BFG has been incredibly patient and polite to you, but I don't suppose you'd ever admit that you're the one who has been getting abusive and emotional, would you? You haven't actually done any benchmarking of a 5770 yet you have the gall and arrogance to call BFG a 'shit author', simply because his testing doesn't fit with your argument.
Polite? Making personal insults is polite I suppose. :/
His testings methods only showed average frame rates and doesn't even compare to other cards before it to come up with a conclusion and I've asked him simply to investigate further with minimum frames as well as average. He doesn't seem to see all the angles where I'm sitting from. Does that make him a shit author? Did I call BFG a shit author personally?
Do you actually need to benchmark 5770 to get a grasp how A functions with B when other cards before it behave the same way? Isn't this same reason why I was able to tell how BFG's ultra was going to perform with core/sp/bandwidth even though I've never tested it? or is that just a lucky guess? You tell me!
1) You have not produced a single shred of evidence that the 4890 is faster than the 5770 entirely due to memory bandwidth. Evidence means that you can demonstrate that it is the case, not simply point to the fact that the 4890 has higher minimums than the 5770 and just repeating that because this is the case, the difference must be due to memory bandwidth. Can you see that you haven't actually eliminated any other possibilities there?
Nobody is disputing that the 4890 doesn't have much better minimums generally, but they are disputing your argument that because this is the case, it must be due to memory bandwidth, and can't be due to anything else. You haven't actually eliminated drivers or architectural problems, you have just thrown them out because they don't suit your arguments. I have to ask, but you don't happen to work in a climate change research centre do you?
I suppose all the benchmarks shown in this thread doesn't count as evidence. I gave plenty of examples within same architecture to show how bandwidth effects minimum frame rates but that's not evidence enough according to you.
Now now. there you go putting words in my mouth. "entirely" is a strong word I never used to describe what was discussed.
They are disputing it's not bandwidth but "entirely" on drivers that is the cause which I disagree with. You in the other hand said it could be bandwidth much later in your post which is quite clever of you only to respond me with this post.
2) Your comparisons between the overclocked 4850 and 4870 are strawmen, because the 5xxx cards are a quite significantly revised core, with some quite meaningful changes to it at a hardware level:
http://www.anandtech.com/video/showdoc.aspx?i=3643&p=5
But of course it's strawman. Any example I put out is and zealot one at that. :/
Never mind bandwidth behaves similar within all GPU architecture up to this point but hey don't let the factual information stop you calling it a strawman.
You have argued earlier that the 5770 is faster (both minimum and avg) than the 4890 in some games due to architectural changes (NFS: Shift and STALKER I believe), so you look hypocritical arguing that they're essentially identical. Those must be reasonably significant changes to allow the 5770 to obliterate the 4890 at a memory bandwidth hungry 2560x1600 with 4xAA/16AF...with the 5770 delivering much better quality AF as I understand it?
Why would that be hypocritical? When most games do not show this behavior to SP? There are exceptions though and I never ruled that out. The change are very minor at best and showing just that in those 2 games which doesn't bottom out minimum fps to unplayable scenarios. Who buys a 5770 to play 2560x1600? That's kind of an absurd observation for you to mention it.
In case of Juniper it supports fused multiply add instructions which benefit single precision applications which doesn't even support double precision math BTW which RV770 supports. Inter process communication cache doubled as well as data share quadrippled the cache of RV770 to improve shader performance but far as fillrate and bandwidth behaving hasn't changed. If anything cypress has slight texture fetching abilities as L1 cache has been doubled on the cypress in its memory controller. I'm not too sure about juniper as this chip is only half the RV770 256bit bus which pretty much means it has equal amount of L1 cache since Juniper is 128bit.
You don't know that any one of the considerable number of changes mentioned couldn't be causing an issue with current drivers (or indeed, with any drivers, it might be an irrepairable architectural problem). People who write programs are only human and it is to be expected that there will be issues that need to be ironed out in the relatively early days. Bugs in hardware happen, although they are not always fatal.
If you call that a considerable amount I wonder what you would call an entirely different architecture like R500 to R600? Nothing much has changed except more cache similar to E6600 to E8400. Obviously you can't seem to understand that or see it for that matter.
Number of changes? Now let's look back at RV670 vs RV770 or G80 to G92? Now why didn't BFG or YOU mention drivers during those times? Was it because the performance was better than previous iterations with more or less GPU?
You come accross as a zealot because you seem to have decided that it has to be memory bandwidth, it absolutely possibly cannot be anything else.
BFG says there's no way it could be bandwidth but drivers. Yet I don't hear you yelling zealot to BFG. It seems you made up your mind before the first post as you just agreed with BFG from the very start. Could this be favoritism or hypocritical? I think it's both.
3) BFG has indicated that his minimum FPS is consistent with his average FPS, and thus his results that show the 5770 a affected to a greater extent by core than by mem bandwidth. Will you accept that you are wrong if he posts this?
BFG says a lot of things without even really testing things through and just believe whatever web site tells him in lot of these cases. His whole sleuth of benchmarks on bottleneck with his ultra was the result of our arguments. He seems to be asking the wrong questions to come up with his conclusions although it's on the right path.
2 hardware sites already reinforced what I was saying. Now if somehow BFG show other wise that bandwidth doesn't make any difference in minimum frame rates and prove those 2 other sites wrong by giving examples and write them email to change their articles I would be willing to say I was wrong. Then again I'm 100% sure with this subject or else I wouldn't be here posting 10 page threads arguing so.
Facebook
Twitter