Quote from: william blake on February 01, 2020, 00:13:34
Quote from: Valantar on January 31, 2020, 23:37:32
What facts? You were pointing to the CPU having a larger increase in performance than the GPU and somehow making that into the GPU being "whatever". That, my friend, is as pure opinion as you will find.
look at the market. strong cpu+whatever gpu rules the roost. whatever cpu+strong gpu is a low cost and small segment.
its not my opinion.
or this way-why there is no really strong integrated graphics, never? they can do it easily, right? because integrated graphics are
-slower or much slower than the same piece of silicon with own memory and tdp
-there are not too many people that needs EXACTLY this level of performance.
-its impossible to make a fast cpu and a fast gpu with shared tdp. one will limit the other.
-people are not interested in paying for igpu. they want them for free of for a very low price. this mean very low margins for a manufacturer.
-and finally big chip is not good for increasing margins, small segment not good fot it also and wafers are limited
all that above are not my opinions but FACTS.
Well, let's conveniently skip the fact that you stated none of this in your previous "factual" post, but never mind that. Let's address this point by point:
-Sure, low-power CPUs with weak iGPUs are the best-selling chips out there. Part of this is (obviously) because there are no really powerful iGPUs, but besides that fact, that wasn't what you said in your previous statement. You didn't mention market segments, sales or anything of that nature, you made a qualitative judgement of the relative performance increase between the CPU and iGPU parts of Renoir. You can't say one thing and then claim you said something else.
-Why there have been no powerful iGPUs up until now? That's a very complex question. Market forces is one factor, other bottlenecks (RAM etc.) is another. Most of these are however changing in one way or another. Gaming has exploded over the past decade, and unlike five years ago (when development of currently available CPUs likely began) it wasn't reasonable to expect gaming on a 15W or 25W chip - they struggled enough getting acceptable CPU performance in that TDP back then. Today, this has changed. So the possible performance acheived within the TDP is greater, while the public expectation/demand for performance has also shifted towards the GPU. Have you looked at Steam Survey data and seen how many people actually game on Intel iGPUs? It's crazy.
-Can this be made easily? Of course not. Part of what makes this hard is memory bandwidth. Up until the past few months the fastest memory available for PCs has been DDR4, which in mobile low-power applications is quite slow (dual channel 2400MT/s at best, as higher speed requires a lot more power), especially in a GPU context. Even a Threadripper with four channels of high-speed (3600 or more) DDR4 pales in comparison to a midrange dGPU in terms of bandwidth. LPDDR4x serves to alleviate some of this bottleneck, though not remove it entirely - it's still about twice as fast as most low power DDR4, with AMD aiming for 4266MT/s. Another part is the power, obviously, as there are limits to how much performance you can squeeze out of a low number of total watts. But laptop cooling is improving, and 25W cTDP-up designs are becoming a lot more common in the thin-and-light space than previously, so there's definitely room to grow here. We're obviously not talking anything matching a powerful dGPU, but given that 3000-series Ryzen fell behind the MX 250 by about 25% it's not unreasonable to expect this new generation to surpass that - and that's a very decent level of performance. Zen2 on 7nm is also extremely efficient, allowing for lots of performance at just a few watts per core, so it's more possible now to not throttle the CPU quite as much while the GPU is being loaded.
-Yes, obviously an iGPU with shared memory and a given set of cores in, say, a 25W shared TDP will be slower than a dGPU with the same number of cores, a higher TDP, and faster dedicated memory. That's as obvious as stating the sun will rise tomorrow. The problem is that not every laptop can house a dGPU, let alone cool it, and you can't make dGPUs that match the low power of an iGPU as the added chip and memory add an unavoidable power draw. It is therefore possible to get a lot more performance out of a low total power - say 25W - with an APU than with a CPU+dGPU. Given that previous gen 15W APUs nearly matched setups with 15W Intel CPUs + 25W MX150 and MX250 dGPUs, this is a given. If you achieve X performance at 15W and 1,2X performance at 40W, the latter is rather disappointing IMO. And that's the reality with current hardware, even if you get a lot more performance by moving to, say, a 50W GPU like a 1050 - but then you're also looking at a 65W total power budget, or >4x the APU. The point is that dGPUs can't scale down low enough to address what is becoming very useful performance that iGPUs today can achieve.
-An addendum to the previous point: of course sharing power between two things will cause them to limit each other, but your statement is patently false even as a stand-alone fact. It is entirely possible to make a fast CPU and a fast GPU with a shared TDP - the TDP just needs to be high enough for both to perform optimally.
-Laptop sales and the use of Intel iGPUs on the Steam Hardware Survey disagrees with you. Sure, large premiums are generally out of the question (as that would bring you into dGPU laptop territory; it's a tight market), but there are definitely a lot of people out there willing to pay premiums for premium functionality in thin-and-light laptops.
-Again, yes, but so what? 150mm2 is not particularly large, especially when taking into consideration that it's a monolithic SoC rather than the MCM solutions of AMD's desktop chips or Intel's mobile offerings (they tack on the chipset alongside the CPU on the package, AMD integrates the chipset in the SoC, lowering packaging costs). If they were making a 250mm2 7nm APU that would likely be very expensive, yes, but they aren't. And the upcoming 150mm2 ones are likely to perform admirably.
In short, all you are really saying is that you want more performance than is currently available in the 15-25W class, and that you want to move up further than upcoming parts are likely to allow. This is entirely fine, but it doesn't mean that the performance of these parts won't be perfectly fine for a lot of people. I'm looking forward to playing a lot of Rocket League on an upcoming APU, personally.
Deutsch
English
Español
Français
Italiano
Nederlands
Polski
Português
Русский
Türkçe
Svenska
Chinese
Magyar