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

[Termie]

Tom's has a new set of charts up that some forum members here might find rather interesting: http://www.tomshardware.com/charts/x86-core-performance-comparison/benchmarks,128.html.

It provides a number of benchmarks using a single core, with HT/Turbo disabled, for sixteen CPUs, from the Pentium 4 to the 2600K, running at 3GHz. This is not a comparison of actual performance in benchmarks, as we all know that extra cores, turbo, and hyper-threading definitely make a difference at this point. But it gives a great perspective on how single-core performance has changed since 2005, and helps compare various processor classes on a clock-for-clock basis (i.e. various dual-cores, various quad-cores).

This particular chart gives a nice summary: http://www.tomshardware.com/charts/...parison/All-time-based-values-added,2779.html. Overall, SB is the biggest jump (11&#37😉 in an Intel design since Pentium 4 to Conroe (an amazing 98%). Also, you'll see that AMD's newest CPUs still trail Conroe in single-core performance.

Edited to correct percentage calculations.

 

[LOL_Wut_Axel]

Hmm... kinda expected, really. In other words: Core 2 was the most awesome 'recent' CPU in terms of performance improvements.

 

[OCGuy]

Yep AMD has yet to answer for Conroe or Yorky. Kind of amazing when you think about it.

 

[birthdaymonkey]

Interesting to see the E8600 besting first gen Core i5 and i3 dual cores in many situations... I wouldn't have predicted that. I wonder how the i5 quads would fare?

It's also interesting to be reminded of just how bad (i.e. inefficient, clock speed advantage aside) the P4 really was.

 

[Arachnotronic]

Yep AMD has yet to answer for Conroe or Yorky. Kind of amazing when you think about it.

Phenom II...?

 

[-Slacker-]

I'm confused as to how the athlon IIs perform the same as the phenom IIs in this chart.

Wouldn't that mean they'd have to perform equally in multi threaded apps?

 

[LOL_Wut_Axel]

Phenom II...?

Well, yes. But the Phenom IIs would have to be clocked 200MHz higher to match them in IPC.

 

[LOL_Wut_Axel]

I'm confused as to how the athlon IIs perform the same as the phenom IIs in this chart.

Wouldn't that mean they'd have to perform equally in multi threaded apps?

Err, they don't. The Phenom IIs are around 4% faster.

 

[Termie]

I'm confused as to how the athlon IIs perform the same as the phenom IIs in this chart.

Wouldn't that mean they'd have to perform equally in multi threaded apps?

Well, depending on the benchmark, the Phenom II is actually much faster: http://www.tomshardware.com/charts/x86-core-performance-comparison/3DMark11,2761.html. As Wut Axel points out, overall the difference is only 4%, however. The fact is that the differences between Athlon and Phenom (i.e. L3 cache) only present themselves in certain circumstances (particularly games). It's not an entirely different design.

To AMD's credit, I should note that the original Athlon 64 X2 design was far more efficient than the Pentium 4, but AMD couldn't get the clock speeds up to 3GHz for quite some time. So we don't see the fantastic 3800+ to 4800+ CPUs here, which were the leaders at their time, because they just couldn't hit 3GHz. I'm sure they could beat Pentium 4 on single-core benches even with a big clockspeed deficit, and destroyed Pentium 4 on multi-core apps.

 

[LOL_Wut_Axel]

Well, depending on the benchmark, the Phenom II is actually much faster: http://www.tomshardware.com/charts/x86-core-performance-comparison/3DMark11,2761.html. As Wut Axel points out, overall the difference is only 4%, however. The fact is that the differences between Athlon and Phenom (i.e. L3 cache) only present themselves in certain circumstances (particularly games). It's not an entirely different design.

To AMD's credit, I should note that the original Athlon 64 X2 design was far more efficient than the Pentium 4, but AMD couldn't get the clock speeds up to 3GHz for quite some time. So we don't see the fantastic 3800+ to 4800+ CPUs here, which were the leaders at their time, because they just couldn't hit 3GHz. I'm sure they could beat Pentium 4 on single-core benches even with a big clockspeed deficit, and destroyed Pentium 4 on multi-core apps.

Myself, I'm looking forward to seeing Bulldozer match Nehalem in IPC to be satisfied. If it does, multi-threaded performance will be much better than Sandy Bridge Core i7. If it can't do that,

 

[Kalessian]

Well, yes. But the Phenom IIs would have to be clocked 200MHz higher to match them in IPC.

Instead, couldn't you keep the 3ghz core clock and overclock the CPU-NB to ~2700mhz? I feel like this alone would match wolfdale. I can't really find many useful benchmarks on this.

 

[busydude]

It would be interesting for them to test Llano.. as it about 3% faster than Phenom II. :ninja:

Instead, couldn't you keep the 3ghz core clock and overclock the CPU-NB to ~2700mhz? I feel like this alone would match wolfdale. I can't really find many useful benchmarks on this.

It would beat conroe for sure.. if NB is OC'd.

Interesting to see the E8600 besting first gen Core i5 and i3 dual cores in many situations... I wouldn't have predicted that.

I think the turbo boost helped them massively here.

 

[LOL_Wut_Axel]

Instead, couldn't you keep the 3ghz core clock and overclock the CPU-NB to ~2700mhz? I feel like this alone would match wolfdale. I can't really find many useful benchmarks on this.

Higher CPU-NB affects file compression, encoding, and gaming performance mostly. I guess you could make the argument that the architecture was made to run at 2GHz CPU-NB, hence why it'd be an "unfair comparison".

Also, since some are mentioning Turbo, it was turned off in all CPUs.

 

[-Slacker-]

Err, they don't. The Phenom IIs are around 4% faster.

Ok, then let me rephrase that.

I'm confused as to how the athlon IIs perform almost the same as the phenom IIs in this chart.

Wouldn't that mean they'd have to perform almost equally in multi threaded apps?

 

[LOL_Wut_Axel]

Ok, then let me rephrase that.

I'm confused as to how the athlon IIs perform almost the same as the phenom IIs in this chart.

Wouldn't that mean they'd have to perform almost equally in multi threaded apps?

Yes, your point is...? The Phenom II X4 955 has around a 10% advantage from the get-go against an Athlon II X4 640 and costs $20 more plus it can OC on average to 4GHz, instead of the average 3.7GHz OC of the Athlon II X4 640. If you're looking to get the most out of them, around 15% faster overall because something some people forget is that Deneb scales much better after 3GHz than Propus. Seems like a good deal for only $20 more.

If it's for an average Joe, though, he'll be better served by the Athlon II X4 640 because of lower power consumption, heat, and therefore noise.

 

[Voo]

So all in all Phenom 2s are at best as fast as Wolfdales but more often 10% or more slower. Although Wolfdales hold their head quite nicely even compared to SB - loosing usually less than 20% for two generations and a die shrink is decent.

Hopefully they'll update those benches when BD comes around

 

[Cogman]

Well, yes. But the Phenom IIs would have to be clocked 200MHz higher to match them in IPC.

I don't think you understand what IPC means. IPC = Instructions per clock. It is a factor that is independent of clock frequency.

 

[LOL_Wut_Axel]

I don't think you understand what IPC means. IPC = Instructions per clock. It is a factor that is independent of clock frequency.

Yeah, I know what it means. Got confused on the terminology. Core 2 has slightly higher IPC.

 

[386DX]

Ok, then let me rephrase that.

I'm confused as to how the athlon IIs perform almost the same as the phenom IIs in this chart.

Wouldn't that mean they'd have to perform almost equally in multi threaded apps?

The Athlon II and Phenom II are pretty much the same, the Athlon just lacks the 6MB L3 cache. For single core the 2MB of L2 cache on the Athlon is probably enough to keep the CPU from stalls when only using 1 core. Which is why the performance is similar to that of the Phenom II.

Once you start adding more cores the Athlon II doesn't scale as well as the Phenom II because the lack of cache means the CPU is spending more time getting data from the much slower RAM.

 

[nonameo]

L2 on the athlon II is not shared, its 512 per core.

edit: native dual cores are 1mb per core

 

[sangyup81]

I can't seem to find what RAM was used. I do suspect that DDR3-1600 was used rather than DDR3-1333 or DDR3-1066. The DDR3-1600 would put the Phenom II at a disadvantage if it's running at 2000 MHz CPU-NB whereas Intel's architecture is good to go with higher bandwidth RAM.

Also, what RAM and Chipset would the Conroe and Pentium 4 be running?

 

[Ben90]

Unfortunately you cannot just extrapolate these results to how many cores your program uses. Different architectures scale differently once you start stressing the entire chip.

For example, the Conroe results and Kentsfield by association won't scale nearly as well as Nehalem results once that FSB starts getting overloaded. Nehalem's entire design was largely based around creating a monolithic CPU that keeps performance high when ALL the cores are stressed.

 

[Cerb]

I'm confused as to how the athlon IIs perform the same as the phenom IIs in this chart.

Wouldn't that mean they'd have to perform equally in multi threaded apps?

No. Improved cache, more cache, and improved prefetchers, really shine in multithreaded applications, and multitasking scenarios.

Only a small amount of multithreaded programs can be treated like separate programs running on their own CPUs. Those that do are the ones where the Phenom II still performs OK. Most have dependencies between threads, so what thread A is doing with at this moment can affect what thread B does the next moment, and thread C could be waiting on both of them. In most programs with several threads, cache updates, and reading from far caches, can be major performance issues.

This set of results really shows how diminishing per-core IPC improvements have been, compared to improvements across chips and sockets.

 

[JimmiG]

While somewhat interesting, the results are what you'd expect them to be.

Netburst was a complete failure. Without those pesky laws of physics, it would have been a decent CPU - Intel expected Netburst to hit 10 GHz in 2010, which combined with some slight IPC enhancements would perhaps have put it at around the same performance level as a 3.x GHz i7. This never happened. It's like designing a formula one car, only to discover you're actually going to participate in an off road rally. It just wasn't the right design for the job.

AMD competed favorably only because of those circumstances. When Intel got back on track, AMD was pushed back into the same position they were at with the K6-x, fighting for the low-end and mid-range market. BD will help AMD to regain a hold of this market, but don't expect it to compete with $300+ Intel CPUs...

 

[pantsaregood]

Does no one else find it odd that K8 at 3GHz outperforms K10 at 3GHz according to this chart? Maybe it just happens in some isolated cases, but it is still odd.