I've ran AMD since the 90's.....I remember having a 300Mhz Thunderbird CPU. But as I look to replace my aging Phenom II system, I'm just not seeing any worthwhile alternatives.
AMD's latest Vishera chips, are just a complete disappointment to me. I would have gone with Thuban, but a system fry required a motherboard replacement, so I went with the cheapest one, which was AM3 and not AM3+ anymore. So that upgrade path was gone. And now I hear AMD is pretty much giving up on higher end processors in favor of APU's and mobile.
First I'll explain as simply as possible, why Vishera is such a bad processor design, which many of you may already know.
![bulldozer.jpg]( "bulldozer.jpg")
One look at the processor design shows a major issue. Each module shares 1 set of resources between 2 processors. This is something completely outside traditional CPU design, and for good reason: It tremendously cripples a CPU's potential. My main focus here is obviously on the shared Floating Point Unit (FPU). A lot of people see processors as just the "cores" but the FPU is a processor on its own. It calculates more complex math faster than what it would take to be done on a core, and it's soo specialized in its capabilities, that much if not all of the stuff done on the FPU, cannot even be done on the CPU at all, except in some cases with emulation software. Anyways, it's a very important part of a processor.
Now you may say "Oh I don't really use FPU heavy programs anyways", but virtually all programs use it to some extent. A CPU and FPU is a bit like multi-threading....one part must wait on the other. With Vishera having an FPU shared between 2 processors, not only do both processors have to wait on each other (short waits but nonetheless it adds up since they happen many times a second), both of them also have to further wait when the FPU is needed. So when the FPU is doing its thing, both cores in a module are essentially waiting. But it's even worse since each processor also has to wait for the other processor to get done with the FPU. Granted, they did up the speed of the FPU's in Vischera, but it comes nowhere near compensating for the design scheme.
What AMD seems to have done is shot itself in the foot, without even caring. Most regular users may not even care about this design flaw (though they should) but this also severely harms AMD's potential in the mid-high end and professional market as well. If I was for example, a 3D professional and needed a machine to render with, while being budget friendly, AMD wouldn't even be a good choice in the slightest. That's a big segment of the market to lose, even if AMD is doing well in mobile. It's essentially unnecessary abandonment. Let's keep in mind that most top end 3D rendering software still does use the CPU, and the FPU comes into play heavily in that work load scenario.
Now the FPU wouldn't be a problem at all if GPU computation had taken off.....but it really hasn't. Other than things like Folding@Home and some GPU video transcoding programs and Photoshop, General Purpose GPU computing is largely non existent. The FPU issue wouldn't even be an issue if the opposite was true. What this means is that AMD has really tied the hands of those in the professional market, and pretty much blocked out that segment altogether. Obviously, when it comes to business, time equals money. But it's not just about that, it puts AMD in a position where it has rarely been before: A loss of price to performance ratio. The shared FPU pretty much seals this fact.
A good example of this is the 8350 vs the i7 4770k. Now of course, they're different processors in different price ranges, but the FPU performance in comparison, shows the 8350 to be only 60% of that of a 4770k. at $200 for the 8350 vs $300 for the 4770, that's a price difference of 33%. With the 4770K being 40% better. That seems like a win to me by itself. Then if you include power usage, the 4770K wins again, massively, at 84 watts vs 125 watts. Doesn't sound like a big issue, until you consider that could equal at least $5 a month in electricity, either in extra cost or savings. Five dollars doesn't sound like a whole lot, but after 2 years, that could equal $120......which would completely negate any extra costs of the 4770k! The energy savings alone make up the difference in pricing. So for essentially the same price, you get a 40% faster FPU performance and at least 25% more overall performance of the CPU too!
In their defense, I'd say they were looking at a typical usage scenario when deciding how to design the chip, and obviously the CPU is used a whole lot more than the FPU, to the point where the FPU can be very much under utilized. But their mistake is clear. Yes Vischera seems to be specialized more for the home PC market....but even on that segment it barely makes the cut, especially since most home PC's have work loads that would just as easily be fulfilled by an old fashioned quad core. The funny thing is, while gaming may be their largest target audience, gaming actually is one of the most FPU intensive applications of a home user. I think what's most sad is that if each core did have its own FPU, there would actually be quite a performance boost to be had overall, not just on the FPU but the CPU too. When I think of Vischera, I think of an insect with their legs half cut off on one side.
A little off topic, but when it comes to APU's, while they may be the future, they don't come close to satisfying any mid-high end expectations. The problem with APU's is they will always use system memory like an onboard GPU would use. Until they start building gigabytes of RAM into APU's themselves (which would be awesome to have 100GB/s+ onboard system memory for a CPU to have access to), they're still going to be largely limited by system RAM speeds, which are not even close to the speed capable on a dedicated GPU.
Since APU's aren't up to snuff to replace dedicated GPU's, and AMD is pretty much pulling out of the high end CPU market, us AMD users will most likely be forced to go intel for our future builds. And with the advent of Maxwell GPU's, Haswell and Maxwell seem poised to dominate the market in both cost and performance segments.
And if intel combined its Maxwell and Haswell into an APU with onboard memory....AMD wouldn't have a fighting chance. Of course that would require a new socket type, but it seems to be the next logical step in the evolution of PC design. Heck, at that point, most of the use of an independent motherboard goes out the window, might as well be a system on a chip style. I'd like to think AMD has something up their sleeve, but it's not looking like they do or even care to.
AMD's latest Vishera chips, are just a complete disappointment to me. I would have gone with Thuban, but a system fry required a motherboard replacement, so I went with the cheapest one, which was AM3 and not AM3+ anymore. So that upgrade path was gone. And now I hear AMD is pretty much giving up on higher end processors in favor of APU's and mobile.
First I'll explain as simply as possible, why Vishera is such a bad processor design, which many of you may already know.
One look at the processor design shows a major issue. Each module shares 1 set of resources between 2 processors. This is something completely outside traditional CPU design, and for good reason: It tremendously cripples a CPU's potential. My main focus here is obviously on the shared Floating Point Unit (FPU). A lot of people see processors as just the "cores" but the FPU is a processor on its own. It calculates more complex math faster than what it would take to be done on a core, and it's soo specialized in its capabilities, that much if not all of the stuff done on the FPU, cannot even be done on the CPU at all, except in some cases with emulation software. Anyways, it's a very important part of a processor.
Now you may say "Oh I don't really use FPU heavy programs anyways", but virtually all programs use it to some extent. A CPU and FPU is a bit like multi-threading....one part must wait on the other. With Vishera having an FPU shared between 2 processors, not only do both processors have to wait on each other (short waits but nonetheless it adds up since they happen many times a second), both of them also have to further wait when the FPU is needed. So when the FPU is doing its thing, both cores in a module are essentially waiting. But it's even worse since each processor also has to wait for the other processor to get done with the FPU. Granted, they did up the speed of the FPU's in Vischera, but it comes nowhere near compensating for the design scheme.
What AMD seems to have done is shot itself in the foot, without even caring. Most regular users may not even care about this design flaw (though they should) but this also severely harms AMD's potential in the mid-high end and professional market as well. If I was for example, a 3D professional and needed a machine to render with, while being budget friendly, AMD wouldn't even be a good choice in the slightest. That's a big segment of the market to lose, even if AMD is doing well in mobile. It's essentially unnecessary abandonment. Let's keep in mind that most top end 3D rendering software still does use the CPU, and the FPU comes into play heavily in that work load scenario.
Now the FPU wouldn't be a problem at all if GPU computation had taken off.....but it really hasn't. Other than things like Folding@Home and some GPU video transcoding programs and Photoshop, General Purpose GPU computing is largely non existent. The FPU issue wouldn't even be an issue if the opposite was true. What this means is that AMD has really tied the hands of those in the professional market, and pretty much blocked out that segment altogether. Obviously, when it comes to business, time equals money. But it's not just about that, it puts AMD in a position where it has rarely been before: A loss of price to performance ratio. The shared FPU pretty much seals this fact.
A good example of this is the 8350 vs the i7 4770k. Now of course, they're different processors in different price ranges, but the FPU performance in comparison, shows the 8350 to be only 60% of that of a 4770k. at $200 for the 8350 vs $300 for the 4770, that's a price difference of 33%. With the 4770K being 40% better. That seems like a win to me by itself. Then if you include power usage, the 4770K wins again, massively, at 84 watts vs 125 watts. Doesn't sound like a big issue, until you consider that could equal at least $5 a month in electricity, either in extra cost or savings. Five dollars doesn't sound like a whole lot, but after 2 years, that could equal $120......which would completely negate any extra costs of the 4770k! The energy savings alone make up the difference in pricing. So for essentially the same price, you get a 40% faster FPU performance and at least 25% more overall performance of the CPU too!
In their defense, I'd say they were looking at a typical usage scenario when deciding how to design the chip, and obviously the CPU is used a whole lot more than the FPU, to the point where the FPU can be very much under utilized. But their mistake is clear. Yes Vischera seems to be specialized more for the home PC market....but even on that segment it barely makes the cut, especially since most home PC's have work loads that would just as easily be fulfilled by an old fashioned quad core. The funny thing is, while gaming may be their largest target audience, gaming actually is one of the most FPU intensive applications of a home user. I think what's most sad is that if each core did have its own FPU, there would actually be quite a performance boost to be had overall, not just on the FPU but the CPU too. When I think of Vischera, I think of an insect with their legs half cut off on one side.
A little off topic, but when it comes to APU's, while they may be the future, they don't come close to satisfying any mid-high end expectations. The problem with APU's is they will always use system memory like an onboard GPU would use. Until they start building gigabytes of RAM into APU's themselves (which would be awesome to have 100GB/s+ onboard system memory for a CPU to have access to), they're still going to be largely limited by system RAM speeds, which are not even close to the speed capable on a dedicated GPU.
Since APU's aren't up to snuff to replace dedicated GPU's, and AMD is pretty much pulling out of the high end CPU market, us AMD users will most likely be forced to go intel for our future builds. And with the advent of Maxwell GPU's, Haswell and Maxwell seem poised to dominate the market in both cost and performance segments.
And if intel combined its Maxwell and Haswell into an APU with onboard memory....AMD wouldn't have a fighting chance. Of course that would require a new socket type, but it seems to be the next logical step in the evolution of PC design. Heck, at that point, most of the use of an independent motherboard goes out the window, might as well be a system on a chip style. I'd like to think AMD has something up their sleeve, but it's not looking like they do or even care to.
