4850 also perform faster than GTX285 in 1 or 2 games so does mean drivers need work?
Yes, some games still need work. Some games are far slower than they should be on the 4850 too. Ive known this for months actually.
Its not like I need to guess this given I submitted an extensive bug report to Terry Makedon (ATis Catalyst Maker). Anyone that has spent any reasonable time with the 5770 can clearly see performance is all over the place and isnt as consistent as it should be, especially with super-sampling.
But then this requires analysis beyond 3DMark fillrate tests from Tech-Report, LOL.
LOL> So cypress that has upgraded core with more cache, unchanged raster setup, etc over RV770 has problems with scheduling , caches, interpolators, raster setup? Then a card with weaker 4870x2 core somehow beats the better cypress core that is clocked higher with more memory bandwidth?
Again, the 5xxx series has changes to the hardware over the 4xxx.
Like? more cache? multi-GPU setup that is also weaker than a single card setup?
When you type responses, can you please actually make sure they make sense and are related to what you quote? Posting some cryptically irrelevant response doesnt make you look clever.
But then again 3dmark is based on a REAL game engine.
Not the fillrate test, which is synthetic by nature. No game operates like a fillrate test because otherwise it wouldnt be a fillrate test.
So a scheduler somehow better on the RV770 with lower cache vs cypress?
No, the scheduler could be different and hence requires different driver optimization to get optimal performance from it.
And what the hell are you trying to claim anyway? You keep saying drivers but there's no proof.
Actually theres plenty of proof. Like I said, I submitted an extensive bug report to Terry Makedon detailing my findings with evidence to back my claims. What tests have you run on the 5770 to conclude the drivers are 100% optimal, Azn?
Ive provided evidence that the drivers need work. You claim they dont, now provide evidence. Put up or retract your claims.
While there are plenty of examples of bandwidth making difference in case of 4870x2 vs 5870, 5770 vs 4890 and so forth.
Again, were talking about the effect of bandwidth on the 5770. Not the effect of apples with oranges.
Quite frankly lot of the graphs were done without AA. NFS is another game that is also shader intensive which has an edge on the 5770 as its shader has more caches with upgraded prefetching abilities.
Again Ill ask for you to explain pre-fetching in your own words, and describe how it correlates to the specific issue of the 5770s memory bandwidth.
Using big words doesnt mean you know what youre talking about.
Hardocp had a track records of not eliminating margin of errors with their FRAP testing procedures.
Describe to us in your own words how to eliminate the margin of error in their graphs.
Now describe in your own words how to do it to a single minimum data point the same data point you advocate as the end-all of performance - which doesnt even have a graph plot putting it into context.
It's obvious to someone who has done lot of testing.
LMFAO.
I don't have to accept all of your results but I can agree with some as different cards behave differently to your method of testing. That's must be it I have a biased agenda!
You accept the 8800s results because they back your claims. You dont accept the others because they dont, despite the methodology being the same. The problem isnt the benchmarks here, its the person looking at them and refusing to believe reality.
It's relevant because fillrate corresponds with bandwidth all cards to date.
Again, if we want to see the effect of bandwidth on the 5770, we test the 5770 with changed bandwidth. We dont check the 4870X2 for 3DMark fillrate tests from Tech-Report, LOL.
i'm telling you that all ATI had to do was change only few numbers to get same optimization as RV770.
Please provide evidence of your claims, or retract them. Code snippets from the driver code and/or commentary from the developer stating the drivers are completely optimal will suffice.
It's no different with Nvidia when G92 came about. you weren't doubting drivers then but why now?
Actually its very different, and someone with a basic understanding of the architecture would understand why, as would someone who has actually used the card in comparison to a 4850.
Cypress has precision enhanced to already their double precision math. It also supports fused multiply add instructions which benefit single precision applications. Inter process communication cache has doubled as well as data share 4x the cache of RV770.
This has what to do with memory bandwidth?
Perhaps it's the wrong word to use. But you do get 2 256bit when you add it up it's double the 256bit. Now I never implied bandwidth was shared between cards while these guys with 3 posts said it does like a raid array.
You dont add it up like that, otherwise my E6850 is a 6 GHz processor. While the aggregate bandwidth may be more than a single cards somewhat, the same applies to the core/shader too. Since both go up proportionally, if youre claiming one card is limited by bandwidth then two cards must limited by bandwidth in the same way.
Again, its been repeatedly pointed out to you that the 5770 has the same SP/bandwidth ratio as the 5870.
GTX260+ is a bandwidth saturated card. Core can't be limited as a card is based on a set core. Either it can be saturated or bandwidth limited. In case of 5770 it's bandwidth limited to a degree as more bandwidth would give much better results as shown in real benchmarks like 4890.
Yet again I see random English words that dont appear to answer the simple question I asked.
Answer the question Azn: do you accept my GTX260+s findings that showed the part is balanced, findings that had the same methodology as the 8800s, which you agreed with?
This is a very simple question so please answer it, and stop trolling.
Which I've also repeatedly explained to you as well.
Your explanation is wrong. Again, a fundamental statistics book may point you in the right direction.
You might imagine but it's reality where I'm sitting as I've done tests after tests to come up with the conclusion as xbit benchmarks only reinforce what I've been saying.
What 5770 test have you run, exactly?
What 5770 driver analysis have you done to conclude the drivers are 100% optimal, exactly?
How much time have you spent gaming on the 5770, exactly?
Which is that SP made biggest difference in games only to be corrected.
Im still waiting for retractions from you that:
1. The 5770 isnt primarily limited by bandwidth.
2. The core doesnt affect minimums as much as the memory.
3. An average doesnt correlate to a minimum.
4. The GTX260+ is a balanced part.
Your arguments were wrong on all four counts, so please retract them.
Again with the personal referendum.
Calling out your arguments is not personal.
Holes with proof and benchmarks to back them up I suppose.
You can lead a horse to water, but you cant make it drink. Weve shown you the results and explained them to you, but we cant do anything if you refuse to accept them.
Facebook
Twitter