Multitasking, physical core vs virtual cores

[Scali]

I backed my stuff like the performance gains with Hyper Threading with links.

And got corrected by various people after that.

The same thing goes with Power consumption, unless if Anand is lying, I just posted the exact things that he wrote in his review.

Yes, you compared with a MUCH faster Intel CPU to show how AMD still consumed more power...

 

[Sp12]

ITT: People with lots of brand loyalty.

Intel has the highest absolute performance, a newer architecture, and a generation advantage in process tech.

Amd has the best price/performance ratio, and the longest platform longevity.



If I'm looking for a strictly gaming platform, I'll take an i5 750. If I'm looking for a strong gaming with significant productivity capability, I'll take an X6. If I'm on a tight budget, I'll take an X4 or an i3.

If money is no issue a Gulftown makes the most sense.

They each cater to target market segments, and claiming superiority for one brand over another is silly, it's not about performance, it's about value.

 

[Scali]

it's about value.

Exactly, and once people understand that value is not necessarily expressed in money, my work here is done.

 

[evolucion8]

Well, why not necessarily can be expressed in money. Tell me, which $100 Intel processor can match the Athlon II X4 630 in terms of performance?

 

[Scali]

Time can be money aswell.
Eg, a Gulftown may seem very expensive, but if you use it for heavy duty processing where it can reduce your compile time, rendering time, or something like that, it will earn itself back for you. It makes you more productive.

Or how about gaming.. what if I like to game with a certain minimum of framerate, resolution, image quality etc... Call it the 'experience', popular marketing term a few years ago... What if that gaming experience is more important to me than the actual cost of the hardware? That any hardware, no matter how cheap, which can not deliver to my minimum expectations, is actually worthless to me? That I'd rather pay a few hundred bucks extra to get a computer that meets, let alone exceeds my expectations? Money isn't that important.

 

[aigomorla]

Well, why not necessarily can be expressed in money. Tell me, which $100 Intel processor can match the Athlon II X4 630 in terms of performance?

Ok.. then tell me when your a company that looks at efficiency and also the cost of maintaining a server room.

Would you want 2x AMD processors in there which cost more energy, also causes air conditioner to be used harder, and takes up more footprint?

Or would you rather replace them with a more efficient, faster, yet cooler running processor which will end up saving you money on the long run?

As i said you need to look at which niche you are aiming at.

BOTH of you guys are blindy looking at once side.

The Intel is a Van... the AMD is a truck.

Sometimes VAN is better then a Truck.. and others Truck is more practical then a VAN.

 

[Scali]

BOTH of you guys are blindy looking at once side.

I'm not, I just don't agree with your view on Intel vs AMD...

The Intel is a Van... the AMD is a truck.

Sometimes VAN is better then a Truck.. and others Truck is more practical then a VAN.

I don't think that analogy is right at this stage.
Intel has AMD beat at everything but price (and even that is stupidly close in some cases, not worth arguing about). No matter what you're looking for... performance, power consumption, size (Intel's 2P server can rival the performance of AMD's significantly larger and more expensive 4P servers... for example), the latest features (special SSE instructions to accelerate video encoding, encryption etc)...
Intel has AMD beat pretty much everywhere... except price. But there's no point arguing about price. Every practical issue, Intel wins.

 

[evolucion8]

BOTH of you guys are blindy looking at once side.

The Intel is a Van... the AMD is a truck.

Sometimes VAN is better then a Truck.. and others Truck is more practical then a VAN.

I'm not, I just don't agree with Scali's point of view on Intel vs AMD...

 

[aigomorla]

Every practical issue, Intel wins.

nope... does not.

I have been nailed enough by AMD people to learn that in some Apps, AMD is better.

Its the niche the product is going into.

 

[Scali]

nope... does not.

I have been nailed enough by AMD people to learn that in some Apps, AMD is better.

Its the niche the product is going into.

Yea right. The performance difference between AMD's fastest and Intel's fastest is so large at this point that there really isn't ANYTHING AMD is significantly better at.

If anyone manages to find the exception to the rule, it's either a rigged test, or the niche is too insignificant to mention.

 

[LoneNinja]

Yea right. The performance difference between AMD's fastest and Intel's fastest is so large at this point that there really isn't ANYTHING AMD is significantly better at.

If anyone manages to find the exception to the rule, it's either a rigged test, or the niche is too insignificant to mention.

Problem is most people excluding you don't compare best product to best product. We're talking $300 vs $1000. I personally don't care for either processor, both out of my price range, but I wouldn't blindly say Intel is better in all situations because their processor that costs 3 times as much performs 40% better or whatever the difference between Thuban and Gulftown actually is.

Intel has the more advanced architecture right now, I won't deny that, but below the I5 750 they have nothing that would even interest me. Price/Performance is important to everyone I know, every one has a budget. Intel could lower prices and take that advantage from AMD if they wanted, but they have clearly choses not to because people still pay the premium for their products.

 

[evolucion8]

Yea right. The performance difference between AMD's fastest and Intel's fastest is so large at this point that there really isn't ANYTHING AMD is significantly better at.

If anyone manages to find the exception to the rule, it's either a rigged test, or the niche is too insignificant to mention.

Denial in its maximum expression. Give it a rest man, in things like Virtualization, stuff that uses FPU, Encryption, AMD platform is quite competitive, specially in multi CPU platforms, HyperTransport scales better than QPI.

 

[Scali]

Problem is most people excluding you don't compare best product to best product.

I think we already covered how money isn't everything...
I pointed out specifically that with everything BUT price, Intel has AMD beat.
But for as long as I can remember (and I go WAY back to the early Am386DX days), the AMD fanboys sound like a broken record, harping on price this, price that.... ignoring everything else.

I wouldn't blindly say Intel is better in all situations because their processor that costs 3 times as much performs 40% better or whatever the difference between Thuban and Gulftown actually is.

It was just one example. I can come up with endless examples, as long as price is left out of the equation. Intel wins on EVERYTHING, as I said. Try to keep up with the CONTEXT please, okay? Thanks.

but below the I5 750 they have nothing that would even interest me.

Who cares anyway? The i5 750 is a piss-cheap CPU already. As I said, you have to be a poor student or a jobless person if you can't afford one. You're not representative of the average consumer.
Heck, way back when, that Am386DX was in a higher pricerange than the i5 750 is today. What are we talking about, really?

but they have clearly choses not to because people still pay the premium for their products.

Yes, period. No point in reiterating that point over and over and over and over and over and over again. So why why WHY WHY WHY does every AMD fanboy have keep bringing it up in EVERY thread AGAIN and AGAIN and AGAIN and AGAIN?
Annoying when the same thing is repeated, isn't it?

 

[Scali]

Denial in its maximum expression. Give it a rest man, in things like Virtualization, stuff that uses FPU, Encryption, AMD platform is quite competitive, specially in multi CPU platforms, HyperTransport scales better than QPI.

Oh god... don't tell me you don't even know that Intel has special instructions for encryption that AMD doesn't have...
Luckily Anand covered it: http://www.anandtech.com/show/2901/5
"...or at least AES-NI will probably keep AMD out of the running for consideration."
Or how about a synthetic test to show the REAL power of those AES instructions?
http://www.anandtech.com/Show/Index...gny-cours-opteron-6174-vs-intel-s-6-core-xeon
AMD is NOWHERE in that test.
"Once the Xeon X5670's AES instructions can do their work, encryption is lightening fast. Here the new Xeon is 19 times faster than its older brother and 9 times faster than the best Opteron."

As for the other stuff: yea right, dream on. I recall a blog post of... I believe it was Ryan Smith, talking about potential performance problems with virtualization on Nehalem because of a disadvantage in the TLB caching on paper. But once they benchmarked actual hardware, Nehalem showed no weaknesses.
FPU? Yea right... Not that anyone cares anyway, that's what GPUs are for.
HyperTransport scales better than QPI? In what alternative reality is that?

Just more of the same rubbish as you had been spouting about HyperThreading earlier.

 

[evolucion8]

Just more of the same rubbish as you had been spouting about HyperThreading earlier.

For the third time, I proved you wrong with the Hyper Threading, with links and stuff and you just come out dodging the results, definitively you need help, get a life.

Before those instructions, AMD used to be the king of encryption, time changes buddy.

Look at this bandwidth comparison table;

HyperTransport (800 MHz, 16-pair) 25,600 Mbit/s 3,200 MB/s

HyperTransport (1 GHz, 16-pair) 32,000 Mbit/s 4,000 MB/s

PCI Express 1.0 (x16 link)[28] 32,000 Mbit/s 4,000 MB/s

PCI Express 2.0 (x8 link)[29] 32,000 Mbit/s 4,000 MB/s

PCI-X QDR 34,133 Mbit/s 4,266 MB/s

AGP 8x 64-bit 34,133 Mbit/s 4,266 MB/s

PCI Express (x32 link)[28] 64,000 Mbit/s 8,000 MB/s

PCI Express 2.0 (x16 link)[29] 64,000 Mbit/s 8,000 MB/s

QuickPath Interconnect (2.4 GHz) 76,800 Mbit/s 9,600 MB/s

QuickPath Interconnect (3.2 GHz) 102,400 Mbit/s 12,800 MB/s

PCI Express 2.0 (x32 link)[29] 128,000 Mbit/s 16,000 MB/s

HyperTransport (2.8 GHz, 32-pair) 179,200 Mbit/s 22,400 MB/s

HyperTransport 3.1 (3.2 GHz, 32-pair) 409,600 Mbit/s 51,200 MB/s

Now which is faster? AMD processor might not be able to take full advantage of it, but large scale projects of multi CPU's, HyperTransport is the way to go.

http://hardforum.com/showthread.php?t=1409990

http://www.overclockers.com/forums/showthread.php?t=603778

Kuroimaho: WRONG. Things DO use HT... HT, unlike QPI, supports a physical slot. Devices currently available for use in HTX(3) slots include FPGA accelerators and dual-port 10Gb routers.

1. QPI at 3.2GHz = 12.8GB/s
2. HT at 3.2GHz = 12.8GB/s (32-bit does 25.6GB/s, but AMD CPUs only use 16-bit)

There's a thread of comparing both showing that QPI has lower latency, and that HyperTransport is bottlenecked by AMD own memory controller and links.

 

[Scali]

For the third time, I proved you wrong with the Hyper Threading, with links and stuff and you just come out dodging the results, definitively you need help, get a life.

Various people have corrected you after posting those links. You're the one in denial.

Look at this bandwidth comparison table;

I don't know where you got it from, but it's wrong.
QPI is a factor 2 too low:
http://en.wikipedia.org/wiki/Intel_QuickPath_Interconnect
And HT 3.1 is a factor 2 too high (32-bit is supported by the HT3.1 protocol, but all current CPUs implement only 16-bit links):
http://en.wikipedia.org/wiki/HyperTransport

After these corrections, QPI has the highest bandwidth.

Besides, since when do theoretical maximum bandwidth numbers represent scalability?
You are completely skipping the effect of the topology and the bus protocol.

 

[evolucion8]

Various people have corrected you after posting those links. You're the one in denial.

Reference: http://www.anandtech.com/show/2832

I've never stated that you were totally wrong, is just that you pretend that Hyper Threading is the holy gray of Intel's performance and that isn't true. A similar clocked i7 920 is slightly faster than a similar clocked i5 750 which is identical in everything except the lack of Hyper Threading. The peformance gains are good considering the little logic that it takes, but that's it, they're not impressive.

 

[Scali]

And again he cherry-picks ONE benchmark, when all this time the others have been arguing that the value of HT depends on the usage scenario.
You're just not getting through to this guy. Waste of time.

"I've never stated that you were totally wrong, is just that you pretend that Hyper Threading is the holy gray of Intel's performance and that isn't true."

I haven't said anything even close.
You are the one claiming that HT has virtually no use at all... I merely pointed out that in certain scenarios (mostly server-oriented, so most Anandtech reviews are pretty useless for this anyway), the amount of cores can be more important than whether they are virtual or physical.
I also pointed out that both Sun and IBM are doing this aswell (in fact, IBM was the first, not Intel). So it's not like it's an Intel thing.
But again I'm just repeating myself over and over again. It's easier to just insult you. You don't deserve better, really.
You've proven over and over again that you don't even respond to arguments, you don't bother to understand what the other is talking about (case in point: Sun Niagara.. another case in point: Intel's encryption extensions), and you just post a lot of incorrect information (such as the bandwidth table to 'prove' that QPI is bad).
I don't know where the moderators are, but they should have done something about you a long time ago.

 

[evolucion8]

Whatever, I posted the most demanding tests in that benchmark, didn't bother to post games since most are GPU limited, neither Excel or Archiving performance which is bandwidth limited and Sysmark is just a benchmark, I believe in real usage scenario. You are in total denial, so you are just a waste of time and air, believe whatever you need to believe. Your Anti AMD antics will only work on fanboys, customers will buy whatever they need without caring if its Intel or AMD.

 

[Scali]

Anti-AMD? You wish.
I'm just pro-SMT, just like Intel, IBM and Sun. I'd love to see AMD add SMT to their CPU line and become more competitive.
But have you seen SMT on their roadmap? I haven't.

 

[iCyborg]

I have an i7 920 and I keep HT off, that's how much I value it

It provides 20-40% gain in well-threaded memory-bound apps like media processing, and negligible gain and even loss in pretty much everything else. Since I don't stare at the screen waiting for encoding to finish (and I do media encoding stuff perhaps once a month...) nor is my PC a VM host for multiple users, I see no point in using it. And it may be only a small die increase, but I'm not sure about power savings, running a 100% load test that has benefits from HT pushes the load temps from low-mid 60s into 70s on my OC-ed CPU, that's a quite a noticeable difference in thermals.

Cliffs: I really don't see lack of SMT as something that's holding AMD down and completely understand if it's not a priority for them...

 

[Voo]

We need better programers who write machine code direct instead of use a set of open source codes to make a new program.

Yeah sure, ever written any larger program? (also not sure how "set of open soure code" has anything to do with the language you use.. there are asm libraries [and open source] as well)

I argue that nobody has ever written a larger multithreaded program in asm, because it's just completely crazy and compilers actually doing a pretty decent job these days with optimizations, though if you know the machine the program will be running you can optimize stuff lots afterwards (register allocation, stuff like optimizing FMA use, etc.) - but in the consumer space? Do you really want to write different programs for those CPUs that use SSE2, 3, etc.?

Assembly has its uses at the bottom level to get all those tools/libs/VMs needed to program efficiently and in those situations where you want to optimize some code and don't care about portability at all, which is usually only interesting for the HPC crowd but not standard consumer programs, why bother about a 20% sequential performance increase with all the associated problems if you can make the program scale from 2 to 4/6 cores and profit everywhere?

PS: And "Money is only important to those who don't have it. Poor students, jobless people etc. Not the average consumer or business."? Ah that's the reason why most people drive ferraris and nobody cares about price?
And the average business usually very much cares about the most efficient buy, though there are more variables then just performance/price that also play into that equation..

 

[Scali]

I have an i7 920 and I keep HT off, that's how much I value it

Bingo!
HT is an option. You can turn it off if it isn't for you. And you can turn it on if it is.

Cliffs: I really don't see lack of SMT as something that's holding AMD down and completely understand if it's not a priority for them...

I don't understand that conclusion when you've just said that SMT can give 20-40% gain in certain situations. That's a LOT of gain, and not something you can easily compensate with more efficient cores or higher clockspeed. Especially not in AMD's case, since they are already behind Intel in core efficiency, and don't have a clockspeed advantage either.

 

[Scali]

PS: And "Money is only important to those who don't have it. Poor students, jobless people etc. Not the average consumer or business."? Ah that's the reason why most people drive ferraris and nobody cares about price?

There we have the one-dimensional thinkers again.
So you're saying that there are only ultra-cheap cars and Ferrari's? No middle ground? I hope you realize how stupid that sounds.

And the average business usually very much cares about the most efficient buy, though there are more variables then just performance/price that also play into that equation..

My point exactly. AMD wins performance/price, but not necessarily the other variables, which may be more important, as I have tried to explain, but I don't seem to be able to get through to people.

 

[iCyborg]

I don't understand that conclusion when you've just said that SMT can give 20-40% gain in certain situations. That's a LOT of gain, and not something you can easily compensate with more efficient cores or higher clockspeed. Especially not in AMD's case, since they are already behind Intel in core efficiency, and don't have a clockspeed advantage either.

It is a 20-40% gain in specific scenarios that might not be important to most people. I am not so sure that the 5% die increase couldn't be used in a better way for a more general performance increase, e.g. I'd easily take a 3% increase in everything over 30% in encoding. Maybe it really can't and HT is well worth, but I've not been convinced so far, and for sure I don't see it as a game changer.

And there was some discussion here with people arguing that large HT gains might be due to apps not being well optimized. I can see how SMT can be used to e.g. hide memory latencies, but scheduling more CPU-bound threads than physical cores on a SMT enabled CPU will probably do more damage than good, even if you will see a gain compared to running the same number of threads with SMT off.