Nintendo Maniac 64 Wrote:I don't really feel like going into a whole CPU architecture discussion,
But that's the entire point of this thread.....look at the title.
Nintendo Maniac 64 Wrote:but I'd just like to say that the Pentium 4 comparison is only half correct. Northwood, viewed as the only "good" Pentium 4, had a 20 stage pipeline
Incorrect. Yet more marketing fluff. The entire idea of the pipeline was originally to help university students grasp the basics of modern (this is in the 1980s so modern in those days) microprocessor design. But it is a flawed model meant to illustrate the concepts quickly, rather than accurately. A more accurate (thought still not fully accurate) model would be to imagine many different pipelines with different lengths all woven into each other. Unfortunately this is difficult to visualize or illustrate effectively. Different operations have a different number of stages and take a different number of clock cycles to complete even if their are no bubbles or flushes. In scalar designs their are many different "pipelines" for different operations but only one of them can be active at a time since they all share the same data input and control input transmission lines and are behind multiple layers of multiplexers and demultiplexers. In a superscalar design the execution resources are organized into blocks that can be active in parallel. Within those blocks are functional units that contain subunits, these cannot be active in parallel. These are sometimes referred to as subpipelines or execution pipelines to avoid confusion but usually I see documents simply avoid using the word pipeline at all.
Certain companies (mainly apple and IBM) began big marketing campaigns to convince pc users that short pipelines where important as a way to justify the lower clock rates of the chips that they were using. Of course consumers need to have a single number to use to easily compare products with. The only way that they could provide an easy to use single number was to operate under the assumption that execution takes only 1 clock cycle (which it only does when doing certain simple operations). All pipeline lengths listed on any processor operate under the assumption that the execution stage takes only 1 cycle. This is misleading as it completely ignores the advantages that some chips may have over others that can pass instructions through the "execution stage" in fewer cycles (lower instruction latencies).
Intel only added to this with their own fluff. The original netburst architecture used in willamette, northwood, and gallatin had an unknown number of frontend stages and 20 backend stages in the pipeline even by that logic of 1 cycle execution stage. Intel attempted to backpedal away from revealing the true length of their pipeline by stating that their "critical execution path" was only 20 stages long since the microcode trace cache allowed the frontend stages to remain inactive so long as the program did not try to access instructions that weren't in the trace cache. Basically it was their way of saying that only backend stages mattered. Which is only true to a certain degree. But unfortunately they also listed their pipeline length as 20 stages to the press stating that as far as they were concerned the frontend stages shouldn't count towards the pipeline length since they didn't matter in the same way that they had in previous architectures. While this is probably a better number to use when comparing it to previous architectures since the frontend stages don't have anywhere near as much impact on power consumption and performance as the backend stages do it is misleading to label it as the "pipeline length" since they used a different logic to come up with this number than other companies. The prescott core actually significantly reduced the instruction latencies of most instructions, though it compensated for this by lengthening the rest of the pipeline. Thus making the comparison even more misleading.
Nintendo Maniac 64 Wrote:- the exact same amount as Bulldozer and Piledriver. It was Prescott that was the really power-hungry one and had a stage pipeline that supposedly varied anywhere from 30 to 100.
Yeah.....no.
They were both power hungry architectures build off of the same general concepts. The top end prescott chips had higher TDPs mainly because of their higher clock rates. If you compare models that ran at the same clock rates prescott consumed only slightly more power (granted most of that is due to the new manufacturing process that they used for prescott) than northwood. Pentium 4s power consumption mainly ramped up as they raised clock rates, not when they added more pipeline stages.
"Anywhere from 30 to 100" is actually an accurate description of the pipeline length for both generations, not just prescott. They still to this day have not told us how long the frontend was and there is no way to figure it out via software testing because of the microcode trace cache. I have a feeling that it was at least 10 stages so 30-100 total is actually pretty spot on.
Nintendo Maniac 64 Wrote:Also, I think you'd be surprised at how many people play PC games without being a "hardcore gamer". You don't need to be playing Battlefield 3 or anything - do you know realize many people play Minecraft and World of Warcraft? And Intel iGPU are particularly bad at OpenGL performance, so Minecraft pretty much requires AMD or Nvidia graphics.
There are very few serious gamers that are willing to deal with integrated graphics. If you count anybody who plays video games as a gamer than most gamers don't play anything more demanding than flash games on their PCs and are perfectly fine with a shitty Intel IGP. For those users Intel cpus offer them better performance in the applications that they will actually use.
If you only play WoW or minecraft sure that APU is fine. But why the hell would you limit yourself to just a handful of modern games? What if you suddenly find that you enjoy a new game that is very demanding down the road? I realize that lot of people play games that aren't very demanding like minecraft or WoW but I doubt that most of those people only play those games and will only want to play light weight games like that for the next couple of years. Gamers tend to think ahead when they plan their builds because if they're poor they know that they're going to be stuck with it for awhile. Just look at how few steam users use integrated graphics. Hell I don't know any gamers that bother with it unless they were forced to by being given an OEM system for free. Building a system with discrete graphics barely increases the total system cost and massively increases video game performance. It gives serious gamers the performance that they need/want. A 2x lead over Intel IGPs is just not enough to make gamers turn away from low/medium end discrete solutions.
AMDs current APU lineup is only really good for HTPCs in my opinion. It's too GPU focused for regular users and doesn't offer enough GPU performance for "hardcore gamers" (I hate using that word but I really don't know what else to call them, serious doesn't sound any less elitist).
Nintendo Maniac 64 Wrote:EDIT: And I'm finding out that apparently Sandy/Ivy Bridge's iGP + Windows 8 = problems with OpenGL support. Yet again, AMD or Nvidia is pretty much required.
I can confirm this firsthand.
But it's not just openGL. Intels drivers suck in general when it comes to 3D stuff. OpenGL is particularly problematic but I have had issues with d3d applications as well. They're fine for anything related to video or sprite rendering though.
Nintendo Maniac 64 Wrote:Minecraft benchmark
For the record minecraft runs fine on most shitty Intel IGPs as long as you don't jack the settings up to max (like they did in that benchmark). I literally have it running at tolerable performance levels on a mobile GMA 950 from 2006.
And now I've been typing for waaayyy too long so I'll read through page 25 in 2 days.
) , AMD 5450 & Nvidia GT 520 are garbage as well