[Tom's] Normalized single-core CPU performance at 3GHz

[Voo]

On top of it, is this test a true measure of single core performance when there are other factors like L2 and L3 cache? Should those be crippled down to the lowest common denominator? Would that still be an accurate measure of single core performance?

As can be seen perfectly fine from the Athlon vs. Phenom example most of those CPUs aren't hampered by too few cache. Now a smaller, faster cache may get you better results while for multi threaded programs this may be different, but for most programs the difference will be small enough - also not much one can do about that.

Also they list the times for every single program so I don't see your problem - sure adding up times together has some obvious flaws but it's not as if you didn't have the raw numbers to normalize them and make a more useful summary - but then it's not as if there were any surprising outlier that would seriously distort the data.

And why IPC is important? Well because apart from some classes of problems (encoding, raytracing,.. - basically anything for which CUDA programs exist by now ) many, many algorithms don't scale well to more than maybe a dozen or two threads. Not to forget that even for perfectly parallel problems, there are many real world reasons why you still need a good baseline performance and not hundreds of extremely weak cores. There are enough papers from google and MS on that topic. Also neither AMD nor Intel have any idea how to scale beyond a few dozen cores at best with their current architectures, which will be a much more interesting problem than who has two cores more or less right now (the advance of NUMA for PCs? that'll give developers some headaches)

Also if you look at the usual game you'll see that even for only four cores the work isn't equally distributed and that'll only get worse with more and more cores.

I'm confused, how do the i5-2500k and the i7-2600k perform differently if these benchmarks are supposed to show 1 thread, at the exact same frequency?

They do? In all tests I looked at the difference is minimal and easily below the margin of error.

 

[toyota]

Single-threaded performance is still king both with regard to gaming and in many enterprise applications, so comparing it across different processor generations isn't exactly pointless.

Games have been made to utilize more than one core (for example to make physics calculations), but the main game thread is still a single-threaded application. There is little that can be done to change that fact because of the nature of what a game is, where everything that happens and which has to be computed and rendered is dependent on what the player decides to do next. Therefore it's not a problem that can be effectively parallellized (whereas the rendering of each individual frame can).

that makes no sense at all. the fact is games use more than one core and many can completely max two cores and beyond. I dont care how much IPC you have, a single core of any cpu is not enough to run most modern games properly.

 

[LOL_Wut_Axel]

Three CPU heavy threads?

Yes. Some take advantage of four, but most of three. Go down to two and you lose a considerable amount of FPS. Want easy confirmation? Look at the Athlon II X2 vs the Athlon II X3.

Sure, wise guy. But in the software field, IPC does mean InterProcess Communication.

Search MSDN sometime.

Except we're discussing hardware, not software, "wise guy".

 

[Voo]

that makes no sense at all. the fact is games use more than one core and many can completely max two cores and beyond. I dont care how much IPC you have, a single core of any cpu is not enough to run most modern games properly.

Sure you may get games to utilize two cores completely - maybe even four (although so far I haven't seen one) - but eight, twelve? It's just getting harder and harder.

Also every core profit from a higher IPC and contrary to more cores there's no diminishing return for it or it can only be used for certain programs. It's not as if Intel couldn't release a 12core CPU if they really wanted - adding cores is a relatively simple problem (up to a certain number of cores) compared to increasing the IPC.

And right now we're still having extremely tame core counts and not much problems with cache coherency overhead, etc. But as soon as the number of cores really starts to get up, both camps will have to come up with some really good ideas - and so far nobody has solved this problem satisfactorily (programming a game to use 60 NUMA cores completely? That'll be an extremely hard problem)

 

[toyota]

Sure you may get games to utilize two cores completely - maybe even four (although so far I haven't seen one) - but eight, twelve? It's just getting harder and harder.

Also every core profit from a higher IPC and contrary to more cores there's no diminishing return for it or it can only be used for certain programs. It's not as if Intel couldn't release a 12core CPU if they really wanted - adding cores is a relatively simple problem (up to a certain number of cores) compared to increasing the IPC.

And right now we're still having extremely tame core counts and not much problems with cache coherency overhead, etc. But as soon as the number of cores really starts to get up, both camps will have to come up with some really good ideas - and so far nobody has solved this problem satisfactorily (programming a game to use 60 NUMA cores completely? That'll be an extremely hard problem)

um okay but that has little to do with my reply to him. most games cannot get by on one core any longer and that's all I was saying.

 

[VirtualLarry]

Except we're discussing hardware, not software, "wise guy".

We are discussing benchmark performance, and someone brought up multitasking. IPC (interprocess communication), is at the very heart of multi-tasking software design. How a processor handles the various forms of IPC, and it's performance at such, is key to understanding the performance of the CPU in multi-tasking scenarios. Amhdal's law is key here.

(where's IDC when you need him.)

 

[Munky]

Other than for academic curiosity, clock-for-clock performance is a non-issue even more so than single-core performance. The P4 was designed to operate at higher clocks than a C2D, even using a larger lithography process. How about we compare a 90nm P4 at 3.6GHz to whatever clocks a 90nm C2D could theoretically muster?

 

[LOL_Wut_Axel]

We are discussing benchmark performance, and someone brought up multitasking. IPC (interprocess communication), is at the very heart of multi-tasking software design. How a processor handles the various forms of IPC, and it's performance at such, is key to understanding the performance of the CPU in multi-tasking scenarios. Amhdal's law is key here.

(where's IDC when you need him.)

Perhaps you should learn to make your terminology clear, then. Anyone else would've thought you confused the normal definition of IPC (Instructions Per Cycle) for something else.

 

[Phynaz]

Yes. Some take advantage of four, but most of three. Go down to two and you lose a considerable amount of FPS. Want easy confirmation? Look at the Athlon II X2 vs the Athlon II X3.



Except we're discussing hardware, not software, "wise guy".

Any game made since 2009 makes heavy use of three threads?

Citation please.

 

[LOL_Wut_Axel]

Any game made since 2009 makes heavy use of three threads?

Citation please.

It's not literal. Stop trying to argue over stupid crap. Most games from 2009 onwards take advantage of three threads. It's just stupid seeing you on every thread with the same BS. I'm not gonna waste my time looking for a long list. If you want to, look at most of the popular games from 2009 onwards and you'll see they perform better on more than two threads.

 

[Voo]

Perhaps you should learn to make your terminology clear, then. Anyone else would've thought you confused the normal definition of IPC (Instructions Per Cycle) for something else.

Wait you think that people would interpret "IPC" in a paragraph about the fact that reviewers should add multi-threaded benchmarks in the same sentence with cache coherency and co as instructions per cycle?
Uh, no I think you're projecting there, sorry.

Also there's a BIG difference between taking advantage of three threads and making "heavy use" of three threads, so to support your claim you'd have to show some games that balanced the three threads about equally - which will be a good bit harder to find..

 

[LOL_Wut_Axel]

Any game made since 2009 makes heavy use of three threads?

Citation please.

Just because you'll inevitably say it's false and all you need is two threads, games include F1 2010, DIRT 2 and 3, Mafia 2, Metro 2033, Civilization V, Resident Evil 5, and there's a lot of others too.

There, you have your list. All those games show marked improvements with an Athlon II X3 over an Athlon II X2.

Include Grand Theft Auto IV, Supreme Commander, Race Driver GRID, Battlefield: Bad Company 2, Dragon Age: Origins. There's some more.

 

[toyota]

Just because you'll inevitably say it's false and all you need is two threads, games include F1 2010, DIRT 2 and 3, Mafia 2, Metro 2033, Civilization V, Resident Evil 5, and there's a lot of others too.

There, you have your list. All those games show marked improvements with an Athlon II X3 over an Athlon II X2.

Include Grand Theft Auto IV, Supreme Commander, Race Driver GRID, Battlefield: Bad Company 2, Dragon Age: Origins. There's some more.

Anno 1404, Dragon Age Origins, Dragon Age 2, Ghostbusters, Prototype, Red Faction Guerrilla, Red Faction Armageddon, Witcher 2, Lost Planet 2 and Crysis 2 would be a few more to add to your list.

 

[pantsaregood]

One interesting point to take away from this is that K8 was far ahead of its time. Athlon 64 and Athlon 64 X2, minus the addition of SSE3, barely changed from 2003 to 2007. The K8s actually stand up quite well to K10s, and by extension, Conroe.

Maybe Conroe wasn't as huge of a jump forward as once thought?

 

[LOL_Wut_Axel]

Anno 1404, Dragon Age Origins, Dragon Age 2, Ghostbusters, Prototype, Red Faction Guerrilla, Red Faction Armageddon, Witcher 2, Lost Planet 2 and Crysis 2 would be a few more to add to your list.

Hopefully he'll realize most everything, including games, has decent multi-threading now.

 

[Termie]

One interesting point to take away from this is that K8 was far ahead of its time. Athlon 64 and Athlon 64 X2, minus the addition of SSE3, barely changed from 2003 to 2007. The K8s actually stand up quite well to K10s, and by extension, Conroe.

Maybe Conroe wasn't as huge of a jump forward as once thought?

Absolutely. The X2 was dominant for quite some time. The problem was not so much the effectiveness of the architecture, it was the inability to get clocks up. The FX-62 @2.8GHz was just about equal to an e6600 @2.4GHz, so indeed, an old design was actually keeping up with Conroe pretty well. Link: http://techreport.com/articles.x/10351/1

The problem was that just about anyone could take an e6600 up to 3GHz and beyond, and therefore take performance up at least 25%, whereas the X2 was tapped out by the time Conroe came along. Conroe's combination of overclocking headroom and lower power consumption is what made it superior to X2. It wasn't a radical jump in performance at stock frequencies.

 

[Cerb]

Can someone explain to me why this measurement is relevant? And I'm not talking just about 'single cpu performance', I'm talking about the methodology used (adding up times from a bunch of tests?) and why this is a valid way to test CPU performance.

It is not relevant to any purchasing decisions any human should make.

When people talk about how we've hit diminishing returns in per-core IPC, this just shows that we indeed have, and how much designing CPUs with more cores working together in mind affects performance, today.

 

[classy]

Is this list even relevant?

 

[KamiXkaze]

Wow very interesting to see.

kXk

 

[Phynaz]

I'm not gonna waste my time looking for a long list.

And yet you did :thumbsup:

 

[cusideabelincoln]

I don't get how Phenom II and Athlon II do so badly in this test:

http://www.tomshardware.com/charts/x86-core-performance-comparison/3DMark11,2762.html

Relative to the Athlon X2s and P4. No other test shows that discrepancy.

 

[hamunaptra]

I don't get how Phenom II and Athlon II do so badly in this test:

http://www.tomshardware.com/charts/x86-core-performance-comparison/3DMark11,2762.html

Relative to the Athlon X2s and P4. No other test shows that discrepancy.

Cuz the whole damn article is a bunch of BS! .. ugh .. I hate review sites more n more.

 

[sangyup81]

I don't get how Phenom II and Athlon II do so badly in this test:

http://www.tomshardware.com/charts/x86-core-performance-comparison/3DMark11,2762.html

Relative to the Athlon X2s and P4. No other test shows that discrepancy.

Use the DDR3 RAM that K10's do poorly with relative to Intels

 

[Jawadali]

Personally, it'd be interesting to see how an Athlon XP (K7) core would compare to these, as that was the first CPU that got me into computer DIY.

I think they topped out at 2.2Ghz stock though, and required a sophisticated match of cooling and hardware to reach speeds above 2.5-2.6Ghz (even using mobile variants).

 

[Munky]

I used to have a K7 Mobile at 2.45GHz using average air cooling. According to old school cpumark, it scored about the same as a 2GHz K8. In real world performance, it wasn't really a contest at all.