Are AMD cpus really that bad?

[LAwLz]

2 minutes ago, Space Reptile said:

okay let me clarify

 

FX 9590 , a 220w motherboard exploding beast at 5ghz is on par w/ my i7 2600 , a locked 65w 3.5ghz quadcore w/ HT in benchmarks

That might be true, but then you're not talking about IPC.

[Prysin]

3 minutes ago, LAwLz said:

Well let's see:

1) Dead platform (AM3).

2) Really, really low single core performance.

3) Bad multi-core performance except in some pretty rare cases (and in those rare cases it's better than an i3, but not at i5).

4) Inefficient

 

 

Not even for that. The IPC is so incredibly low that even when all cores can be utilized, a dual core from Intel is usually not that far behind. When less than 8 cores can be utilized, the dual core Intel chip will obliterate the AMD one.

 

 

It's not even close to the IPC of Sandy Bridge.

Sandy Bridge is about 30-40% faster clock for clock.

 

Sandy Bridge vs Vishera. Both at 3.3GHz.

Cinebench 10 (single core):

AMD - 3948

Intel - 5860

 

Cinebench 11.5 (single core):

AMD - 1.1

Intel - 1.47

Single core is not the alpha omega of this world. Pretty much every piece of software these days use more then one thread. We just don't notice it as operating systems has become pretty good at resource management.

 

Bad multicore? No
bad SCALING? YES.

R15 results

 

 

[Dabombinable]

2 minutes ago, Prysin said:

Single core is not the alpha omega of this world. Pretty much every piece of software these days use more then one thread. We just don't notice it as operating systems has become pretty good at resource management.

 

Bad multicore? No
bad SCALING? YES.

R15 results

 

 

Is that the maximum listed Turbo clock speed, or the actual 4 thread clock speed?

[Prysin]

4 minutes ago, Dabombinable said:

Is that the maximum listed Turbo clock speed, or the actual 4 thread clock speed?

my ASUS HERO board will allow for 4c at max turbo (aka 4.4GHz), the "stock" is 4Ghz, but CB only sees the stock, not the turbo. So its actually 4.4GHz... some Intel boards has this function, mainly the higher end ones.

 

EDIT: Also for reference, a i5 6500 = 555 multi thread

[Curious Pineapple]

How long was the Pentium 4 on sale before the Core architecture was released? IIRC, the Athlon absolutely swept the floor with them for years in performance, heat output and power consumption.

[Prysin]

Also, lets not forget FireStrike (normal) scores...

 

8320 = http://www.3dmark.com/fs/7714404

4790k = http://www.3dmark.com/fs/10504294

[ybriK]

5 minutes ago, Curious Pineapple said:

How long was the Pentium 4 on sale before the Core architecture was released? IIRC, the Athlon absolutely swept the floor with them for years in performance, heat output and power consumption.

Yep AMD was beating Intel during the Athlon days until Intel decided to be dirty.

 

https://www.reddit.com/r/pcmasterrace/comments/3s5r4d/is_nvidia_sabotaging_performance_for_no_visual/cwukpuc/

[LAwLz]

7 minutes ago, Prysin said:

Single core is not the alpha omega of this world. Pretty much every piece of software these days use more then one thread. We just don't notice it as operating systems has become pretty good at resource management.

 

Bad multicore? No
bad SCALING? YES.

R15 results
<pictures>

I just want to point out to everyone who might look at this post, that the AMD chip is heavily overclocked and the Intel ones all run at stock settings.

Those images are very misleading if you're not careful and notices that.

 

 

1 minute ago, Prysin said:

Also, lets not forget FireStrike (normal) scores...

 

8320 = http://www.3dmark.com/fs/7714404

4790k = http://www.3dmark.com/fs/10504294

This is also comparing an overclocked AMD chip to stock Intel chip.

 

 

How about you show stock vs stock, or overclocked vs overclock results?

[Curious Pineapple]

4 minutes ago, ybriK said:

Yep AMD was beating Intel during the Athlon days until Intel decided to be dirty.

 

https://www.reddit.com/r/pcmasterrace/comments/3s5r4d/is_nvidia_sabotaging_performance_for_no_visual/cwukpuc/

I had a friend back then who was a total Intel fanboy. His Pentium 2 was obviously better than my K6-2, because it was an Intel chip. He upgraded to a P4 around the same time I upgraded to a Socket A board with a Duron. Again, ignoring the fact that said Duron was a 1.6 ghz low end chip, his was better because Intel. When I managed to get hold of an Athlon XP 3200 Barton and paired it with a Radeon x700, his P4 and Geforce 4 machine was slightly obliterated in performance playing Half Life 2 and other Source based games. Obviously the game engine was crap and badly written and must have been compiled with a shit compiler because an AMD machine cant be faster than a P4.

[Prysin]

15 minutes ago, LAwLz said:

I just want to point out to everyone who might look at this post, that the AMD chip is heavily overclocked and the Intel ones all run at stock settings.

Those images are very misleading if you're not careful and notices that.

 

 

This is also comparing an overclocked AMD chip to stock Intel chip.

 

 

How about you show stock vs stock, or overclocked vs overclock results?

in that case, all i got is these two.

http://www.3dmark.com/compare/fs/7627265/fs/10504294

 

if you want equal clocks (EQ: both running 4.4GHz) then i got these:
http://www.3dmark.com/compare/fs/5200697/fs/4954588

 

 

Then again, it is easy to get a FX 8 to 4.7GHz. Not so much with some intel CPUs. If you do not believe me, click here;
I have my FX and i7 under water: http://hwbot.org/hardware/processor/fx_8320/

http://hwbot.org/hardware/processor/core_i7_4790k/

 

for reference:

FX 6300 = http://hwbot.org/hardware/processor/fx_6300/

FX 6350 = http://hwbot.org/hardware/processor/fx_6350/

FX 8300 = http://hwbot.org/hardware/processor/fx_8300/

FX 8320e = http://hwbot.org/hardware/processor/fx_8320e/

FX 8350 = http://hwbot.org/hardware/processor/fx_8350/

FX 8370 = http://hwbot.org/hardware/processor/fx_8370/

FX 8370e = http://hwbot.org/hardware/processor/fx_8370e/

 

But let us not ignore the fact that some Intel CPUs CAN be overclocked.

i5 3670k = http://hwbot.org/hardware/processor/core_i5_3570k/

i5 4690k = http://hwbot.org/hardware/processor/core_i5_4690k/

i5 5675c = http://hwbot.org/hardware/processor/core_i5_5675c/

i5 6600k = http://hwbot.org/hardware/processor/core_i5_6600k/

 

i7 3770k = http://hwbot.org/hardware/processor/core_i7_3770k/

i7 4770k = http://hwbot.org/hardware/processor/core_i7_4770k/

i7 5775c = http://hwbot.org/hardware/processor/core_i7_5775c/

i7 6700k = http://hwbot.org/hardware/processor/core_i7_6700k/

 

So excuse me if i dont care about "OC" or not. It's so easy to get them to that speed that pretty much anyone should if they have the board to do it. The fact that every single FX CPU is overclockable further cements my position that you either run them overclocked, or you dont bother and go buy a new CPU.

[Phentos]

From a technical perspective, AMD's Bulldozer line and its successors are all hindered by sharing cache memory between core modules. Also each core shares a floating point unit, which slows down their throughput even more. 

 

Basically they share all of their resources among the cores, which only serves to slow all of them down when it comes to clock-for-clock performance. AMD gambled everything on tasks becoming predominantly multithreaded and highly parallel... and lost, as the majority of games and productivity workloads are more dependent on strong single core performance.

[Dabombinable]

9 minutes ago, Phentos said:

From a technical perspective, AMD's Bulldozer line and its successors are all hindered by sharing cache memory between core modules. Also each core shares a floating point unit, which slows down their throughput even more. 

 

Basically they share all of their resources among the cores, which only serves to slow all of them down when it comes to clock-for-clock performance. AMD gambled everything on tasks becoming predominantly multithreaded and highly parallel... and lost, as the majority of games and productivity workloads are more dependent on strong single core performance.

And for the most part, no more than 4 threads (ever seen an FX 8350 with only 1 ALU enabled in each module? The performance increase can be substantial).

25 minutes ago, Prysin said:

in that case, all i got is these two.

http://www.3dmark.com/compare/fs/7627265/fs/10504294

 

if you want equal clocks (EQ: both running 4.4GHz) then i got these:
http://www.3dmark.com/compare/fs/5200697/fs/4954588

 

 

Then again, it is easy to get a FX 8 to 4.7GHz. Not so much with some intel CPUs. If you do not believe me, click here;
I have my FX and i7 under water: http://hwbot.org/hardware/processor/fx_8320/

http://hwbot.org/hardware/processor/core_i7_4790k/

 

for reference:

FX 6300 = http://hwbot.org/hardware/processor/fx_6300/

FX 6350 = http://hwbot.org/hardware/processor/fx_6350/

FX 8300 = http://hwbot.org/hardware/processor/fx_8300/

FX 8320e = http://hwbot.org/hardware/processor/fx_8320e/

FX 8350 = http://hwbot.org/hardware/processor/fx_8350/

FX 8370 = http://hwbot.org/hardware/processor/fx_8370/

FX 8370e = http://hwbot.org/hardware/processor/fx_8370e/

 

But let us not ignore the fact that some Intel CPUs CAN be overclocked.

i5 3670k = http://hwbot.org/hardware/processor/core_i5_3570k/

i5 4690k = http://hwbot.org/hardware/processor/core_i5_4690k/

i5 5675c = http://hwbot.org/hardware/processor/core_i5_5675c/

i5 6600k = http://hwbot.org/hardware/processor/core_i5_6600k/

 

i7 3770k = http://hwbot.org/hardware/processor/core_i7_3770k/

i7 4770k = http://hwbot.org/hardware/processor/core_i7_4770k/

i7 5775c = http://hwbot.org/hardware/processor/core_i7_5775c/

i7 6700k = http://hwbot.org/hardware/processor/core_i7_6700k/

 

So excuse me if i dont care about "OC" or not. It's so easy to get them to that speed that pretty much anyone should if they have the board to do it. The fact that every single FX CPU is overclockable further cements my position that you either run them overclocked, or you dont bother and go buy a new CPU.

So what? I could also overclock my Pentium 4 540J, and 631 65W further than an of those unlocked i7. Oh and BTW, AMD allowing higher overclocks doesn't mean shit when a stock i5 4440 with DDR3 1333 has better IPC than an FX 8320/8350 at almost 5.4GHz. And look at the Core 2 uad, lower clock speed, far higher IPC-and it predates the FX 8150 itself by 3 years, and the 8350 by 4!
 

[JoostinOnline]

AMD doesn't focus on CPU's like Intel.  They also make GPU's, cutting their resources.

[Kc7vwc]

3 hours ago, v0nn_toaster said:

NOTICE *im not trying to start a war on intel vs AMD*

 

I have seen all of the builds and most of the powerful gaming PCs

Rarely use AMD cpus

 

I mean when you goon amazon, a six core 3 ghz CPU for $90 REALLY?

 

SO why is AMD THAT BAD?

 

 

AGAIN *im not trying to start a war on intel vs AMD*

Do you want a Civic, or a Porsche? Both are good at what they do, and get you to the same point. One is just a bit faster & easier on gas.

[Prysin]

6 hours ago, Dabombinable said:

And for the most part, no more than 4 threads (ever seen an FX 8350 with only 1 ALU enabled in each module? The performance increase can be substantial).

So what? I could also overclock my Pentium 4 540J, and 631 65W further than an of those unlocked i7. Oh and BTW, AMD allowing higher overclocks doesn't mean shit when a stock i5 4440 with DDR3 1333 has better IPC than an FX 8320/8350 at almost 5.4GHz. And look at the Core 2 uad, lower clock speed, far higher IPC-and it predates the FX 8150 itself by 3 years, and the 8350 by 4!
 

Lol.  Clock speeds and memory has nothing to do with IPC.  

 

Again you are just posting single thread performance. A metric that isn't alpha omega to computers.  Almost all software these days are using more than one thread. So single core isn't that important,  however scaling is. Aka the ratio between single core and multi core performance 

 

Whenever you post, it is clear you never have a clue, you're just bashing products left and right. 

[Dabombinable]

3 minutes ago, Prysin said:

Lol.  Clock speeds and memory has nothing to do with IPC.  

 

Whenever you post, it is clear you never have a clue, you're just bashing products left and right. 

Yes, it does. The memory because it does make a difference in benchmarks, and the IPC is uite obviously shit because at a significantly lower clock speed an architecture 4 years older than it.

Here is another comparison, note that just under a year separates Vishera from Haswell, and well over 4 years separates Vishera from Wolfdale/Yorkfield.

http://www.cpu-world.com/Compare_CPUs/AMD_FD8350FRW8KHK,Intel_AT80569PJ080N,Intel_CM8064601464800/

 

Then look at the SS again:

 

 

 

Facts+evidence>hearsay+lack of evidence

[Prysin]

2 minutes ago, Dabombinable said:

Yes, it does. The memory because it does make a difference in benchmarks, and the IPC is uite obviously shit because at a significantly lower clock speed an architecture 4 years older than it.

Here is another comparison, note that just under a year separates Vishera from Haswell, and well over 4 years separates Vishera from Wolfdale/Yorkfield.

http://www.cpu-world.com/Compare_CPUs/AMD_FD8350FRW8KHK,Intel_AT80569PJ080N,Intel_CM8064601464800/

 

Then look at the SS again:

 

 

 

Facts+evidence>hearsay+lack of evidence

Ipc is constant, it only varies depending on the instruction used,  not clock speeds or memory speeds. 

 

IPC means Instructions per clock 

Say you seek to execute a SS4e instruction. 

 

Processor A can execute 1 SS4e instruction per clock. 

Processor B however can only do 0.50 SS4e instruction per clock. 

 

In this instance processor B need 2x clocks of processor A to have the same single core performance. 

 

But on top of clocks and ipc comes system latency.  In the case of Cinebench,  the benchmark see no real benefit from memory speeds.  The benchmark is so small it can easily fit in the cache of most modern cpus

[Dabombinable]

10 minutes ago, Prysin said:

Ipc is constant, it only varies depending on the instruction used,  not clock speeds or memory speeds. 

 

IPC means Instructions per clock 

Say you seek to execute a SS4e instruction. 

 

Processor A can execute 1 SS4e instruction per clock. 

Processor B however can only do 0.50 SS4e instruction per clock. 

 

In this instance processor B need 2x clocks of processor A to have the same single core performance. 

 

But on top of clocks and ipc comes system latency.  In the case of Cinebench,  the benchmark see no real benefit from memory speeds.  The benchmark is so small it can easily fit in the cache of most modern cpus

IPC isn't constant, and it varies between architectures as well (there are some exceptions of course, Deschutes & Coppermine have the same IPC as does K6 & K6-2, as well as Skylake & Kabylake), not just the instruction set. Why else does having Vishera perform a task at the same speed as Yorkfield/Wolfdale mean that Vishera has to run at a higher clock speed? Because Vishera takes more cycles to perform the same task using the same instructions at the same clock speed, ergo its IPC is lower.

And BTW-there is a reason why those with 4790K paired with RAM that is faster than DDR3 1333 have their 4790K outperform mine at the same task-Cinebench does benefit from faster RAM.

[Biggerboot]

To put it simply, AMD just needs to update their platform.  It's old.  AMD is not a bad company or manufacturer, but they put their eggs in the wrong basket when it came to the 'core wars'.

[SteveGrabowski0]

Wait a couple of months for Zen and AMD cpus should be awesome again.

[Dabombinable]

3 minutes ago, Biggerboot said:

To put it simply, AMD just needs to update their platform.  It's old.  AMD is not a bad company or manufacturer, but they put their eggs in the wrong basket.

Not a bad company? They've pulled the same kind of shit as Intel and Nvidia-plus they love playing the victim card instead of admitting their faults.

[Techno_Reverend]

8 hours ago, v0nn_toaster said:

NOTICE *im not trying to start a war on intel vs AMD*

 

I have seen all of the builds and most of the powerful gaming PCs

Rarely use AMD cpus

 

I mean when you goon amazon, a six core 3 ghz CPU for $90 REALLY?

 

SO why is AMD THAT BAD?

 

 

AGAIN *im not trying to start a war on intel vs AMD*

yOU ARE TRYING TO CREATE A FLAME WAR, A BAIT THREAD.

[Biggerboot]

3 minutes ago, Dabombinable said:

Not a bad company? They've pulled the same kind of shit as Intel and Nvidia-plus they love playing the victim card instead of admitting their faults.

I'm separating their PR and fans from the actual product.  I wasn't trying to defend them in that sense.  It's just that there's a simple reason nobody is building AMD gaming PCs which is what the OP was asking.  They've released a few stinkers though (FX 9000, reference rx 480).

[Dabombinable]

1 minute ago, Biggerboot said:

I'm separating their PR and fans from the actual product.  I wasn't trying to defend them in that sense.  It's just that there's a simple reason nobody is building AMD gaming PCs which is what the OP was asking.  They've released a few stinkers though (FX 9000).

Dat 220W TDP+$1000 USD launch price. And looking at when it was launched-AMD really hasn't been thinking things through, the FX 9590 launched not that far ahead of Hawswell:

http://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series FX-9590.html

[Biggerboot]

8 minutes ago, Dabombinable said:

Dat 220W TDP+$1000 USD launch price. And looking at when it was launched-AMD really hasn't been thinking things through, the FX 9590 launched not that far ahead of Hawswell:

http://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series FX-9590.html

My feeling is that they just wanted a certain corner of the market covered without developing a whole new architecture, but yeah, it wasn't well thought out.

 

I think they'll turn things around with Ryzen though.  They may be in a position where they can ride this platform out for a good while because of the slowing die shrinks.  Intel is going to increase the cores because of it.