almost puts the 2 cards side by side. The point is here that unlike the HD2900 that was released 6 months after the G80 which pretty much had developed as much as possible by then so only the 2900 moved forwards and after initial dissapointment did brilliantly through good driver releases. On this occasion both cards are pretty much brand new, both have the opportunity to compete in driver development but the 8800GT has in places a 15-20% headstart, at least in performance terms, if ATi catch up and overtake them and in 6 months time the 3870 is the faster of the 2 cards (which I suspect is unlikely looking at the differences in Architecture) then it really does go to show that ATi's driver development is wayyyyyyy better than NVidia's!

Or waaaaay worse. It depends on how you look at it. Maybe Nvidia does it better from start, while Amd needs some bits of time to make them work properly. With 320 SP the card shouldn't perform better? Just kiddin. Well kinda.

 

[Tatty_Two]

Or waaaaay worse. It depends on how you look at it. Maybe Nvidia does it better from start, while Amd needs some bits of time to make them work properly. With 320 SP the card shouldn't perform better? Just kiddin. Well kinda.

The SP's are not the key to the performance here, if that was the only factor they would be well over twice as quick as the 8800Ultra, half the Sp's running twice as fast can make up for that easily but it's more complicated than that with this architecture, does it not kind of speak for itself at the moment seeing as the 8800Gt has way less than half the SP's but at the moment at least is quicker?

 

[cefurkan]

more sp doesnt mean anything

since the whole architecture is different

 

[DarkMatter]

Eeeeh!! Calm down about SPs!!
Well I suppose I deserve this for trying to make a cheap joke on this...

But you know what I think about SPs, and that on the same architecture they do matter, more than anything else.

EDIT: Please visit my thread about how a midrange card should be and post.

http://forums.techpowerup.com/showthread.php?t=44564

No one except Wile has cared about it. It is my firts thread.

 

[Tatty_Two]

Eeeeh!! Calm down about SPs!!
Well I suppose I deserve this for trying to make a cheap joke on this...

But you know what I think about SPs, and that on the same architecture they do matter, more than anything else.

EDIT: Please visit my thread about how a midrange card should be and post.

http://forums.techpowerup.com/showthread.php?t=44564

No one except Wile has cared about it. It is my firts thread.

Then on this one we will have to disagree, it's not the quantity that matter as much as the speed they run, the quicker...the more cycles blah blah, if your theory was correct the 2600 would be quicker then the 8600 blah blah.....hopefully I have bored you to death now and U are at least asleep!

 

[DarkMatter]

Then on this one we will have to disagree, it's not the quantity that matter as much as the speed they run, the quicker...the more cycles blah blah, if your theory was correct the 2600 would be quicker then the 8600 blah blah.....hopefully I have bored you to death now and U are at least asleep!

As I said it only apllies to the same architecture. And about quantity or speed, it depends on the aplication but total power (SP number x clock) is what most matters overall
. Some games will benefit from SP quantity more and some will benefit from higher closcks. If you ask me, I would say quantity matter more. Just look at 8800GT/GTX comparison.

1500 x 112 = 168000
1350 x 128 = 172000

But the difference between them, even on low resolution and AA (where memory bandwith and quantity, and fill rate don't matter as much), seems to be higher than that. But it could take hours to discuss this, so let's move on, we both know what the other thinks about, after all.

 

[Tatty_Two]

As I said it only apllies to the same architecture. And about quantity or speed, it depends on the aplication but total power (SP number x clock) is what most matters overall
. Some games will benefit from SP quantity more and some will benefit from higher closcks. If you ask me, I would say quantity matter more. Just look at 8800GT/GTX comparison.

1500 x 112 = 168000
1350 x 128 = 172000

But the difference between them, even on low resolution and AA (where memory bandwith and quantity, and fill rate don't matter as much), seems to be higher than that. But it could take hours to discuss this, so let's move on, we both know what the other thinks about, after all.

Yup but the problem with having 320 Sp's (that incidentially cannot render on each cycle) and that dont have (ideally) enuff ROP's and Texture units to feed them means that their potential is being throttled, now I know in comparison to the 8800GTX they offer brilliant performance > price but ATi could really have killed the competition for years with say 24 ROP's in there, all I am really saying I spose is that for all those SP's, the architecture is not efficient enuff to make the best use of them where as, for NVidia with so few SP's, the architecture DOES make the best use of them......well kind of.

 

[SSXeon]

http://www.guru3d.com/article/Videocards/472/10/

didnt want to make a whole new thread, but this review is the best. Check out crysis vs 700/2000Mhz 88GT

 

[Silverel]

http://www.guru3d.com/article/Videocards/472/10/

didnt want to make a whole new thread, but this review is the best. Check out crysis vs 700/2000Mhz 88GT

Not a problem, the premise of the thread was having to reviews as of yesterday. The first one found was at LegionHardware, which was all f*cked up, and I found this one which was piles better in comparison.

Nice review. btw. Love the Crysis comparison, I probably won't play it for another year though.

I do my gaming at 12x10 for fps's and 16x12 for less intense games. I'm about a few years behind in the gaming curve, simply because I'm not going to spend that much on the "brand new shiny". The 38xx series will suit me just fine, and I've been waiting a good while for this level of performance to drop down in cost. Been hanging on to my 9600 for far too long.

The 2900pro made me spin a little, but I couldn't get a hold of one. The 8800Gt kinda killed my buzz over AMD at first when I saw the crazy numbers. Now that these two are out, they're cheap, comparable, and will work for me. I'll probably grab the 3850 and slap on some pretty aftermarket cooling, then OC the piss out of it.

 

[DarkMatter]

Yup but the problem with having 320 Sp's (that incidentially cannot render on each cycle) and that dont have (ideally) enuff ROP's and Texture units to feed them means that their potential is being throttled, now I know in comparison to the 8800GTX they offer brilliant performance > price but ATi could really have killed the competition for years with say 24 ROP's in there, all I am really saying I spose is that for all those SP's, the architecture is not efficient enuff to make the best use of them where as, for NVidia with so few SP's, the architecture DOES make the best use of them......well kind of.

You don't read what I write don't you? What I say is that efficient or not shaders are responsible of graphics cards performance. :shadedshu
I know very well r600 architecture, it has one scalar and one vec4 units per SP, leaving it with only 64 SP in the worst case and having a (more realistic) max of 256 SPs when vec4 is used. I know that this equates to the 8800 Ultra power, because the latter runs twice as fast. And I know that r600 isn't as fast as 8800 Ultra, because it's impossible to make all shaders work 100% all the time. I know all that.

I said that given an architecture, be it G80 or R600, the more the shaders faster the card. That shaders affect performance more than ROPs and even TMUs. Or do you explain 2900pro - 2900gt difference based on memory bandwidth? I don't and that is the only difference between them, that and that the GT only has 240 SPs. I have seen many reviews and the differece between them is around 25% average. Go see some benchmarks if you don't believe me. I have found this on a quick search, it's about 8800GT but both are there. Look at any game and resolution. 2900gt is always around 75% of the Pro.

http://www.firingsquad.com/hardware/nvidia_geforce_8800_gt_performance/page14.asp

Now do 240/320*100. Interesting? 25% Performance penalty, same ROPs, same TMUs, same clocks, 75% SP quantity... No... It' not possible... 75% performance and 75% SPs. What are the chances? Yeah there's also the lower memory and memory bandwidth. But, what a chance! And as I said before, it happens the same with 8800GTX and GT, where gtx has much bigger memory bandwidth and fillrate. And only similitudes come in SP power, and yes TMUs. But shouldn't G92 perform much better with twice the texture address units if bottleneck was on texturing?

Conclusion: if 8800GT had had 384bit interface and 24 ROPs, it wouldn't have performed better. And 3870 with so neither. Both Nvidia and Amd know very well what they make, it's not like they go and say: Ey let's make a card with this and that, just to see how it performs...

See you!

 

[PVTCaboose1337]

Man I am so excited about the potential of this card! If we had good drivers, who knows how good it would be?!?!

 

[DarkMatter]

Tatty_One. Sorry because I'm being annoying, but I still have something to say to try prove my point.
Let's see it from manufacturing point of view. G80 as an example:
You decide after researching that the chip you are making is going to be fast and well balanced at around 128 SP, 32 TMU and 24 ROP. Yeah those are 8800gtx specs, just imagine they were orientative numbers when designing the chip.
Those units are basically Floating Point units aconditioned to an especific job. So they aren't going to be very different between them in shape, and if there is a difference (that there is) its going to make SPs more complicated (and bigger) than TMUs and ROPs, because they need to perform more different tasks. Right?
But let's consider them similar. So 128+32+24= 184 units. 128 out of 184, 70% of your die size is going to be dedicated to SPs, 17% to TMU, and 13% to ROPs.
Now you know that one of those is going to be the bottleneck, because is something inherent to complex systems. Wich of them do you want to keep out of being bottlenecked by other components at all cost? And in wich can you add more, while not adding too much in die size? The answer is that simple: you don't want 30% of your transistors bottlenecking the others. So if more power was needed there you would add more.

 

[SSXeon]

Yeah i think i can run tri-fire with this board, around christmas i think im going that way. Never did multicards before, but i need all i can get for crysis. The card is also faster per clock in DX10 from what Dave "wavey" said.

If anyone wants to read a some what technical preview here ya go:

http://www.beyond3d.com/content/reviews/50/

 

[deddyeka]

seems that 8800 is more STABLE speed overall

does it need to add like PHYSX CARD ?