What did you think of the AMD Hawaii Tech Showcase today?

AMD can compete plenty well in the CPU space because people often don't want to fork out an extra $100 for 5 frames per second better performance, and in video rendering which is a perfect example of highly multi-threaded workload AMD beats Intel in nearly every situation when comparing the two mainstream platforms. I can honestly say that owning both and FX-6300 and an i5 3570k that I have only encountered a single situation where I have ever been able to tell the difference between the two systems in any way, and that situation is so far out there that no one should ever really see it. On graphics cards, AMD has done very well for a long time, continually trading the crown with Nvidia. Actually many of the 7000 series cards are far better than their Nvidia equivalent, you just don't see it in a lot of the really old reviews because those don't use the drivers that gave the 15% performance jump to the 7k series. The only time Nvidia is clearly a better choice in all generations so far is for multi-graphics card systems. AMD is working hard on fixing that, but it is a long way from beating Nvidia in that situation.

AMD graphics cards had 1-2 bad years when the cards that had started work already when AMD bought out ATI did not do very well because of merging complications. Those would be the 2k and 3k series. Starting with the first AMD made generation, the 4k series, the graphics cards have done very well. "The" card to have back at that point was the 4870 and then the 4890. Nvidia had no answer whatsoever and was flailing helplessly for almost 2 generations before finally coming back hard with Fermi based cards. Do you even remember the GeForce 100 series? A few remember the terrible GeForce 200 series cards, the power sucking, heat dumping, buggy behemoths. Then the stopgap GeForce 300 series to keep OEM's happy and pretend they had a product. Finally Nvidia got their feet 4 generations later with Fermi and GeForce 400 series, GTX 460, 470 and 480. The whole argument about AMD not being competitive is nonsense.



Many new games are AMD optimized because AMD graphics chips power the Wii U, XBox 360, XBox One, and PS4. AMD CPU's power the XBox One and PS4. Developers are forced to massively optimized for highly parallel architecture now because AMD only is providing ~1.6GHz cores, but 8 of them. The only way to have good performance is to use all the cores you have available. The graphics chip in the new consoles is crap compared to our dedicated cards, but is still something like 15x more powerful than the old generation graphics chips in the consoles in real world usage. On the CPU side again, if you calculate strictly by flops performance between generations we see a big problem, the new CPUs only provide something like 22.x gigaflops, and that is from all 8 cores. Last gen XBox for example theoretically provides about 115 gigaflops of performance. The catch though is that the way the old CPU is designed and coded for actually has real world performance that is pitiful to even look at, something closer to 15 gigaflops real world usage. SO the new CPU should be just slightly more powerful, but only when the game is properly coded for multi-threading. Comparing the Cell processor in the PS3 though, it is a HUGE step down, but much easier to code for.

 

Nvidia's response that will beat these top end cards with a single GPU for the same price with better performance will be in the Maxwell cards, which are late and will be out in the 1st quarter of 2014. A GTX 790 would be a stopgap to maintain superiority, but unless Nvidia has already been working on such a card "just in case" it wont happen in a timeframe that makes any sense since Maxwell cards would be out before a product could be designed, tested, and manufactured. It also would make Titan owners very max, because the card would sell for probably about $1100, which is Titan range but would be near equivalent to running dual Titans. To pull a stunt like that would bring more hate towards them than any benefit from keeping a performance crown a couple months longer when Nvidia will just take it back anyway in short order. But then again, Nvidia loves to piss off anyone and everyone possible in the name of money, so it wouldn't be unrealistic to think them capable.

 

True, but people seem to get quite mad if their new $1000+ toy is replaced by something much better for the same price or barely anything more if it is within a year of release.

 

This article talks about everything I was thinking last night in regards to Mantle when I heard about it:
http://www.anandtech.com/show/7371/understanding-amds-mantle-a-lowlevel-graphics-api-for-gcn

 

Unfortunately AMD is in a bit of a bind, SoI process advantage is quickly running out which is one of the reasons for switching to Bulk. Intel is already used to Bulk and has their designs down well, AMD will be starting fresh. Also due to international decisions AMD must change their entire designs from a gate-first over to a gate-last architecture, another large change where Intel has already been doing it for years and AMD must play catch up. Sorting out the architecture manufacturing process may be more complicated than most people estimate.

 

Power is not exactly heat, although usually the two are linked. That formula you posted has to do with actual power consumption, and power is dissipated as heat. However there are other factors that affect heat output such as the thermal density of the design. We can see from the Sandy to Ivy switch that they have the same power design envelope, yet Ivy is hotter. This is partly because of moving from solder to TIM, but it is also that Intel is at such a small process node that their thermal density is really high, this is compounded by the FinFET 3D design in their processors also allowing more power through in a smaller area. At stock speeds and voltage you are right that Intel does fine. And you are right it does help a lot for the power consumption of their processors, but even tiny changes in voltage drastically increase heat output because everything is so dense and heatsinks have a very hard time moving the heat away. All objects are easier to cool if the heat is spread out more because you can make more efficient use of the larger contact patch. As things get smaller your method of cooling must also change in that you have to come up with a way to get more efficient at cooling a single smaller area. Unfortunately standard heatsinks are not very good at that because the design concept of modern heatsinks is opposite of what needs to be done. We are right at the tipping point these last couple years in that although we can get better OC potential from smaller process node's giving better improvement in speed for less voltage but we are getting so small that the shrinks are causing OC headroom to start to decrease at a higher rate than it is increasing due to the thermal density and lack of cooling efficiency. Intel is at its limits for how much performance they can squeeze out of the Pentium 3 architecture and they will need to do a complete redesign from the ground up if they want to get significantly more performance from the processors. Or Intel can instead do what they have been focusing on, which is improving power consumption to become more relevant to the mobile space. Still, doesn't matter how much anyone loves or hates Intel, it is very impressive they could take an architecture so far and tweak such performance out of it for this many years.


EDIT: Intel could also hope that new materials become mature enough to use as some of the new up and coming materials can be switched faster for less power and this would also bring better performance to their processors even without really changing the architecture.