vishalaestro

Distinguished
Jun 29, 2011
1,446
0
19,310
here is the specs
Intel Core i5-3450 Ivy Bridge 3.1GHz
GIGABYTE GA-H61M-DS2
corsair value select 2x4gb(1333 mhz)
amd 7770 1gb
XFX SMPS ProSeries 450W PSU
if i buy this system i won't be changing it for 4 years so will a four core cpu will be enough or i will have to wait for amd piledriver 8 core cpu
 

InvalidError

Titan
Moderator
The FX-8150 gets creamed in most games by the i5-3570k so waiting for AMD's next 8-core design seems completely pointless, even more so considering that AMD dropped out of the high-end desktop CPU race, which makes it unlikely that they will ever regain the upper hand in typical games which are likely to remain lightly threaded for the foreseeable future.
 
8 real cores would be nice, but that's not what a 81xx truly is (and any "8 core" CPU based on that architecture will be the same). My thought on a 81xx CPU is that it's basically a quad core with 4 more that are pretty much AMD's version of HT (not AT ALL the same, obviously, but performance wise is what I'm talking about). 8 Intel cores (NOT including HT) would be awesome though.
 

InvalidError

Titan
Moderator
HT shares register files, execution queues, address calculation units, retirement units, FPU resources, etc. between the two threads, only a few elements are reserved for either logical thread.

Modules in AMD CPUs have two completely independent integer execution units paired with one shared floating point unit, which makes it a true 8 cores integer chip. All execution resources within each integer core are dedicated to a single thread.

i2/i3/i7 takes a performance hit from enabling HT due to resource contention between the two threads in the execution pipeline while 81xx may take a performance hit due to increased load on caches, memory controllers and arbitration logic between cores/modules.

They are two very different approaches with very different implications. HT yields more work done per transistor while AMD's integer module should yield much higher peak integer throughput on properly optimized mostly-integer code.
 


Exactly what I meant by not AT ALL the same... I know that HT isn't exactly the same performance as the integer cores on the BD CPU's, but they don't equal the performance of 8 actual cores either.
 

InvalidError

Titan
Moderator

What is an 'actual core' on a fundamental basis? IMO, this would be the smallest non-reducible functional execution resource in a CPU.

Each BD integer core is a fully viable self-contained integer decode and execution unit, it does not get any more 'real' than that. You could yank a BD integer core, slap an IO/memory controller on it and you would have a working (albeit hellishly slow due to not having L2/L3 cache or FPU) CPU. Physical or logical internal grouping and additional resource sharing as modules is nothing more than that, grouping with extra resources.

How would a "real-er" BD be different? Remove the module boundary and plug them directly in the main fabric? Not much of a change except for quadrupling interconnect complexity between cores.

Optimize multi-threaded integer code to avoid cache thrashing on BD and you would get practically the same integer throughput you would get with "real-er" cores since the raw int-power would be unchanged. The "real-er" cores may actually end up SLOWER in a properly optimized workload due to higher latency between cores and more bandwidth wasted across the internal fabric due to having twice as many devices snooping on it.

Same goes for GPUs, except architectures change so much every few generations that what could pass as a fundamental execution block changes every few years, unlike CPUs where this has not really changed in 50 years.

And before I get accused of AMD fanboyism, I'm actually an Intel fanboy.
 


Just talking about a "normal" core like all AMD CPU's pre-BD (and Intel cores without HT, too, of course). I suppose my main point is that a single BD module doesn't equal the performance of 2 pre-BD cores, so one "core" on each module might as well be equated to HT, just slightly better.

If BD's IPC wasn't so bad, it's not a terrible design per se, (although I wouldn't have chosen to go in that direction, if I were designing it), but the low IPC just cripples it.

Still, if it were 8 Phenom II-like cores (possibly a newer architecture and with a die shrink, but on the same plane), clocked at the same speed as a BD chip (not necessarily viable or possible, I know), it would be a much more capable chip, and something that could have truly challenged Intel, IMO.

AMD tried to do something innovative, which I guess I applaud, but so far it's bitten them.
 

so let me get this straight... you haven't bought a pc yet, yes? i thought you would have made decision already... i guess you're doing more research.
if the software and the pc industry as a whole changes towards multicore computing, then yes, a cpu with mroe cores would be viable.
it depends on what you want to do with the pc. if you can utilize more than 4 cores, then a 4+ core cpu would seem feasible. otherwise it'd be overkill.
problem with amd cpus is, they already had 6 core thuban cpus available and those didn't really catch on with everybody. same with bulldozer in which they messed up the execution. amd's ideals sound great on paper e.g. bulldozer architecture is good, but the cpus are not.
as for piledriver's performance, you'll have to wait till it comes out, gets reviewed then see how it turns out to be.
to answer your question: the is no absolute way to predict the future. if you think that the next technology upgrade will bring more performance and be more 'futureproof' then you'll end up waiting forever.
 

InvalidError

Titan
Moderator

While there is no doubt that many may be making SOME use of additional threads to run miscellaneous background tasks such as prefetching data, garbage collection, audio processing, etc., dividing critical game logic in a way that can fully leverage multiple cores is going to take a lot more effort and I do not expect many games to push that far any time soon.

While delegating housekeeping and background tasks to threads may not allow games to make full use of multiple cores, it does take mundane stuff off the main game logic thread which I suspect is what most software developers will be focusing on.
 


Considering that not everyone even has a quad core CPU yet, I'd say so. Programmers aren't gonna make something that the majority of people can't make the best use of (even in 4 years).

Core count won't be your problem. Overall speed will. That's the main problem in buying a non-K CPU because you can't really OC much to make up for the speed difference in newer architectures.
 

They're focusing on consoles, because this is where the profits come from. If you want to play games, I suggest you get a console and a low end laptop to surf the internet. There really are no profits coming in from the PC crowd compared to the console side, so PC development hasn't really been optimized. Even the newest games can barely take advantage of more than 2 cores at most.
 
as long as games require less than 4 threads then a basic quad will be enough, when they move to 8 threads then intel would struggle in some cases. ht will reduce the pain but executing 8 threads on 4 cores will always be slower than executing 8 threads on 8 cores regardless of manufacturer.

intel will release 8 core cpus soon enough but will they be useful for gaming? not for another 3-5 years if current trends are anything to go by.
really its out of intels hands and more in microsoft and sonys. if they move to 6 or 8 cores for there new hardware and take full advantage of them then pc's will respond in kind. but saying as the new gen of console is likely to be 4 cores and maybe 8 threads i dont see things changing dramatically for pc users...
if the rumors are true and im not saying they are. any current midrange pc will match the new consoles so there will be no need to upgrade any part of that machine for at lest 5 more years. im not saying pc's wont move on just that there will be less pressure to do so other than wanting more power.

as it stands console is set to hold back pc gaming for at lest another 6 years because the next gen may be dx11 but they will be low midrange cards to keep production costs manageable. this is what will hold pc gaming pack as pc's are looking at dx12 next year but most games will still be being produced for dx10-11 on consoles for the next 5 or more years.

 

thats not actually true m8, pc gaming is worth more than both consoles combined... this year pc gaming will surpass both if this article is correct. http://forums.electronicarts.co.uk/pc/1427752-pc-take-over-console-revenue-say-nvidia.html consoles will have there place but its unlikley they will be lead platform unless its a console exclusive which there aint many of now.
im not saying your entirley wrong about pc's being under developed but consoles arnt as attractive as they used to be and the next gen is likely to be even less so...
 
lol that happens a long time ago... just not in public... intel do have a 48 core cpu http://www.wired.com/gadgetlab/2009/12/intel-48-core-processor/ and i heard runours of a 64 core cpu

http://www.wired.com/gadgetlab/2007/08/64-core-chips-a/ announced in 2007 some are talking 1000 cores. not that they are of any use to gamers but still 1000 cores... :lol:
 

InvalidError

Titan
Moderator

Apart from the high-end gamers who play with vsync off so they can have a chance to see things one torn-up frame sooner in FPS games, this is a non-issue.

Many of the most popular non-FPS games like WoW, D3 and SC2 have rather quite modest hardware requirements and still work perfectly fine on a C2D-E8400 and HD5770.
 

amdfangirl

Expert
Ambassador


They build it for consoles because they target the lowest denomination.

It needs to run well on the console first. No point making a really graphically amazing gaming then making crappier textures for the console.

Making them crappy to begin with means less effort.
 


That will change somewhat with the next generation of consoles (coming in the relatively near future), but for now, I agree. Programmers are horribly handicapped by inferior hardware (compared to most current gen PC's) in current consoles, but that won't last for too much longer.
 


It certainly won't be more than 4 as a minimum. I may regret saying that, but nothing right now tells me otherwise. We've just recently gotten to the point of dual core being the minimum for most games.

Now, will most games be ABLE to use more than 4 cores in 4 years? Possibly, but that doesn't mean they won't run just as happily on 4, they'll just be written to use more than that, if they're there.

I still think you're hung up on the fact that there's an 8 core consumer CPU out there right now that most people can afford, when there's not (the FX-81xx doesn't count for reasons that have been explained here).

That time won't come for a while.

Besides all that, in 4 years, it'll basically be time for an upgrade anyway, so I'm not sure why you're so worried about it.
 

amdfangirl

Expert
Ambassador
Well according to Steam hardware survey, 50% of all gamers use dual core CPUs as of May 2012. Not going to change drastically soon. If you made a quad-core only game, you would be stopping all but 43% of gamers from playing your game.

http://store.steampowered.com/hwsurvey/cpus/
 

vishalaestro

Distinguished
Jun 29, 2011
1,446
0
19,310

i was worried because i can buy a new cpu only after every four years..i will not have enough money to buy a new one..finally go with six or eight
 

InvalidError

Titan
Moderator

Would the 81xx provide twice as much INT processing power as four identical same-clocked INT cores in whatever other configuration assuming the code is adequately optimized to avoid bottlenecks? Yes. True 8x INT cores design.

You will never achieve perfect scaling at high thread and core count without taking increasingly greater care to avoid creating inter-dependencies and counter-productive interactions between threads and cores no matter how the cores are organized within the CPU or computer. Although the 81xx's internal organization may make fully leveraging the extra INT cores somewhat more challenging in some aspects, the biggest challenge still lies in successfully splitting the software into multiple seamless threads.

You would run into even worse problems with optimizing inter-processor communications/snooping if we were talking about 8-socket type NUMA systems due to much slower busses between CPUs which makes careful software design even more critical for successful scaling, and it gets worse still when you step up to clusters and supercomputer - spending even 1% of the time waiting for other threads effectively prevents the software design from scaling much beyond 100 cores, which would be unacceptable in a 50 000 cores supercomputer.