AMD real talk

[Coaxialgamer]

3 minutes ago, Gikero said:

I see a lot of people who express how poorly they consider AMD's FX8XXX series, but I've never seen anyone compare to a Pentium 4. Damn.

yeah i've seen a few people talk about that. To be fair though , the fx cpus aren't as bad as the pentium 4 , which was a Huge failure ( despite selling well because of how intel forced tier one builders to use it )

[Flowey]

43 minutes ago, branden_lucero said:

NVidia developed an asshole attitude which is what made me stick with AMD for GPUs, not to mention the "performance percentage" between  a non TI, a TI and a Titan card recently. 

Yeah that's a fair point, lately they fell like they know there on top and acts like it, but I cannot believe the whole company's like that. I'm really trying to give them a chance here. like, do sumting fast 'cause otherwise I won't even be thinking about Nvidia when I need to uprade later down the road.

 

42 minutes ago, Gikero said:

I see a lot of people who express how poorly they consider AMD's FX8XXX series, but I've never seen anyone compare to a Pentium 4. Damn.

is that SHODAN by the way?

[KemoKa]

Just now, Flowey said:

Wow, I chuckled quite a bit here sir, ¨Greased lightning¨

 

Still, if their architecture is so old, how can it still keep up with Nvidia's? 

because Nvidia shilled and hyped the hell out of Maxwell, screaming at the top of its lungs about how wonderful Maxwell was going to be. And it... wasn't that much more spectacular. All AMD had to do was release the Fury lineup and they were pretty much set. For the price, GCN has held its ground against Nvidia for even longer. My 7970 is 4 years old and it still stands toe-to-toe with many of Nvidia's mid to high-end cards from the past few years. 680, 770, 960, it beats them all in most level benchmarks.

And GCN architectures eats OpenCL applications for breakfast. Have you seen a Hawaii/Tahiti/Fiji XT card go at something that uses OpenCL? They're absolute monsters. Look up what happened during the scrypt mining craze of Q1 2014. Radeon cards were being price-gouged because they were so proficient at OpenCL that they could mine several to a dozen times faster than Nvidia's equivalent Kepler cards.

 

There's a reason why the 295x2 is still the fastest single-slot card on the planet.

[AnnoyedShelf]

1 hour ago, Flowey said:

Alright so I'm tired of hearing either Nvidia's or Intel's fan swearing by their products and how much AMD are 2-3 years behind everybody. I myself opted for the R9 390, not because I'm a fanboy, but because for the sake of future proofing my build. So, my question is, are AMD really that far behing in GPU/CPU technology? I know for sure that their GPU's are lacking in the efficiency departement but otherwise I cannot find any explanation of how AMD are lacking behing their competitors. So yeah, thanks for your time!

nope.  In fact it looks like Polaris will push them further than where Nvidia is.

 

 

[TidaLWaveZ]

1 hour ago, Flowey said:

Alright so I'm tired of hearing either Nvidia's or Intel's fan swearing by their products and how much AMD are 2-3 years behind everybody. I myself opted for the R9 390, not because I'm a fanboy, but because for the sake of future proofing my build. So, my question is, are AMD really that far behing in GPU/CPU technology? I know for sure that their GPU's are lacking in the efficiency departement but otherwise I cannot find any explanation of how AMD are lacking behing their competitors. So yeah, thanks for your time!

 

I thought someone should finally mention that with the exception of the 980ti the higher end Nvidia GPU's were released a lot earlier than their AMD counterparts.

 

I'm not taking either side, I get what I can get a deal on at the time I'm buying.

[Citadelen]

1 hour ago, DevilishBooster said:

Their architecture is several years behind, and that is why they are so lacking in effeciency. Hopefully the new line of AMD products will bring them back into the fighting ring.

Wrong statement is wrong. Ripping out compute doesn't count as efficiency.

[Flowey]

7 hours ago, TidaLWaveZ said:

 

I thought someone should finally mention that with the exception of the 980ti the higher end Nvidia GPU's were released a lot earlier than their AMD counterparts.

 

I'm not taking either side, I get what I can get a deal on at the time I'm buying.

That's where I'm sitting trying to invest my money into the best price/performance ratio I can get.

[Prysin]

41 minutes ago, Flowey said:

Wow 100% agree, look at TrueAudio, were is it today? Consoles maybe?

True Audio is not what you think it is.

It isnt a "sound card". It is a suite to create 3D sound.

If you own a AMD GPU, and has tried to hook it up with a 5.1 surround system, you would realize how powerful it is...

[Citadelen]

2 minutes ago, Prysin said:

True Audio is not what you think it is.

It isnt a "sound card". It is a suite to create 3D sound.

If you own a AMD GPU, and has tried to hook it up with a 5.1 surround system, you would realize how powerful it is...

It isn't a suite, it's a co-processor.

[Flowey]

7 hours ago, Prysin said:

True Audio is not what you think it is.

It isnt a "sound card". It is a suite to create 3D sound.

If you own a AMD GPU, and has tried to hook it up with a 5.1 surround system, you would realize how powerful it is...

Nah bud I knew that, I also know it's going to be implemented to a certain degree in VR pretty darn soon I believe. No offense taken do, that short sentence sure was misleading

 

[Prysin]

30 minutes ago, Gikero said:

I see a lot of people who express how poorly they consider AMD's FX8XXX series, but I've never seen anyone compare to a Pentium 4. Damn.

Netburst and BUlldozer is NOT THE FUCKING SAME....

 

Fun fact, but nowehere is it stated that Netburst was CMT... CMT itself was "invented" in 1996 by DEC under the name "Clustered Integer Core". Netburst is never mentioned as CIC or CMT. It is what happened when intel fucked up their HT.

 

The reason why Netburst fucked up was TDP. It had TWO ALUs, but unlike Bulldozer, these ran at TWICE the clock rate of the rest of the CPU.

If a "CPU" was 3GHz, the ALUs would each run at 6GHz...

 

What is an ALU? What does it do?
 

Quote

Functions

A number of basic arithmetic and bitwise logic functions are commonly supported by ALUs. Basic, general purpose ALUs typically include these operations in their repertoires:

Arithmetic operations

• Add: A and B are summed and the sum appears at Y and carry-out.
• Add with carry: A, B and carry-in are summed and the sum appears at Y and carry-out.
• Subtract: B is subtracted from A (or vice-versa) and the difference appears at Y and carry-out. For this function, carry-out is effectively a "borrow" indicator. This operation may also be used to compare the magnitudes of A and B; in such cases the Y output may be ignored by the processor, which is only interested in the status bits (particularly zero and negative) that result from the operation.
• Subtract with borrow: B is subtracted from A (or vice-versa) with borrow (carry-in) and the difference appears at Y and carry-out (borrow out).
• Two's complement (negate): A (or B) is subtracted from zero and the difference appears at Y.
• Increment: A (or B) is increased by one and the resulting value appears at Y.
• Decrement: A (or B) is decreased by one and the resulting value appears at Y.
• Pass through: all bits of A (or B) appear unmodified at Y. This operation is typically used to determine the parity of the operand or whether it is zero or negative.

Bitwise logical operations

• AND: the bitwise AND of A and B appears at Y.
• OR: the bitwise OR of A and B appears at Y.
• Exclusive-OR: the bitwise XOR of A and B appears at Y.
• One's complement: all bits of A (or B) are inverted and appear at Y.

Bit shift operations

ALU shift operations cause operand A (or B) to shift left or right (depending on the opcode) and the shifted operand appears at Y. Simple ALUs typically can shift the operand by only one bit position, whereas more complex ALUs employ barrel shifters that allow them to shift the operand by an arbitrary number of bits in one operation. In all single-bit shift operations, the bit shifted out of the operand appears on carry-out; the value of the bit shifted into the operand depends on the type of shift.

• Arithmetic shift: the operand is treated as a two's complement integer, meaning that the most significant bit is a "sign" bit and is preserved.
• Logical shift: a logic zero is shifted into the operand. This is used to shift unsigned integers.
• Rotate: the operand is treated as a circular buffer of bits so its least and most significant bits are effectively adjacent.
• Rotate through carry: the carry bit and operand are collectively treated as a circular buffer of bits.

 

 

Quote

Complex operations

Although an ALU can be designed to perform complex functions, the resulting higher circuit complexity, cost, power consumption and larger size makes this impractical in many cases. Consequently, ALUs are often limited to simple functions that can be executed at very high speeds (i.e., very short propagation delays), and the external processor circuitry is responsible for performing complex functions by orchestrating a sequence of simpler ALU operations.

For example, computing the square root of a number might be implemented in various ways, depending on ALU complexity:

• Calculation in a single clock: a very complex ALU that calculates a square root in one operation.
• Calculation pipeline: a group of simple ALUs that calculates a square root in stages, with intermediate results passing through ALUs arranged like a factory production line. This circuit can accept new operands before finishing the previous ones and produces results as fast as the very complex ALU, though the results are delayed by the sum of the propagation delays of the ALU stages.
• Iterative calculation: a simple ALU that calculates the square root through several steps under the direction of a control unit.

The implementations above transition from fastest and most expensive to slowest and least costly. The square root is calculated in all cases, but processors with simple ALUs will take longer to perform the calculation because multiple ALU operations must be performed.

 

 

As you can see, an ALU can do A LOT of CPU tasks, if you design it properly. Whether AMD did this in bulldozer we will NEVER know, because that is basically the most secret of secret "sauce" to their CMT linup. IF their ALUs indeed CAN make complex calculations, then that may explain SOME of the latencies within Bulldozer and also why it isnt as bad as it SHOULD be.

 

back to netburst.

Northwood was the only family of P4s that had hyperthreading, BUT, unlike Bulldozer, it is not CMT...

Here is a diagram of Netburst:

Here is bulldozer

Here is Core architecture

And here is a picture of a Clustered Integer Core.
https://en.wikipedia.org/wiki/Alpha_21264#/media/File:Alpha_21264.png

 

As you can see, there is a difference, not a huge one though. WHich is another fun fact. "AMDs" CMT is actually a basterdization of Simultaneous Multithreading and Clustered Integer Core. It has half the setup for a SMT configuration, but share the other half with a CIC configuration. Which is why it is called CMT. One could perhaps argue that Netburst is CMT, but it lacks some of the shared

 

As you can see, the Core architecture ALSO features multiple ALUs...

 

So what is netburst?

It is what you get when you fuck up Hyperthreading....

 

What is CMT?
It is what you get when you try to take the best of two worlds and fuck it up.

 

what is SMT?

An old ass idea that should have been implemented half a century ago rathen then 14 years ago.

 

 

shared cache is an issue, because AMD used a slower cache. Intels cache operates at much tighter latencies and much higher speeds. If AMD had had faster and more responsive cache, ideally, you shouldnt have noticed much. But faster = more clocks = hotter.
There is also the question of OS optimization. In the beginning it was evident that Windows 7 and earlier OSs had no clue how FX operated, and as a result though it was a quad "core" rather then 4 modules/8 threads.

 

Then there is a HUGE "maybe".... Intel has a set of liberaries specifically tuned and tailored to THEIR products. Whilst intel HAS to optimize a little bit for AMDs sake (after they got busted for cheating in benchmarks) they dont really do much to help AMD. A LOT of software uses parts of those liberarires. Whilst it wont be hurting AMD much, it falls under a saying we got here where i live "Mange bekker små blir en stor å"... which translates to "many small streams form a large river".... I honestly think there is more then ONE issue with bulldozer. Probably 20+ issues. That in varying degree has been fixed (Piledriver -> Steamroller -> Excavator).

If you remember the thread of AMD bitching about some benchmark strongly favoring intel over them, well, said benchmark is using Intels highly optimized liberaries. The same liberaries (although updated) Intel used when they were busted for cheating in benchmarks.

Is it a coincidence? I think not. but why?

Most of the stuff we do daily ISNT very complex. Rather, it is incredibly simple stuff. Computer AI, quicktime event, typing in word, browsing the web and such isnt complex at all, and should be well within the capability of an basic ALU to compute. There is ZERO reason why the ALU would be struggling in these tasks, and AMD knew this.

So what is the culprint?

Latencies... AMDs CMT implementation is VERY slow. Not the ALUs, but the combined latencies of Cache, IMC, Hyper-Transport (although this one aint THAT slow as it maxes out at 51GB/s bandwidth).

 

Looking at how AMD changed their CMT over time, it is obvious that things HAVE improved.

 

In all honesty, just going from Bulldozer to Steamroller, we can see clear changes. The splitting of decode and dispatch has probably helped usher along a much greater efficiency and cut down latency from queueing issues. Still, they got a way to go.

 

It also doesnt help that pretty much every modern instruction is made by Intel Corp.... With the exception of AMD64 (x86-64).... so yeah. Conspiracies or not, aslong as a instruction CAN run on AMD hardware, no matter how unoptimized, it is "ok" in the eyes of the government to not hammer intel for anti competitiveness.
So if it so happens that AMD has any sort of efficiency overhead using Intel instructions (which we will never know), then that doesnt help AMD.

 

In the end, ill say that whatever plagues Bulldozer, it probably isnt entirely AMDs fault, although 75%+ of the blame is definetively theirs to take. BUT, there IS outside factors not helping their case for sure.

 

 

Just for the hell of it. here is K10 aka Phenom II

 

 

This is copies from a private discussion i had with someone. This is why it may seem a bit weird in the way it is laid out

[TidaLWaveZ]

Just now, Flowey said:

That's where I'm sitting trying to invest my money into the best price/performance ratio I can get.

 

Always the best decision in my opinion(assuming the software/drivers are up to par).

 

I just thought it was strange that nobody mentioned anything about the difference in release dates between AMD and Nvidia's cards whether it be in AMD or Nvidia's favor.

 

 

[Flowey]

10 minutes ago, Prysin said:

Netburst and BUlldozer is NOT THE FUCKING SAME....

 

Fun fact, but nowehere is it stated that Netburst was CMT... CMT itself was "invented" in 1996 by DEC under the name "Clustered Integer Core". Netburst is never mentioned as CIC or CMT. It is what happened when intel fucked up their HT.

 

The reason why Netburst fucked up was TDP. It had TWO ALUs, but unlike Bulldozer, these ran at TWICE the clock rate of the rest of the CPU.

If a "CPU" was 3GHz, the ALUs would each run at 6GHz...

 

What is an ALU? What does it do?
 

 

 

 

 

As you can see, an ALU can do A LOT of CPU tasks, if you design it properly. Whether AMD did this in bulldozer we will NEVER know, because that is basically the most secret of secret "sauce" to their CMT linup. IF their ALUs indeed CAN make complex calculations, then that may explain SOME of the latencies within Bulldozer and also why it isnt as bad as it SHOULD be.

 

back to netburst.

Northwood was the only family of P4s that had hyperthreading, BUT, unlike Bulldozer, it is not CMT...

Here is a diagram of Netburst:

Here is bulldozer
 

Here is Core architecture
 

And here is a picture of a Clustered Integer Core.
 

As you can see, there is a difference, not a huge one though. WHich is another fun fact. "AMDs" CMT is actually a basterdization of Simultaneous Multithreading and Clustered Integer Core. It has half the setup for a SMT configuration, but share the other half with a CIC configuration. Which is why it is called CMT. One could perhaps argue that Netburst is CMT, but it lacks some of the shared

 

As you can see, the Core architecture ALSO features multiple ALUs...

 

So what is netburst?

It is what you get when you fuck up Hyperthreading....

 

What is CMT?
It is what you get when you try to take the best of two worlds and fuck it up.

 

what is SMT?

An old ass idea that should have been implemented half a century ago rathen then 14 years ago.

 

 

shared cache is an issue, because AMD used a slower cache. Intels cache operates at much tighter latencies and much higher speeds. If AMD had had faster and more responsive cache, ideally, you shouldnt have noticed much. But faster = more clocks = hotter.
There is also the question of OS optimization. In the beginning it was evident that Windows 7 and earlier OSs had no clue how FX operated, and as a result though it was a quad "core" rather then 4 modules/8 threads.

 

Then there is a HUGE "maybe".... Intel has a set of liberaries specifically tuned and tailored to THEIR products. Whilst intel HAS to optimize a little bit for AMDs sake (after they got busted for cheating in benchmarks) they dont really do much to help AMD. A LOT of software uses parts of those liberarires. Whilst it wont be hurting AMD much, it falls under a saying we got here where i live "Mange bekker små blir en stor å"... which translates to "many small streams form a large river".... I honestly think there is more then ONE issue with bulldozer. Probably 20+ issues. That in varying degree has been fixed (Piledriver -> Steamroller -> Excavator).

If you remember the thread of AMD bitching about some benchmark strongly favoring intel over them, well, said benchmark is using Intels highly optimized liberaries. The same liberaries (although updated) Intel used when they were busted for cheating in benchmarks.

Is it a coincidence? I think not. but why?

Most of the stuff we do daily ISNT very complex. Rather, it is incredibly simple stuff. Computer AI, quicktime event, typing in word, browsing the web and such isnt complex at all, and should be well within the capability of an basic ALU to compute. There is ZERO reason why the ALU would be struggling in these tasks, and AMD knew this.

So what is the culprint?

Latencies... AMDs CMT implementation is VERY slow. Not the ALUs, but the combined latencies of Cache, IMC, Hyper-Transport (although this one aint THAT slow as it maxes out at 51GB/s bandwidth).

 

Looking at how AMD changed their CMT over time, it is obvious that things HAVE improved.

 

 

In all honesty, just going from Bulldozer to Steamroller, we can see clear changes. The splitting of decode and dispatch has probably helped usher along a much greater efficiency and cut down latency from queueing issues. Still, they got a way to go.

 

It also doesnt help that pretty much every modern instruction is made by Intel Corp.... With the exception of AMD64 (x86-64).... so yeah. Conspiracies or not, aslong as a instruction CAN run on AMD hardware, no matter how unoptimized, it is "ok" in the eyes of the government to not hammer intel for anti competitiveness.
So if it so happens that AMD has any sort of efficiency overhead using Intel instructions (which we will never know), then that doesnt help AMD.

 

In the end, ill say that whatever plagues Bulldozer, it probably isnt entirely AMDs fault, although 75%+ of the blame is definetively theirs to take. BUT, there IS outside factors not helping their case for sure.

 

 

Just for the hell of it. here is K10 aka Phenom II

 

 

 

This is copies from a private discussion i had with someone. This is why it may seem a bit weird in the way it is laid out

also was wondering how you'd managed to make such a post in a matter of minutes haha

[Prysin]

1 minute ago, Citadelen said:

It isn't a suite, it's a co-processor.

that runs off its own API and SDK.

 

aka, a "suite to create 3D sound"

[Citadelen]

1 minute ago, Prysin said:

that runs off its own API and SDK.

 

aka, a "suite to create 3D sound"

So it's still a co-processor, it just has a suite to develop stuff on it.

[Citadelen]

2 minutes ago, Flowey said:

also was wondering how you'd managed to make such a post in a matter of minutes haha

 

pls anybody don't quote this post again

Please snip that post.

[DevilishBooster]

11 minutes ago, Citadelen said:

Wrong statement is wrong. Ripping out compute doesn't count as efficiency.

AMD Bulldozer micro-architecture: 32nm released in late 2011

Intel Skylake micro-architecture: 14nm released in 2015

 

Yeah, I'm wrong... /s

[Flowey]

7 hours ago, TidaLWaveZ said:

 

Always the best decision in my opinion(assuming the software/drivers are up to par).

 

I just thought it was strange that nobody mentioned anything about the difference in release dates between AMD and Nvidia's cards whether it be in AMD or Nvidia's favor.

 

 

A lot of people cannot bear to monitor pricing for like minimum three months before buying something like a GPU. I've got a friend who was buying his GPU about 2 months ago. I had been monitoring and researching but even when I tipped him off that the R9 390 probably performing better in VR then the 970, he went for a 970. Now he can't keep from lamenting about how I got a better card for like 50 buck less. Thatl teach him.

[Citadelen]

1 minute ago, DevilishBooster said:

AMD Bulldozer micro-architecture: 32nm released in late 2011

Intel Skylake micro-architecture: 14nm released in 2015

 

Yeah, I'm wrong... /s

The majority of the OP's post was about GPU's, I thought you were talking about GCN and Maxwell. Also, Bulldozer isn't AMD's latest architecture, it's Excavator on 28nm released in 2015.

[Flowey]

7 hours ago, Citadelen said:

The majority of the OP's post was about GPU's, I thought you were talking about GCN and Maxwell. Also, Bulldozer isn't AMD's latest architecture, it's Excavator on 28nm released in 2015.

Well i knew that AMD's CPU were shite, do explanations are always nice. I believe excavator are also shite, am I right?

[Prysin]

3 minutes ago, Citadelen said:

So it's still a co-processor, it just has a suite to develop stuff on it.

a co-processor is NOT a feature, it is just a co-processor.

However the True Audio API and SDK exposes said co-processor to a developer.

 

that co-processor wouldnt be used for graphics or even ingame sound if the true audio API and SDK didnt exist.

 

that co-processor is an ASIC, it is specialized in tracing sound reverb.

 

what is ASIC?

Application Specific Intergrated Circuit.

 

Meaning the co-processor is BUILT to run within the True Audio API and SDK. WIthout said software is is useless.

 

True audio is not a co-processor. It is a package. The sound COULD be run off the GPU SPs if the software was tuned, but this would come at the cost of FPS. Thus a ASIC Co-processor is added onto the card/GPU silicone to deal specifically with just sound and its own API and SDK. By using its own API and SDK, outside the GPU drivers, it also helps reduce GPU driver overhead, that would otherwise be added if the two weren't separated.

 

 

you are not wrong. But you are not correct either.

[Enderman]

their GPUs are not behind

they are just inefficient compared to nvidia

 

their CPUS are what is behind

their "gaming" CPU FX series was released 7 years ago

[don_svetlio]

2 minutes ago, Enderman said:

their GPUs are not behind

they are just inefficient compared to nvidia

 

their CPUS are what is behind

their "gaming" CPU FX series was released 7 years ago

4*
Bulldozer came 4 years ago
Vishera came 3 years ago
Phenom II came 7 years ago.

[branden_lucero]

57 minutes ago, Gikero said:

I see a lot of people who express how poorly they consider AMD's FX8XXX series, but I've never seen anyone compare to a Pentium 4. Damn.

i still have a Pentium 4 by itself, clocked at 3.0 GHz. i kept it from work.  few bent pins, but i'm sure i can fix that. we had a couple of 3.2GHz ones and i think there was even one of the rarer 3.4GHz (775 socket), which i wish i had kept as well.

[Enderman]

1 minute ago, don_svetlio said:

4*
Bulldozer came 4 years ago
Vishera came 3 years ago
Phenom II came 7 years ago.

well its 5 now since its 2016