Originally Posted by Tsumi
You're acting like they simply increased clock speeds using the same cores to get better performance. They didn't, based on what we've seen Bulldozer capable of. They had to have made some IPC improvements which you're so hung up on, as well as increasing clock speeds, all while keeping within the same thermal envelop. And that's the important thing, that the thermal envelop is the same, yet performance is higher. In the end, what it really comes down to is performance/watt in these systems (besides price/performance), which is what AMD is really losing to Intel at the moment, and what Bulldozer actually lost to in regards to Thuban. Not to say that pure raw performance doesn't matter, but for what these chips are targeted at, it's not about raw performance.
Additionally, higher stock clocks at the same thermal envelop generally means higher overclocks possible as well, since the chip is drawing less power and putting out less heat. If that doesn't translate to improvements, I don't know what else can mean more performance to you, besides getting an Intel.
And IMO, comparing Stars performance to Trinity performance is as valid as comparing Vishera performance to Thuban performance. Both Stars and Trinity do not have L3, while Vishera and Thuban both have L3. So performance differences should be similar, and comparable.
Look, I said in my first post
on this thread that this was good enough for mainstream PCs. The problem with your argument is that Piledriver will have shared L2 cache within modules, while Stars doesn't, so, it's not as you are saying. There is that added benefit to Piledriver.
The main point I said with Piledriver - while maintaining that it would be good for mainstream PCs - is that AMD needs a proper flagship. And I stand by what I said, there is only so much you can do when you have to ramp up clockspeeds to get more performance. Why do you think Intel is staying away from that ? They had a Pentium 4 running at 3.8 Ghz and now they are almost at 22 nm and what is the fastest clockspeed you've seen from them ? All I was saying is that they may very well have been able to keep the thermal envelope the same with a 2 module, 4 core design, but what about when you put Piledriver on the desktop and have to add more modules and more clockspeed ? Having a controlled power envelope for mobile is one thing, nothing guarantees you that with a higher clockspeed and more modules it won't use much more energy, and I'm saying this because thatt is exactly how Bulldozer works on the desktop. Now, have they made any fundamental changes that mean that you can now ramp up clockspeeds with more modules and keep the same or lower envelope ?
Originally Posted by Artikbot
Intel's RD budget is more than tenfold AMD's and
they have to develop monstrous graphics powerplants that fight nVIDIAs at the same time that they develop processors.
Do you seriously call that 'no need to invest lots of money'? I'm surprised they still manage to give fight with so little resources.
By the way, AMD has (I suppose you know it) no fabs. Therefore they rely on whatever GF/TSMC has. Then crap happens when the fab promised you X but they deliver X/2.
There we go again, I already addressed that. You guys are all having problems with the way I phrased it, aren't you ? I'm sorry, I'll try to be more clear the next time. What I meant is: AMD does not have money. Intel does. But not even Intel introduces a new architecture AND a new manufacturing process at the same time. So how does AMD have money to do both sucessfully ? They don't, and Bulldozer is the result. And what if Globalfoundries isn't theirs anymore, that is actually one more reason to go at it with more caution. Intel goes with caution so any eventual problems with a new manufacturing process are not amplified by eventual problems with a new architecture, and vice-versa, why doesn't AMD do the same ? Like someone already said, why is GF being used as an excuse ?
Originally Posted by Clipze
go back to playing games bro all this improvement talk is getting to your head
LOL, you had so many posts to make a joke about and you chose the wrong one. I was being as neutral as possible answering a legitimate question stating the facts of what we know thus far regarding Piledriver. Most of it is paraphrasing what the alleged AMD slide says on the OP. The only thing I added was the 26.6% higher clockspeed, and even that is being neutral, because it's just doing the math from what we allegedly know from the A10 5800K's clockspeed.
Originally Posted by lostmage
Even better, they could have transitioned to 32nm, but kept the same PH2 archetecture. The new architecture has flaws, and I think the only thing keeping it from being next to useless compared to a PH2 is the fact that it's on a 32nm process. A 32nm PH2 would destroy Bulldozer until they get the kinks worked out with their new design.
Exactly, actually yours is a better example, they have to compete on power envelope for mobile, so, yes, the best approach, while being cautious and not spending too much money, would be that, transition to 32 nm and use a mildly improved Thuban with the added Turbo Core 3.0 instead of Turbo Core and adding the new instruction sets.
I honestly don't know why people even suggest this might not be possible. It is and they did exactly that with Llano, which they managed to release more than three months before Bulldozer, even though they had the added complication of having an on-die GPU too, and yet they managed to deliver it sooner.Edited by tpi2007 - 4/5/12 at 10:26am