Quote:- Connecting to a 50” 720p Tv, future proofing system for future upgrade to 1080p, so I think I need 4x IR.
3x IR or higher is recommended for 1080p, 2x IR or higher is recommended for 720p.
Quote:- Would prefer to stick with AMD, (strong Intel preference on this forum)
As others have said. Good luck running the heavier games. The "intel preference" on these forums is based on real verifiable data from dolphin users, not opinions.
Quote:Is Llano APU a good choice or should I go with Zambezi AM3+ setup?
Bulldozer (zambezi) is currently the slowest architecture on the market for branchy single/dual/triple/quad threaded code. It's literally the worst choice you can make. Llano is decent and is the best amd architecture for dolphin currently but it's nothing next to a good sandy bridge/ivy bridge cpu (in fact gulftown/bloomfield/lyynfield/clarkdale/yorkfield/kentsfield/wolfdale/conroe/etc. are all still faster).
Quote:I guess there is no clear "min requirement" setup for 4x IR for MOST games.
......look at the first post in this thread.
Quote:Secondly - Vram setup: thinking of using Windows Ready boost (usb flash drive memory) instead of HD VRam
Vram stands for video ram, please say virtual memory or virtual ram in the future.
Quote: I read it kills HD lifespan, is there a limit of size and number of USB’s used? Is there any benefit to use USB3.0 over USB2.0? Any thoughts would be appreciated.
Nobody uses readyboost because it's almost pointless. It does not significantly improve HDD lifespan, I don't know where you read that.
Quote:I looked around but couldn't determine if AMD is changing the Chipset, or if processor will be upgradeable to FM2 with the same motherboard down the road.
FM1 will be replaced by FM2 when trinity comes out. Trinity will not be backwards compatible with FM1. Bulldozer uses AM3+ (or AM3 with a bios update). Piledriver will likely continue using AM3+.
Quote:The APU (advanced process unit cpu/gpu) has some sort of Hybrid CrossFireX abilities when combined with certain graphics cards. The one I chose being the highest possible combination to gain the benefit.
Like werewolfy said dolphin doesn't benefit from crossfire. However even for pc games it makes very little different since the integrated gpus are so weak compared to discrete gpus.
Quote:I guess I'm not understanding the benefit of Intel.. More L3 banwidth? Multi thread capacity?
Don't compare intel to amd. Compare specific microarchitectures against each other. Intel just happens to have better microarchitectures on the market right now. Microarchitectures are ludicrously complex and most of the details about how they work are kept a secret so that competing companies can't copy them. As a result it's very difficult to determine specific reasons why one architecture is faster than another. You may be able to uncover a few reasons through careful benchmarking, analysis of results, and comparison of those results with public architectural details. But you will never know all of the reasons. Because bulldozers performance was so unexpectedly bad many companies and people performed these studies and have tried to point out some of the architectural flaws.
After reading through lots of reviews, architecture analysis articles, and opinion articles I compiled a short list of some of the visible flaws in bulldozers microarchitecture that could be contributing to its poor performance. So here it is if you want some elaboration:
Quote:Now that I've spent two days pouring over technical details, reviews, and some user analysis of bulldozer I think I've identified all of the major problems with the chips (at least the ones that can be directly observed, obviously some problems may exist at the low level that only the engineers who worked on the chip know about).
1. Cache Thrashing (cores competing for access to cache resources)
Bulldozer has an L2 cache which is shared by a pair of cores while sandy bridge has a dedicated L2 cache for each core. The two cores can sometimes compete for access to the same data in the cache. Some users tests have already shown that disabling every other core on a bulldozer chip can sometimes actually IMPROVE performance for software that is highly multithreaded.
2. Pipeline Flushes (poor branch prediction + long pipeline gives a big penalty)
Bulldozer has a much longer pipeline than deneb/thuban (phenom II) or sandy bridge. Reviews have shown that although AMD promised better branch prediction in bulldozer over its predecessor by decoupling the branch prediction from the pipeline the branch prediction in bulldozer is actually worse than deneb/thuban. Since bulldozer has a longer pipeline these branch misfires (I think that's the word for it) have a much higher performance penalty than in other chips.
3. Only 4 floating point/SIMD units
This isn't so much of a problem as it is a lack of an advantage. Despite having 8 "cores" bulldozer has the same number of FPUs as sandy bridge. This means it has no potential performance gains for x87/SSE applications.
4. Extremely high cache latency/lower cache bandwidth
Cache performance in bulldozer is VERY poor. Bandwidth is lower than the competition and deneb/thuban. Latencies are slightly higher than deneb/thuban and MUCH higher than sandy bridge. L3 cache latency in particular is abysmal. Cache latencies were raised in order to improve yields.
5. Decode/dispatch not able to keep the execution pipelines utilized
It appears that bulldozer is not able to keep the functional units in the execution pipelines occupied. Functional units are being under utilized and poor decode/dispatch may be playing a role in this.
6. Reduced functional units per core
Not really much of a problem since bulldozer isn't even doing a good job at keeping the functional units it does have occupied with work. However I should point out that in deneb/thuban each core has three ALUs, in sandy bridge each core has four ALUs, and in bulldozer each core only has 2 ALUs, less than its predecessor.
7. Clock rates are lower than expected (4.5GHz with turbo core up to 5.0GHz was expected by AMD)
Clock rates were lowered to well below expectations in order to boost yields.
8. Applications and OS not properly designed for the new threading system
Performance for bulldozer on windows 8 appears to be significantly better than windows 7 due to optimizations for bulldozers unique threading system. The problem is that right now multithreaded applications running on bulldozer are distributing their threads in order of available cores, this means that two threads will end up running on the same module rather than two separate modules. Windows 8 fixes this by treating it like a CPU with HT, so threads will be distributed to different modules instead of just different cores until all of the modules are occupied. Luckily dolphin is already set up to do this but in the meantime this is a serious bummer for AMD until windows 8 becomes common (which won't happen for a very long time).
Clock rates were lowered and cache latency was raised in response to very poor yields. Poor yields were a combined result of the ridiculous die size/transistor count and the poor quality 32nm fabrication at global foundries.
Piledriver (2012) is expected to have higher clock rates and better cache latency as yields improve. New instructions will be added, a 5th module will be added (10 cores), power consumption will be reduced, and undisclosed IPC improvements will be made (10-15% according to AMD). We can only hope that this is true, otherwise AMD is in deep shit.
If it wasn't for the additional cores and memory/HT bandwidth bulldozer would actually be worse than its predecessor (deneb/tuban) in every way. The biggest problems seem to be that the chip is too big, the pipeline is too long, the branch prediction is poor, and the yields are poor (resulting in lower clock rates and higher cache latency as well as higher production costs). Performance for dolphin on bulldozer will no doubt be significantly worse than on deneb/thuban (phenom II).
Llano has avoided these problems (except the poor yields). The pipeline is short, branch prediction is decent (although not as good as sandy/iv bridge), the chip size is decent, threading is good, clock rates are decent (although not as good as sandy/ivy bridge), cache performance is almost as good as sandy/ivy bridge, and memory performance is almost as good as sandy/ivy bridge. Llanos inferiority primary come from the cores themselves. The execution pipeline is narrower than sandy/ivy bridge (3 wide instead of 4 wide) and the out-of-order execution control is not as efficient as sandy/ivy bridge. It averages about 2 instructions per clock while sandy bridge gets close to 3 in many situations. SIMD performance is also surprisingly bad on llano (I have no idea why).
Quote:Any other dolphin system building tips, gddr5, high L3, RAM speed?? Anything would be helpful.
-desktop sandy bridge/ivy bridge cpu, or very high end mobile sandy bridge/ivy bridge
-decent gpu, preferably nvidia since HD7000 cards seem to be having issues with dolphin, GTX 550 or higher would be recommended
-at least 2GB of ram
If you have those three things you're pretty much set.