eidairaman1Beyond Shaders, ROPs, TMUs there is the Fact of the Basic Transistor Density.

Beyond all that...developer-level optimisations for games and 3D Apps. The basic architecture of a GPU has diversified very much after the advent of DX 10.

lemonadesodaGiven the same architecture, higher clocks, and more shaders, I think these are the performance implications:

1./ Broadly similar performance at standard resolutions e.g. 1280x1024 and with no AA FSAA effects since no architectural changes
2./ General improvement in line with clock-for-clock increases 10-20%
3./ The increase to 32 TMU will mean that the cards wont CHOKE at higher resolutions. It will be able to handle 1920x1200 without hitting the wall
4./ Currently you can dial up 4x AA without any performance hit. With the extra shaders you can do the same at 1920x1200 now
5./ With the extra shaders, you will be able to dial up 8x or 16x at 1280x1024 without a significant hit.
6./ The GPU will run hotter and require more power
7./ Compensated by using GDDR5 memory that will require less power and run a bit cooler

Net net... get the GDDR5 model.

Will there be a "jump" in performance like we saw between the x19xx series and hd38xx? No.

Want to place a bet? No, seriously, that almost made my day.

So according to you what is what it gives more performance if Gflops, texture fill-rate and memory bandwidth don't increase anything??? Does performance come out off of thin air?

You are not very versed at GPU architectures, are you?

lemonadesodaGiven the same architecture, higher clocks, and more shaders, I think these are the performance implications:

1./ Broadly similar performance at standard resolutions e.g. 1280x1024 and with no AA FSAA effects since no architectural changes
2./ General improvement in line with clock-for-clock increases 10-20%
3./ The increase to 32 TMU will mean that the cards wont CHOKE at higher resolutions. It will be able to handle 1920x1200 without hitting the wall
4./ Currently you can dial up 4x AA without any performance hit. With the extra shaders you can do the same at 1920x1200 now
5./ With the extra shaders, you will be able to dial up 8x or 16x at 1280x1024 without a significant hit.
6./ The GPU will run hotter and require more power
7./ Compensated by using GDDR5 memory that will require less power and run a bit cooler

Net net... get the GDDR5 model.

Will there be a "jump" in performance like we saw between the x19xx series and hd38xx? No.

Oh boy, never have i seen a guy knowing so little about GPU architectures making such a long and bold (and completely wrong) statement.
1- If the CPU can deliver, fps will always increase. Since every aspect of RV770 is almost 2x that of RV670 theoretically it can do 2x the fps. What you say is only correct if the CPU is not fast enough but that's not the point here. Secondly, where do they state that it will (or will not) have completely same architecture?
2- Indeed clock increases!
3- First of all, the 3870 isn't hitting a wall at 1920x1200. Actually it's gaining a lot of ground at 2560x1600. TMU's don't have anything to do with the resolution btw. The increase in TMU's will help a lot with shaders and texture lookups.
4- Enabling 4x AA has a performance hit at any resolution. Fact! Only when your CPU is already too slow to deliver enough fps, only then you won't see a performance hit.
5- Nonsence.
6- Obviously. But anything beyond that is guessing. 55nm has matured a lot over the last year.
7- Compensated by GDDR5? Obviously you love guessing.

Performance increase over RV670? Almost double if not more.

lemonadesodaWill there be a "jump" in performance like we saw between the x19xx series and hd38xx? No.

Everything about the 4xxx series suggests it will be a dramatic performance increase. anything else I have to say is generally covered in the two posts above me and does not need to be elaborated on. :rolleyes:

OK, shithot DarkMatter, if you are going to throw personal insults around. Show how confident you are in your 2.2x performance. Put your money where you mouth is. This is a PUBLIC CHALLENGE.

Let's take a CPU with a GDDR4 HD 3870 at stock. Say Q6600 at 3.0Ghz. Run 3dmark06. Record the result.

Now lets put a GDDR5 HD 4870 in there, at stock. Same Q6600 at 3.0Ghz. Lets run 3dmark06 again. Record the result.

I bet you $100 the result is nowhere near 2.2x. In fact, I'll give you the odds not even at <2.0, but at < 1.7. If it's less than 1.7, I win. If it's more than 1.7, you win. Take on the bet, boyo. If you dont, then take back your personal insults and lick my boots.

This bet is also offered to the Belgian sprout from antwerp. Dont be a chicken.

lemonadesodaOK, shithot DarkMatter, if you are going to throw personal insults around. Show how confident you are in your 2.2x performance. Put your money where you mouth is. This is a PUBLIC CHALLENGE.

Let's take a CPU with a GDDR4 HD 3870 at stock. Say Q6600 at 3.0Ghz. Run 3dmark06. Record the result.

Now lets put a GDDR5 HD 4870 in there, at stock. Same Q6600 at 3.0Ghz. Lets run 3dmark06 again. Record the result.

I bet you $100 the result is nowhere near 2.2x. In fact, I'll give you the odds not even at <2.0, but at < 1.7. If it's less than 1.7, I win. If it's more than 1.7, you win. Take on the bet, boyo. If you dont, then take back your personal insults and lick my boots.

This bet is also offered to the Belgian sprout from antwerp. Dont be a chicken.

Well obviously in this situation you will be right since a Q6600 at 3Ghz is nowhere near fast enough for a two fold increase (it will be a bottleneck). And you made your point, you sir are a moron. 3DMark06 also includes the performance of the CPU in the final score. So when you keep the same CPU, you can't expect 2x the performance in 3DMark since only the GPU is faster.
How 'bout we make the same bet but let's take a real game. Let's say Crysis since that's the hardest game around and we'll use 1680x1050 4xAA/16xAF (very high detail level). How 'bout that? In the Ice level a 3870 gets around 6 fps. 6 x 2 = 12! OK?

MrMilliWell obviously in this situation you will be right since a Q6600 at 3Ghz is nowhere near fast enough for a two fold increase (it will be a bottleneck). And you made your point, you sir are a moron. 3DMark06 also includes the performance of the CPU in the final score. So when you keep the same CPU, you can't expect 2x the performance in 3DMark since only the GPU is faster.
How 'bout we make the same bet but let's take a real game. Let's say Crysis since that's the hardest game around and we'll use 1680x1050 4xAA/16xAF (very high detail level). How 'bout that? In the Ice level a 3870 gets around 6 fps. 6 x 2 = 12! OK?

My 3870 gets about 6fps on the level at 1920x1200 no AA or AF & my X2 gets about 15fps. If the 4870 gets 12 & 4870X2 gets 30 then I'll eat my 3870 :D - I really want to eat my 3870 :D

MegastyMy 3870 gets about 6fps on the level at 1920x1200 no AA or AF & my X2 gets about 15fps. If the 4870 gets 12 & 4870X2 gets 30 then I'll eat my 3870 :D - I really want to eat my 3870 :D

What do you want to go with that?

MegastyMy 3870 gets about 6fps on the level at 1920x1200 no AA or AF & my X2 gets about 15fps. If the 4870 gets 12 & 4870X2 gets 30 then I'll eat my 3870 :D - I really want to eat my 3870 :D

i Quote you on that

Morgothi Quote you on that

Nice, very nice. Now I know I'm gonna have to eat it :cry:

You know I'm thinking someone is definitely going to be eating their card in this situation. On a side note anyone want to buy up a 3870 ;)

lemee guess, switchin to nvidia right

magibegYou know I'm thinking someone is definitely going to be eating their card in this situation. On a side note anyone want to buy up a 3870 ;)

lol, I'm preparing the stew in my sig for the 3870 as we speak :D

Please do not insult other members. If you disagree with the opinion of another member, explain why in a polite and reasonable manner. Insulting others only creates an unpleasant atmosphere in the forums. Please do not create competitions for personal vendettas either.

Thanks

This is a news story treat it as such, we do not like flame wars and insult matches in any section, especially the news section. I have no problem closing this discussion if I have to revisit this thread again for negative purposes.

lemonadesodaOK, shithot DarkMatter, if you are going to throw personal insults around. Show how confident you are in your 2.2x performance. Put your money where you mouth is. This is a PUBLIC CHALLENGE.

Let's take a CPU with a GDDR4 HD 3870 at stock. Say Q6600 at 3.0Ghz. Run 3dmark06. Record the result.

Now lets put a GDDR5 HD 4870 in there, at stock. Same Q6600 at 3.0Ghz. Lets run 3dmark06 again. Record the result.

I bet you $100 the result is nowhere near 2.2x. In fact, I'll give you the odds not even at <2.0, but at < 1.7. If it's less than 1.7, I win. If it's more than 1.7, you win. Take on the bet, boyo. If you dont, then take back your personal insults and lick my boots.

This bet is also offered to the Belgian sprout from antwerp. Dont be a chicken.

First of all I didn't insult you anywhere. I said you are not versed at GPU architectures. We have the proof, it's called "post #200". I'm not very versed at genetics, I'm not very versed at solfegge, hell I could even go and say that I'm not versed at computers compared to what it's left to know. And you know what? I am not insulting myself, because that's the truth. You have a problem if you think that you are versed at computer architectures after what you said. You have a bigger problem if you feel offended when they say you are not. You have even a bigger one if you take such little critics as insults. I really hope you can resolve them.

Second, a bet involving money it's stupid in the net, specially since I live in Europe. Definately I'm not going to give my account number to anyone that I don't know. And as they have already told you, you need faster CPUs and newer games to see the difference. When 6600GT, 7600GT and other midrange cards were launched they offered almost 80% the performance of their high-end cousins, go look if they perform even 50% now.

And finally, I said that 2,2X is the peak power the HD4 series have compared to HD3 when looking at those specs. There are other things to take into account. In fact, I implied a 2x improvement, while you said 1,1x. Middle ground for that is 1,5x. If you really want to place a bet (involving our prestige and honour, I already said I won't exchange money in the net with you. Also I don't want to be a thief robbering your money :p) let's do it on Crysis 1920x1200 4x AA or other new games that have not been released yet, and in an overclocked Quad at 3,6++ Ghz. They have already told you why you are not going to see the improvement on 3DM06 on a 3Ghz...

With the above conditions, if the performance increase is more than 50% I win, if it's less than that you win. The loser will have to show as his avatar whatever the other wants.

EDIT: BTW this bet is if you want to do it at launch day. If you want to wait 6-9 months (until new games, CPU, chipsets, etc are launched) I increase that number to 2x the performance.

And we have demonstrated proof in your lack of diplomacy. And perhaps I over-reacted: Accept my apology.

Not withstanding that, at no point did I say the gains would be limited to 1.1x. Point 2 refers to the gains associated with clock increases. Point 3 refers to 2x performance on texture bound resolutions, like 1920x1200 and higher. Point 5. refers to a application specific improvement associated with AA and FSAA.

Lets sit back with a beer and see how performance pans out. The challenge is 1.7x. If performance is >1.7x, I'll open a beer in your name and drink it with pleasure. And vice-versa. But the tool is 3dmark06. And it will be the same CPU. I'll only be looking at the combination of "SM2.0" + "SM3.0" scores, excl. the CPU score. And no it will NOT be a 1920x1200 test, but the regular demo test on 3dmark06. The 1920x1200 problem which was very clearly identified as being the #1 objective that ATI was trying to solve with the 32 TMU, is covered in my points 3. and 4.

If you misunderstood my original 7 points, that's OK. Perhaps it wasnt clear. But better to say, OK, now I understand what you mean, than to continue this "you dont know anything about xyz", or, "you've got a problem...". It is offensive language. And whether you use it on TPU, or with your friends, or at work, there will be people offended, whether they tell you or not. It's not a good way to start a dialog, let alone, cooperation. And that's what the TPU community is about.

Let's respect Thermo's request to keep flaming off the board. I'll say nothing more about it. Take it easy.

lemonadesodaAnd we have demonstrated proof in your lack of diplomacy. And perhaps I over-reacted: Accept my apology.

Not withstanding that, at no point did I say the gains would be limited to 1.1x. Point 2 refers to the gains associated with clock increases. Point 3 refers to 2x performance on texture bound resolutions, like 1920x1200 and higher. Point 5. refers to a application specific improvement associated with AA and FSAA.

Lets sit back with a beer and see how performance pans out. The challenge is 1.7x. If performance is >1.7x, I'll open a beer in your name and drink it with pleasure. And vice-versa. But the tool is 3dmark06. And it will be the same CPU. I'll only be looking at the combination of "SM2.0" + "SM3.0" scores, excl. the CPU score. And no it will NOT be a 1920x1200 test, but the regular demo test on 3dmark06. The 1920x1200 problem which was very clearly identified as being the #1 objective that ATI was trying to solve with the 32 TMU, is covered in my points 3. and 4.

If you misunderstood my original 7 points, that's OK. Perhaps it wasnt clear. But better to say, OK, now I understand what you mean, than to continue this "you dont know anything about xyz", or, "you've got a problem...". It is offensive language. And whether you use it on TPU, or with your friends, or at work, there will be people offended, whether they tell you or not. It's not a good way to start a dialog, let alone, cooperation. And that's what the TPU community is about.

Let's respect Thermo's request to keep flaming off the board. I'll say nothing more about it. Take it easy.

If there was really something offensive, then sorry. It must be something related to the language, something lost in translatioon, since I don't see any offensive language in what I said in my first reply. But I apologize if there was something offensive there. It would help me a lot if you tell me what exactly was offensive and an insult though. I did was offensive in the second, but only because you directly insulted me before.

But if you are talking about me saying you don't know about GPUs, if that is what you are taking as an insult, then I will take my apologies back. That's not an insult nor offensive and I am definately not going to say sorry for that, considering your reaction. It's just not offensive, I explained that in my previous post. There are lots of things that I don't know and I will never take as an insult if someone tells me so. You are demostrating you don't know about this, mate, and you are being arrogant by acting like a victim and taking offense for that. There's nothing to (miss)understand on your statements, they are just wrong. I'm trying to say this kindly, learn how a GPU works and then we'll discuss if those improvements will yield any gains. Some of the points could be true if they had only improved shaders and kept the rest as is, or if they only improved TMUs, but since they have improved both, plus the bandwidth enough to feed everything well your points are just wrong.

Just to point one of the things you learnt wrong. TMUs load and filter textures. They do their work on pixels. It doesn't matter if the next pixel is from the same frame or the next, it's just the next pixel. For them doing 16x16 pixels at 20 frames is the same as doing 32x16 at 10 FPS. They are just doing their work on 5120 pixels/second. Double the number of TMUs (or double the clock) and you can do either double the frames at same resolution or double the resolution at same frames. It doesn't exist anything like "texture bound resolution". Exactly the same applies to shader processors. Double of their power gives exactly double the performance (for that stage of the graphics pipeline). If we have double the power in every stage, as is the case here, except on pixel-fillrate (ROPs), you will get double the performance.
Now if you know what ROPs do, you know that since Ati does AA with shaders, the only job that ROPs have to do is blend the different fragments together (sub-pixels, which are calculated in the SPs using the data fetched from textures), and that job is only related to the resolution and the number of fragments. RV670 and G92 have demostrated that the bottleneck was not in ROPs. Specially G92 has demostrated this, because it does AA in ROPs (it's a lot of work being done there), and even though fill-rate is smaller than on RV670, G92 is a lot faster. Ati offloads AA work from ROPs meaning that there's still more room. It's difficult to know if a bottleneck occurs on ROPs in an architecture that will relegate so many things to shaders, but it's common sense they wouldn't make all other parts double as fast, just to let this one be a big bottleneck. They have those things resolved before launch.

My first calculations are based on all that and have their logic based on the graphics pipeline. Your statements don't have any sense, they are not based on the reality of how a GPU works. I didn't want to be offensive when I said you didn't know about GPUs, I still don't. We don't have to know about everything in this life, but if we don't know something, we don't know, that's all, we don't have to act as if we knew and when they prove us wrong act as a victim. That is not the way to go. That's what I thought you were doing. If you are not doing that consciously, I apologize. And I'm going to apologize in advance just in case this post is also offensive to you. I'm not trying to offend you, believe me, I just think you don't know enough about what I explained above and that's all.
Let's forget about this until we can compare the cards. :toast:
But not in 3DM06, it's the worst aplication you can use to know the power of a card nowadays. Vantage maybe. And definately not in a 3 Ghz bottleneck...

wow hoe did i miss this...

well il be getting a 4870x2

to simplify the matters between both of you, drop it and Kiss and Makeup.

On a side note you said you needed a quad core clocked to 3.6ghz with benches and such in crysis, i can deliver those benchmarks, just send me a 4870 when it comes available ;)

eidairaman1to simplify the matters between both of you, drop it and Kiss and Makeup.

gah, you're trying to make it worse ;)

On a lighter note, ATI is definitely not pulling any punches with the 4870 & X2. 1 & 2 GB of GDRR5, uber high gpu clocks, independent shaders, double the TMUs, etc...what is this stuff coming to. I don't feel like munching on my 3870 today as I did the other day but those specs really got me wondering why would they finally want to give out something thats impressing me so much on paper. Oh well I need to stop b4 it sounds like I'm complaining :D

lemonadesodaOK, shithot DarkMatter, if you are going to throw personal insults around. Show how confident you are in your 2.2x performance. Put your money where you mouth is. This is a PUBLIC CHALLENGE.

Let's take a CPU with a GDDR4 HD 3870 at stock. Say Q6600 at 3.0Ghz. Run 3dmark06. Record the result.

Now lets put a GDDR5 HD 4870 in there, at stock. Same Q6600 at 3.0Ghz. Lets run 3dmark06 again. Record the result.

I bet you $100 the result is nowhere near 2.2x. In fact, I'll give you the odds not even at <2.0, but at < 1.7. If it's less than 1.7, I win. If it's more than 1.7, you win. Take on the bet, boyo. If you dont, then take back your personal insults and lick my boots.

This bet is also offered to the Belgian sprout from antwerp. Dont be a chicken.

3dmark=utterly useless for anything but comparing tweaks on the same system and as a stab test for overclocks, 3dmark is SYNTHETIC and is only tauted by people on forums to show how big their epeen is, find a real test like some REAL GAMES....

Megastygah, you're trying to make it worse ;)

On a lighter note, ATI is definitely not pulling any punches with the 4870 & X2. 1 & 2 GB of GDRR5, uber high gpu clocks, independent shaders, double the TMUs, etc...what is this stuff coming to. I don't feel like munching on my 3870 today as I did the other day but those specs really got me wondering why would they finally want to give out something thats impressing me so much on paper. Oh well I need to stop b4 it sounds like I'm complaining :D

its a psychological tactic

oh btw, if you wana try and say that gflops and fill rate mean all

google "tomshardware a speedy tiler" and read it, its old but it shows that a cards theoretical numbers mean dick when compared to its acctual numbers, the kyro2 matched its numbers 100%, the cards from other makers fell far short due to memory bandiwth mostly :)