The Best AMD GPU

[Moonzy]

Just now, Citadelen said:

Clock speed is directly related to IPC.

no, clock speed is clock speed, IPC is IPC

 

lets say processor A can process 2 IPS and is 1hz and processor B can process 1 IPS and is 2hz

 

theoretically, both processor would be the same in term of performance since they both process 2 instructions per second

[Citadelen]

1 minute ago, Majestic said:

They're not. Now granted, all those games are very CPU heavy and really bring out the worst.

http://www.gamersnexus.net/game-bench/1911-gta-v-cpu-benchmark-4790k-3570k-9590-more

[ZeusXI]

Just now, Moonzy said:

no, clock speed is clock speed, IPC is IPC

 

lets say processor A can process 2 IPS and is 1hz and processor B can process 1 IPS and is 2hz

 

theoretically, both processor would be the same in term of performance since they both process 2 instructions per second

so....10 ipc for 1 GHz would be better than 5ipc for 1GHz. what your saying is more IPC per GHz means its more efficient with each added GHz?

 

[Moonzy]

1 minute ago, ZeusXI said:

how many IPC does an I7-4790k have?

no idea  and i cant tell you to google it or ill get in trouble

 

but frankly it doesnt matter, real world benchmarks are more important imo

[Majestic]

2 minutes ago, Citadelen said:

http://www.gamersnexus.net/game-bench/1911-gta-v-cpu-benchmark-4790k-3570k-9590-more

Pandering. Even digitalfoundry shows the FX series getting destroyed in GTA 5. Gamernexus are hacks.

 

And not relevant, as I said. Seperate testing.

[Moonzy]

Just now, ZeusXI said:

so....10 ipc for 1 GHz would be better than 5ipc for 1GHz. what your saying is more IPC per GHz means its more efficient with each added GHz?

 

efficient in term of time yes, since it process about double the instruction in a given time

but efficient in electricity, that depends on architecture

[i_build_nanosuits]

2 minutes ago, ZeusXI said:

how many IPC does an I7-4790k have?

it's not as simple...you can't ''quantify'' IPC like that...it's too complicated...IPC stand for ''instructions per cycle'' and those instructions can be of many types...floating point, double precision floating point, integer based operations...so it's not a number you can come up with...only way is to check benchmarks for example...

But YES, ''haswell level IPC'' for Zen would already be very good if you ask me...if priced competitively and offering more threads for example it could be a win for AMD...IMHO skylake is out of reach...

[Citadelen]

1 minute ago, Moonzy said:

no, clock speed is clock speed, IPC is IPC

 

lets say processor A can process 2 IPS and is 1hz and processor B can process 1 IPS and is 2hz

 

theoretically, both processor would be the same in term of performance since they both process 2 instructions per second

IPC isn't a set figure, the higher a CPU is clocked, the more instructions it can process, the better the architecture, the more instructions it can process, cache latency is also important. IPC isn't static, as on different workloads a CPU will process different amounts of instructions.

[i_build_nanosuits]

3 minutes ago, ZeusXI said:

so....10 ipc for 1 GHz would be better than 5ipc for 1GHz. what your saying is more IPC per GHz means its more efficient with each added GHz?

 

yes, exactly...that's why a 5.0ghz FX-9590 is slower than a 3.2ghz i5-4460 for example...IPC.

[Moonzy]

Just now, i_build_nanosuits said:

it's not as simple...you can't ''quantify'' IPC like that...it's too complicated...IPC stand for ''instructions per cycle'' and those instructions can be of many types...floating point, double precision floating point, integer based operations...so it's not a number you can come up with...only way is to check benchmarks for example...

But YES, ''haswell level IPC'' for Zen would already be very good if you ask me...if priced competitively and offering more threads for example it could be a win for AMD...IMHO skylake is out of reach...

yea, some instruction requires 1 cycle and some requires up to 8 or even more

so it really depends

[Majestic]

Quote

In computer architecture, instructions per cycle (IPC) is one aspect of a processor's performance: the average number of instructions executed for each clock cycle. It is the multiplicative inverse of cycles per instruction.

Or you guys just list the wikipedia quote and have everyone understand.

 

It's clearly disconnected from the clockspeed, only multiplied with.

[Moonzy]

1 minute ago, Citadelen said:

IPC isn't a set figure, the higher a CPU is clocked, the more instructions it can process, the better the architecture, the more instructions it can process, cache latency is also important. IPC isn't static, as on different workloads a CPU will process different amounts of instructions.

thats assuming both processing the same data, of course IPC varies upon a lot of variables like what instruction is the cpu processing at the current time when it receive the next, perhaps the logic algorithm can be used parallel with the arithmetic algorithm, theres a lot factors and i cant possibly cover them all here

 

but clock speed does not affect IPC, clock speed only affects the cycle time, not the amounts of things that can be done in a cycle

[Citadelen]

5 minutes ago, Majestic said:

Pandering. Even digitalfoundry shows the FX series getting destroyed in GTA 5. Gamernexus are hacks.

 

And not relevant, as I said. Seperate testing.

Well, thank you for saving me the trouble of finding more benchmarks to post, Digital Foundry shows a difference in averages of 10 fps, as it does in Gamernexus, not 20 like in your ones.

[i_build_nanosuits]

AMD claim a +40% increase in IPC with Zen based processors when compared to their up most recent CPU architecture (which i think is steamroller) so this would put them around haswell or a bit bellow...which would be pretty good for budget builders...if for example they could deliver this in a 6 thread capable chip for around 175$ or so and have good motherboard with modern features and overclocking options for around 100$...this could be a good alternative to intel's i5-6400/6500 for example.

(but then again, what AMD claim AMD usually does not deliver...the FuryX was supposedly faster than the 980ti and very overclockers friendly...they said they were overclockers dream chips...but they overclock like shit...like...not at all...so IMHO we will se a 20% to 30% increase, which honestly would already be quite good so long as it's cheaper)

[Moonzy]

1 minute ago, i_build_nanosuits said:

AMD claim a +40% increase in IPC when compared to their up most recent CPU architecture (which i think is steamroller) so this would put them around haswell or a bit bellow...which would be pretty good for budget builders...if for example they could deliver this in a 6 thread capable chip for around 175$ or so and have good motherboard with modern features and overclocking options for around 100$...this could be a good alternative to intel's i5-6400/6500 for example.

so... amd zen cant compete with the 5960x and the likes?

cant tell you how disapointed im, like an asian dad with his son

"talk to me when you beat 5960x"

[Citadelen]

2 minutes ago, Moonzy said:

thats assuming both processing the same data, of course IPC varies upon a lot of variables like what instruction is the cpu processing at the current time when it receive the next, perhaps the logic algorithm can be used parallel with the arithmetic algorithm, theres a lot factors and i cant possibly cover them all here

 

but clock speed does not affect IPC, clock speed only affects the cycle time, not the amounts of things that can be done in a cycle

Okay I'm changing my stance, seeing as IPC isn't a set figure it can't be measured.

[Majestic]

5 minutes ago, Citadelen said:

Well, thank you for saving me the trouble of finding more benchmarks to post, Digital Foundry shows a difference in averages of 10 fps, as it does in Gamernexus, not 20 like in your ones.

 

Yup, only 10fps. Confirmation biased nobend.... seriously can't stand this behaviour.

[i_build_nanosuits]

Just now, Moonzy said:

so... amd zen cant compete with the 5960x and the likes?

cant tell you how disapointed im, like an asian dad with his son

"talk to me when you beat 5960x"

i really don't see that happening...IMHO i think it will compete with the i5-4690 for example...something along those lines...probably with more ''cores'' again split in modules...it can be good though...like i said it will depend on pricing and motherboards.

[Moonzy]

1 minute ago, i_build_nanosuits said:

i really don't see that happening...IMHO i think it will compete with the i5-4690 for example...something along those lines...probably with more ''cores'' again split in modules...it can be good though...like i said it will depend on pricing and motherboards.

but thats competing with architecture from 4 or 5 years ago... wut

[Citadelen]

3 minutes ago, Majestic said:

-snip-

Ah, I was looking at the 2500k, but still, there's not i7 6700k on that, so I still think PClab has inflated figures.

[i_build_nanosuits]

Just now, Moonzy said:

but thats competing with architecture from 4 or 5 years ago... wut

no the haswell CPU's are from 2014 if i recall correctly...and intel has insane budget for RND...not AMD...they start from so far back you know...they havnt done something good in years...reaching haswell in one generation would already be an immense feat.

[Citadelen]

3 minutes ago, i_build_nanosuits said:

i really don't see that happening...IMHO i think it will compete with the i5-4690 for example...something along those lines...probably with more ''cores'' again split in modules...it can be good though...like i said it will depend on pricing and motherboards.

AMD's abandoning CMT...

[i_build_nanosuits]

4 minutes ago, Majestic said:

 

Yup, only 10fps. Confirmation biased nobend.... seriously can't stand this behaviour.

i see 24FPS from intel's best performing CPU in games...and this is with a single GPU...

[Majestic]

Just now, Citadelen said:

Ah, I was looking at the 2500k, but still, there's not i7 6700k on that, so I still think PClab has inflated figures.

Look at the clockspeeds on that 6700K. 4,7ghz, not a stock 6600K like the DF video. Crysis3 uses HT, so it's hyperthreading (+25-30%) + 30% OC.

 

They're good numbers.

[i_build_nanosuits]

Just now, Citadelen said:

AMD's abandoning CMT...

that could be good or bad...it would be great to still have those two integer processing units per core though...but so long as it does not mean having a CPU with an IPC figure from 10 years ago like the AMD FX line...