Clarkdale proves that HT is NOT a gimmick

[cbn]

I totally agree the benchmark SUCKS but it's..

1. Free
2. Widely Used
3. Not biased towards anyone

I'm not a Fanboi. Have owned both Amd and Intel. Went with an I3 over a Athlon II X4 630 because i like playing with new tech and wanted to see how high it can go on air.

Posting the benchmark proves that Intel's new 32nm Dual core's can trade blow's with Amd's Flagship Quad Core. Nothing more. Nothing less.

Hey Zeus,

Do you happen to own any of the new quad optimized games? Or maybe there is a place where you could download a free benchmarking tool?

It would be especially interesting to see how Core i3 does in Dragon Age with HT toggled on and off for two runs.

 

[aigomorla]

I totally agree the benchmark SUCKS but it's..

try cinebench.

thats what i use..

http://www.maxon.net/index.php?id=162&L=0

 

[JFAMD]

Definitely!! Its true in server platforms, which is why the quad-core Nehalem-based Xeon 5570 can beat quad-core Shanghai by similar amount shown on the benchmark above. The 4 core 5570 is still faster than the 6-core Istanbul by 20-30%, with some database applications reaching nearly 50%!

Of course we are not server guys, so those benches are "irrelevant". But certainly not unrealistic.

What Clarkdale is really showing us is that the number of threads supported by the program is more important than the implementation of Hyperthreading in regards to performance(specifically relating to performance losses here).

Intel really did improve Hyperthreading on the Nehalem compared to the P4. Too bad it showed mostly on the server apps. Home user apps are very picky.

But, to be fair, there are also benchmarks on Anandtech that show HT actually giving negative performance (i.e. better performance with HT turned off).

Now, to be fair, if you are saying that a $1386 x5570 is 20% faster than a $989 Opteron 2435, I would submit that being 20-30% faster and 40% more expensive makes the Intel processor a bad deal. Plus, if draws a lot more power than the AMD processor, so your price/performance/watt is significantly better for the AMD Opteron.

In the enthusiast desktop world, people may actually live by a "raw performance only" decision criteria (though I tend to doubt it, who spends $1K on those high end desktop parts?) However, in the server world, with tight budgets and plenty of power concerns, raw performance is the driver in probably <5% of the server purchases. Server customer buy down the stack, either to get better price performance, better performance per watt, or both. And in that case, I believer our processors are well positioned to win.

 

[shaolin95]

Re-read how they test their coolers. Frostytech has the best testing methodologies for heatsinks bar none.

Sorry but their testing is crap.
You want the 3 top air coolers based on decent websites AND user feedback (you know, zero hidden agendas) then I will give them to you:

Noctua NH-D14
Venomous X
Megahalems

Having a zalman above any of those is not only ridiculous but proves my point about hidden agendas...
End of the story.

 

[IntelUser2000]

It's up to you guys to decide whether I brought up Xeon 5570 vs AMD Opteron 24xx to purely prove it beats it or that I was trying to stay on topic.

It's true value does matter, I'm not going against that.

It's hard to say whether PC users buy more of the enthusiast processors or server users buying top-end CPUs. It's true in both cases that the mainstream CPUs have the largest volume and probably make most money for the CPU manufacturer anyway.

I do believe the point about Hyperthreading still stands though. When an app only recognizes 4 threads yet the CPU can handle 8, and the program is poorly handling the threads, the contention of resources causes performance loss. If similar architecture CPU with Hyperthreading is dual core running the same program though, it won't have the problem, and will benefit from Hyperthreading.

Even completely independent quad cores had a bit of performance loss compared to dual cores: http://www.anandtech.com/showdoc.aspx?i=2866&p=10

 

[JFAMD]

My point was based on 2 assessments, one was by Johan that called out Linpack performance actually being better if you turned off HT (less thread contention) as well as another that I can't find right now but I am sure I could probably track down the link.

I would hope that people would be writing apps to adequately handle HT, though that is probably not always the case. Where I have had the biggest gripe is how it is marketed. When you market a processor as 4 cores/8 threads, there is an implicit position that says you are really getting double. But with a 20&#37; bump in overall throughput, your 8 threads really act like 4.8 cores. The perceptions from the customers that I talk to is that HT is doubling performance (pitched as "twice as many threads") when in reality this is not the case. And in some cases, it is actually negative.

 

[MrK6]

Sorry but their testing is crap.
You want the 3 top air coolers based on decent websites AND user feedback (you know, zero hidden agendas) then I will give them to you:

Noctua NH-D14
Venomous X
Megahalems

Having a zalman above any of those is not only ridiculous but proves my point about hidden agendas...
End of the story.

Before you get yourself in a bunch, look at the Intel results. Your hated Zalman (speaking of hidden agendas, amazing) isn't even in the top 10. You aren't going to find a better way to consistently test the function of a heatsink than to put it on a rated hot plate. Welcome to the scientific process.

Anyway, sorry to go off topic.

 

[Ben90]

Originally Posted by Hey Zeus
I totally agree the benchmark SUCKS but it's..

1. Free
2. Widely Used
3. Not biased towards anyone

I'm not a Fanboi. Have owned both Amd and Intel. Went with an I3 over a Athlon II X4 630 because i like playing with new tech and wanted to see how high it can go on air.

Posting the benchmark proves that Intel's new 32nm Dual core's can trade blow's with Amd's Flagship Quad Core. Nothing more. Nothing less.

wait what?

Originally Posted by Hey Zeus
IF that were the case don't you think i would have an ES of a 32nm Quad in my room?

I'm just an Intel Fanboi. Plain and simple

Im an intel fanboi.
Im an amd fanboi.

Bet you can't process that.... I hate people that can't detect sarcasm

 

[DrMrLordX]

But them if you look at their top 5, the megahalem is number one by temperature !!!

Actually it's only #1 for Intel platforms on their Top 5/10 page. The Iceage Boss II is #1 for AMD platforms.

For some reason the Megahalems is not listed in any of Frostytech's published reviews, yet somehow is #1 on their Intel platform Top 10 list and #5 on the AMD Top 10 list.

Before you get yourself in a bunch, look at the Intel results. Your hated Zalman (speaking of hidden agendas, amazing) isn't even in the top 10. You aren't going to find a better way to consistently test the function of a heatsink than to put it on a rated hot plate. Welcome to the scientific process.

Indeed.

The Zalman HSF seems to perform better with socket AM3 on Frostytech's tests than on Intel platforms. Why? Couldn't tell you. Their testing methodology does a pretty good job of isolating the HSF's performance from variables that exist in "real-world" installations, such as load spikes, fitment/contact problems, warped IHSes, poor/no case airflow, weird eddies/air currents, and so forth.

Some of the variations between relative performance on the two platform groups tested may come down to retention mechanisms. Is Frostytech using the default retention mechanisms, or do they convert everything to bolt-through kits when possible (assuming bolt-through isn't the default)?

In any case, it's up to the builder to make sense of the numbers they produce.

Were I to put together an AMD system tomorrow, I would either go with the Zalman cooler or MAYBE the Venomous X based on the fact that the Thermalright Ultra 120 does pretty well in their AM2/AM2+/AM3 tests and that the Venomous X is the grandchild of the Ultra 120 (in a sense).

The Megahalems is well-loved because of what it can do on Intel platforms. It seems to do pretty well on AMD platforms too, so obviously, one would be well-served by it on an AM3 machine . . . but that doesn't mean it's going to be the best.

And yeah, sorry for going a bit off-topic but the point had to be made.

 

[toyota]

Im an intel fanboi.
Im an amd fanboi.

Bet you can't process that.... I hate people that can't detect sarcasm

so you hate me now? that makes me sad...

 

[Scotteq]

wow..

you guys are calling him a fan boy because he's showing off a 32nm dual core with HT?
Because i think his comment was a meant as a joke...

The definition of Fanboi has changed. It's now an accusation leveled at anyone who has anything positive to say about a company the accusor happens to dislike. Facts, Actual Performance, Reality, and the like have nothing to do with it.

In short - The entire premise is a man made pile of bullsh*t.

 

[MrK6]

The definition of Fanboi has changed. It's now an accusation leveled at anyone who has anything positive to say about a company the accusor happens to dislike. Facts, Actual Performance, Reality, and the like have nothing to do with it.

In short - The entire premise is a man made pile of bullsh*t.

You're such a logic fanboi.

 

[Cogman]

The definition of Fanboi has changed. It's now an accusation leveled at anyone who has anything positive to say about a company the accusor happens to dislike. Facts, Actual Performance, Reality, and the like have nothing to do with it.

In short - The entire premise is a man made pile of bullsh*t.

Nobody accessed him of being a fanboy, he called HIMSELF a fanboy. How does that fit in, self hate?

Im an intel fanboi.
Im an amd fanboi.

Bet you can't process that.... I hate people that can't detect sarcasm

I hate people that don't know what sarcasm is. Here's a hint, it isn't contradiction.

 

[Oxaqata]

Edit

 

[DrMrLordX]

Some "synthetic garbage" is actually useful for gauging performance. CineBench is a pretty good benchmark, in my opinion, and Sciencemark is certainly interesting.

 

[Scotteq]

Yes, stop running this synthetic garbage. >_>

Kinda hard to show HT performance using (for example) a game as a vehicle, since games by and large aren't heavily multithreaded.

 

[Hey Zeus]

DR,

Sciencemark won't load for me in X64 XP

 

[zsdersw]

Now, to be fair, if you are saying that a $1386 x5570 is 20% faster than a $989 Opteron 2435, I would submit that being 20-30% faster and 40% more expensive makes the Intel processor a bad deal. Plus, if draws a lot more power than the AMD processor, so your price/performance/watt is significantly better for the AMD Opteron.

When buying a new server total system cost matters, not CPU price. A server equipped with a Xeon X5570 isn't necessarily going to be 40% more expensive than a server with an Opteron 2435.

Power consumption is about 10-15% in the Opteron's favor.. but then again it's also 20% slower. That's hardly a significant feather in the Opteron's hat. If the Xeon gets the work done faster (before returning to idle) the power consumption advantage becomes less, as well.

 

[JFAMD]

Actually I did a similar comparison back in december:

http://forums.anandtech.com/showthread.php?t=2033369&highlight=2435&page=3

Scroll down the page to entry #61.

You're right that the 40&#37; delta at the processor level doesn't necessarily = 40% delta in total system price. What you'll find is that the Intel based system was 50-60% more expensive.

So, while your statement was factually correct, you were heading in the wrong direction. Total system price is far more important than a component price, and at the system level we deliver MUCH better value.

 

[Enigmatic]

What's with the this obsession with the 965... there are cheaper X4's that OC just as high.

 

[DrMrLordX]

DR,

Sciencemark won't load for me in X64 XP

http://www.techpowerup.com/downloads/170/ScienceMark_2.0_FINAL_32_-_Bit_version_21032005.html

Will that load? Or is that what you tried already?

 

[Hey Zeus]

http://www.techpowerup.com/downloads/170/ScienceMark_2.0_FINAL_32_-_Bit_version_21032005.html

Will that load? Or is that what you tried already?

 

[DrMrLordX]

Glad to see that version worked for you. It would be interesting to see a thread dedicated to Sciencemark 2 scores in XP-64/Vista 64/Win7 64 similar to the Cinebench thread that was started recently.

SM2 is a bit wonky under 64-bit OSes for now and in the foreseeable future, but as long as everyone runs the same version under a 64-bit OS, the results should be useful for comparisons.

 

[Oxaqata]

Edit

 

[JFAMD]

oh...so it is a gimmick?

No, it's real. But there will also be benchmarks that get released showing that HT actually degrades performance and the throughput increase is actually negative.

Now, when that happens, will someone start the thread on that?

HT has its disadvantages just like it has its advantages. That part doesn't get picked up as much.