Did ATI cripple their own cards?

[Drythn]

Alright guys, put on your tin foil hats, I have something interesting to say.
We've all seen the massive improvements using 10.2 over the older versions of drivers. Everything seems fine, right? I mean, new versions are supposed to improve the performance.
However, the big question mark arises over the leaked 10.3 drivers.
Some users are reporting increases of ~50%(even in the extremely demanding Heaven benchmark: from 28 to 45+ on a 5870 at huge resolutions).
If you're thinking about it, 10.3 will probably be released right before Fermi.
Some good months passed since 58xx series were launched, yet it's right NOW when we're seeing some huge improvements.
This makes me wonder a little:
Did ATI cripple their own cards using bad drivers?
Did they do this to fool Nvidia into thinking something like "It's that number we need to reach!"?
Either way, I can't wait for the stable 10.3

 

[zagood]

possible, but more likely that the hardware was ready before the drivers. CF/Eyefinity issues and the gray screen of death point to that.

 

[WhoBeDaPlaya]

No idea, but it would be bloody ironic since this comes right out of nVIDIA's playbook

 

[Udgnim]

ATI has been steadily improving their 5XXX drivers with fixes and performance improvements

I'm not going to be so cynical as to believe ATI was purposely holding back on its users.

one possible reason for the reports of Cat 10.3 Heaven benchmark FPS increases is that before the reports of Fermi performing very well in the DX11 Heaven benchmark, ATI did not view devoting resources to improve Heaven benchmark performance as a priority. after the reported Fermi numbers came out, e-peen became at stake, and e-peen is everything when owning high performance GPUs.

 

[Phil1977]

It is very possible that they have been holding back performance boosts to spoil Nvidias launch a bit...

 

[lifeblood]

ATI's drivers have always been their weak spot. However, a 50% performance increase? I seriously doubt it's true. If it is true then either ATI is playing serious games or their driver writers are incompetent.

 

[happy medium]

ATI's drivers have always been their weak spot. However, a 50% performance increase? I seriously doubt it's true. If it is true then either ATI is playing serious games or their driver writers are incompetent.

I read the 50% increase somewhere. It was in one game under crossfired 5870's.
Don't believe the hype.

It might have been Fudzilla where I read it.

 

[Dark4ng3l]

I read the 50% increase somewhere. It was in one game under crossfired 5870's.
Don't believe the hype.

It might have been Fudzilla where I read it.

Even 20% would be a MASSIVE boost really. Anyways lets all hope we can get 50% more speed out of our cards for free.

 

[soccerballtux]

ATI's drivers have always been their weak spot. However, a 50% performance increase? I seriously doubt it's true. If it is true then either ATI is playing serious games or their driver writers are incompetent.

I read the 50% increase somewhere. It was in one game under crossfired 5870's.
Don't believe the hype.

It might have been Fudzilla where I read it.

if the hardware is exactly the same (from my understanding the only things cut out were stuff like the inter-gpu connection), then what we're postulating is that since AMD has no competition and it's looking like Fermi's not going to bring anything competitive to the table, then AMD would release more powerful cards but cripple their performance on purpose with the drivers. At some time in the future, when they release say a 5890, they could "unlock" that performance that's been there with the 5870, as a reason for people to upgrade.

That would be dirty playing, but it's part of the free market and we need Nvidia to provide some competition to keep ATI competitive on both the performance and price front.

 

[jiffylube1024]

Several things at play here. Most notably, these "50% improvement" miracles are always on pre-release drivers from shady sources. When the actual final drivers come around, it's pretty much always in the 2-5% increase range, and rarely eclipses a double digit improvement, usually when improving something like memory bandwidth efficiency.

As others have said, ATI's had weaknesses with driver optimization traditionally, which is both a gift and a curse. Pick a bunch of their heavy hitters, from the original Radeon (later renamed 7200) to the 8500 to the 2900XT to more recent ones, and they all have had driver issues that slowed them down at launch and those products realized solid performance improvements over their lifespan.

I'm going to go with historical precedent here - any improvements are based on incremental driver revision and optimization.

 

[GodisanAtheist]

No, ATI didn't cripple their own cards, they just had to make sacrifices in order to get meaningful yields and a rational die size. Those sacrifices can be seen in the far less than perfect scaling from the 4xxx series to the 5xxx series, yet folks who just can't seem to come to grips with the real performance of ATI's newest batch of cards (insecurities? I dunno) seem to keep floating this "OMFG mega drivers are coming, hold on to your pants everyone!" spiel.

Additionally, the idea that these drivers were held back to "drop the bomb" on Nvidia seems a little out there: Nvidia is going to try and make the fastest card they reasonably can, regardless of what ATI or anyone else is doing. If ATI had released "Uber Drivers" on launch day or if they release them on March 26th, it would make little to no difference to Nvidia.

If I remember correctly, there were rumors about the 9.11's or the 9.12's fitting the whole "drivers to end all drivers" thing, and I said the same thing then as I say now.

 

[YBS1]

Not commenting really on the 10.2 or 10.3 performance, but something did strike me a bit odd looking at the benchmarks in the new anandtech 5830 review. I don't recall the 4870x2 trading blows with the GTX295 quite like it currently seems to be doing at 1920x1200 back when they were both "new" products. It does seem like something has been uncorked via drivers if not in the 10.2's then at least over the past few releases.

 

[bryanW1995]

4870x2 as ~ 5% slower than gtx295 when 5870 launched, with 5870 another 5%slower than 4870x2. are you saying that 4870x2=gtx295 now, or is it still just a little bit slower? I remember predicting (along with many others) that 5870 would eventually catch up and even exceed thost dual gpu cards much like 8800gtx did to 7950gx2. Have you seen that borne out by recent testing?