AMD FX 8350 vs i5 4670k

[patrickjp93]

I didn't think their was such a thing 0,o.

 

Although admittedly I have no idea how what the difference between the AMD 8 core and Intel's Hyperthreading. I'm guessing it's to do with their architectures.

Hyperthreading is basically super-intelligent scheduling and rapid context-switching (Operating System terminology). Intel makes available 8 schedulers which try to use different resources of the chip for different tasks as best they can. While there aren't suddenly 2 of every logic resource available (effectively a true 8-core chip), the given resources can be used such as to simulate it (to a point). The result is higher heat output, but a smaller overall die to manage as opposed to AMD's modular strategy (also not a true 8-core solution).

 

The Opteron and Xeon lines were the first to have true 8-core solutions (servers/workstations only), but the 5960x will be the first PC-grade true 8-core solution.

[vm'N]

I didn't think their was such a thing 0,o.

 

Although admittedly I have no idea how what the difference between the AMD 8 core and Intel's Hyperthreading. I'm guessing it's to do with their architectures.

Hyper-threading is Intels method to implement SMT. This is done by the fetch stage (the fetch is fetching the complex instructions). So it can fetch two instead of a single.

After then it simply doesn't matter as the complex instruction are decoded (doing the decode stage) to RISC-a-like instructions (simpler instructions). From there the execution stage doesn't care what thread is what. It just execute the instruction to certain data.

Note: IBM have a 8-way SMT processor. A processor where a single core can handle 8 threads simultaneously.

AMD introduced CMT. CMT is again a fetch fetching for two complex instruction. In case of piledriver the complex instructions are send to the shared decoders. AMD is using a shared FPU. They are calling this technology. To simplify things, they simply putted two wired to the FPU. So they are both queueing to the same FPU.

The major difference between CMT and SMT is basically that you have a more "fixed" set of resources with CMT, where SMTs resources are shared more dynamically.

CMT is about duplicating certain stages of the CPU to increase throughput and be spaceefficient.

SMT is about utilizing the execution stage better, to increase throughput and be powerefficient.

[FizzyFantom]

This has now just turned into a fanboy war, my FX-8350 @ 4.8GHz runs everything I throw at it with 60+ fps (except Crysis 3)

It always irritates me to see fanboy "opinions" come into threads trying to convince people. I have seen "AMD is just crap and runs hot" before and it simply isn't true. That being said the AMD FX 8350 is older and won't perform as well in single threaded performance as the 4770K (or 4790K), however it may outperform them in tasks that benefit from multiple CPU cores such as encoding video (but I have no doubts that someone will shout at me for suggesting this...)

I would personally recommend the 4790K as many others have, but the 8350 is a good option too, just note that it won't perform as well in games.

[vm'N]

Hyperthreading is basically super-intelligent scheduling and rapid context-switching (Operating System terminology). Intel makes available 8 schedulers which try to use different resources of the chip for different tasks as best they can. While there aren't suddenly 2 of every logic resource available (effectively a true 8-core chip), the given resources can be used such as to simulate it (to a point). The result is higher heat output, but a smaller overall die to manage as opposed to AMD's modular strategy (also not a true 8-core solution).

 

The Opteron and Xeon lines were the first to have true 8-core solutions (servers/workstations only), but the 5960x will be the first PC-grade true 8-core solution.

Context switching is just as effective on non-SMT chips as SMT chips. SMT is simultaneously, not two threads running separately, constantly switching threads (if that is what you meant, I'm not 100% sure).

[patrickjp93]

Context switching is just as effective on non-SMT chips as SMT chips. SMT is simultaneously, not two threads running separately, constantly switching threads (if that is what you meant, I'm not 100% sure).

but the switching happens more intelligently, so as to find a place where tasks are independent (branch prediction is one example of intelligent) and get ahead of the overall process with out of order execution to find those places where it is convenient and independent to run the same instruction on two sets of data or send independent tasks off to different resources in a manner to increase throughput.

[vm'N]

but the switching happens more intelligently, so as to find a place where tasks are independent (branch prediction is one example of intelligent) and get ahead of the overall process with out of order execution to find those places where it is convenient and independent to run the same instruction on two sets of data or send independent tasks off to different resources in a manner to increase throughput.

Are we talking about the CU allocating more resource to one of the threads? If so, that have nothing to do with thread-switching. This is how SMT behave. SMT have are sharing almost all resource dynamically with eachother, but both threads will mostlikely run simultaneously. You will have a hard time completely starve a threads (it was at a point a problem, however this have now been fixed).

But I do think I get the context. Operative systems now have better support for SMT. It is intellifently queueing two complex instructions which won't try to starve the other thread.

[Faa]

TDP is only the theoretical thermal output limit.

TDP is calculated through alot of variables on the architecture and much more.

Theoretical limit? That would include the thermal output in worst case scenario's, TDP is only measured on normal applications not stuff like prime95 from the heatspreader (Tcase). 

[Faa]

All you did was posted benchmarks of poorly threaded applications, Intel has for a while had stronger single core performance, show me a benchmark of a well threaded game where the 8350 falls "100% behind" as you previously claimed.

I've included them here: http://linustechtips.com/main/topic/184204-amd-fx-8350-vs-i5-4670k/page-3#entry2490056

 

[vm'N]

Theoretical limit? That would include the thermal output in worst case scenario's, TDP is only measured on normal applications not stuff like prime95 from the heatspreader (Tcase).

No, they don't use ANY kind of software to determine the TDP.

They calculate it by the hardware specifications.

This is also why Intel and AMD can call out a decent guess on their future architectures TDP.

[Faa]

No, they don't use ANY kind of software to determine the TDP.

They calculate it by the hardware specifications.

This is also why Intel and AMD can call out a decent guess on their future architectures TDP.

They don't calculate the TDP with software, that's not what I've said. All I am saying is that TDP isn't measured in worst case scenario's like stresstests such as prime95. The thermal output is only measured under normal usage such as productivity things etc. 

Hardware specifications, honestly you're just making it hard. http://www.intel.com/content/www/us/en/intelligent-systems/intel-technology/thermal-design-embedded-appl-paper.html

Page 18 they show how they calculate the cooling performance you need. 

[vm'N]

They don't calculate the TDP with software, that's not what I've said. All I am saying is that TDP isn't measured in worst case scenario's like stresstests such as prime95. The thermal output is only measured under normal usage such as productivity things etc. 

Hardware specifications, honestly you're just making it hard. http://www.intel.com/content/www/us/en/intelligent-systems/intel-technology/thermal-design-embedded-appl-paper.html

Page 18 they show how they calculate the cooling performance you need.

The TDP isn't measured with any kind of workloads. It is measured by some hardware information.

The TDP is calculated through a algorithm using the information. They are not physically testing them.

I am also pretty sure Intel and AMD keeps those algoritm secret for the public. (Not sure)

EDIT: Forget to add that they might adjust the algorithm to a normal use instead of running all at max. They simply are not sitting in a lab with a PC doing workloads, that is my point.

[Razzaa]

The TDP isn't measured with any kind of workloads. It is measured by some hardware information.

The TDP is calculated through a algorithm using the information. They are not physically testing them.

I am also pretty sure Intel and AMD keeps those algoritm secret for the public. (Not sure)

EDIT: Forget to add that they might adjust the algorithm to a normal use instead of running all at max. They simply are not sitting in a lab with a PC doing workloads, that is my point.

Nobody in the general public knows what methods they use to determine TDP. It is measured though using normal applications and not stressing. I believe a chip is capable of producing heat up to 2x more than the set TDP. I could be wrong though.

[Faa]

Nobody in the general public knows what methods they use to determine TDP. It is measured though using normal applications and not stressing. I believe a chip is capable of producing heat up to 2x more than the set TDP. I could be wrong though.

Or even lower. A 4820K has 130W TDP just the same as the 4930K.. That thing won't manage to pull 130W: http://tpucdn.com/reviews/AMD/FX-8350_Piledriver_Review/images/power_eps_load.gif (measured from 8pin socket to exclude psu's AC/DC efficiency crap)

[Razzaa]

Or even lower. A 4820K has 130W TDP just the same as the 4930K.. That thing won't manage to pull 130W: http://tpucdn.com/reviews/AMD/FX-8350_Piledriver_Review/images/power_eps_load.gif (measured from 8pin socket to exclude psu's AC/DC efficiency crap)

TDP is heat dissipation measured in watts, obviously It is not the same as actual power consumption. AMD determines the TDP while it is @ full load. Intel measures differently. 130W wouldnt be out of reach by any means, although probably OC'd

[Sintezza]

Who cares about TDP?

 

Thermal design power (called TDP)  is just the amount of heath in watts (with a safe margin) which your cpu cooler should be capable to dispace from the processor.

 

It doesn´t say a shit about how manny WAAAATS a cpu is actualy using.

You are unable to measure how manny watt's a cpu is actualy using.

You can only measure how manny watt´s a complete system uses, and then make a calculation, but its still no where near accurate!

 

I dont know why people still talking about this crap.

[vm'N]

TDP is simply the max theoretical output of heat by the processor in watts.

It is not directly connected to the power-consumption or heatoutput.

What I think you are forgetting is that technologies similar to turbo boost are allowed to exceed their TDP.

[bamerx98]

To be honest the amd platform is so old now that I think you should just get the i5 4670k, I personally have experience with a 8350 @5+ ghz and it is not nearly as a good as the new i7 like there was a really big boost from coming from the 8350 for games DayZ and Battlefield 

[Lazmarr]

Bloodkil933, on 24 Jul 2014 - 12:07 AM, said:

All you did was posted benchmarks of poorly threaded applications, Intel has for a while had stronger single core performance, show me a benchmark of a well threaded game where the 8350 falls "100% behind" as you previously claimed.

I've included them here: http://linustechtips.com/main/topic/184204-amd-fx-8350-vs-i5-4670k/page-3#entry2490056

None of those benchmarks show that the FX 8350 is "100% behind".

[Razzaa]

None of those benchmarks show that the FX 8350 is "100% behind".

Lol dont get him started........

[APrettyCoolWalrus]

Now what about the FX 9590 vs the i7 4770K?

9590 is basically a higher clocked 8350, still sucks compared to any i5 or i7.

[Razzaa]

To be honest the amd platform is so old now that I think you should just get the i5 4670k, I personally have experience with a 8350 @5+ ghz and it is not nearly as a good as the new i7 like there was a really big boost from coming from the 8350 for games DayZ and Battlefield

You recommend the i5 but then give an opinion based on i7 Vs 8350..........

[Razzaa]

9590 is basically a higher clocked 8350, still sucks compared to any i5 or i7.

9590 bests a i5 anyday......

[APrettyCoolWalrus]

9590 bests a i5 anyday......

uwotm9

oh wait nvm he's right

eh the more mistakes I make the more posts I get fixing them :3

[bamerx98]

You recommend the i5 but then give an opinion based on i7 Vs 8350..........

Yes, I know, but with my experience with the 8350 it gave me a sour taste in my mouth, the single core means so much to games these days, in till games are fully multi-threaded there is no point in buying this outdated 32nm processor. Games just aren't ready for 8 core CPU's yet. 

[Bloodkil933]

Yes, I know, but with my experience with the 8350 it gave me a sour taste in my mouth, the single core means so much to games these days, in till games are fully multi-threaded there is no point in buying this outdated 32nm processor. Games just aren't ready for 8 core CPU's yet. 

Although I agree, it also depends on the optimization of the game, and with some companies PC Ports, they're pretty badly optimized. Just look at Ubisoft and Watch Dogs, that ran like crap on any system.