Saturday, September 08, 2007

Intel's only trick was cache for benchmarketing

A message posted by a reader at ZD-NET. AMD is expected to introduce true quad core on Monday, Intel is expected to follow suit in 2009.

/////////////////////////////QUOTE///////////////////////////////////
You hit the nail on the head
I assume you mistyped "7350" and meant "7320"... If that's the case then, yes, you hit the nail on the head: the only reason the 7345 beats the 7320 is because of its bigger L2 cache.

- Xeon 7320 (2.13 GHz, 4 MB L2): SPECint_rate2006=139
- Xeon 7345 (1.86 GHz, 8 MB L2): SPECint_rate2006=166

The 7345 clock frequency is 13% lower but its score is 19% higher. To better evaluate the impact of the cache size, I went ahead and compared 2 similarly clocked Tigertons:

- Xeon 7330 (2.40 GHz, 6 MB L2): SPECint_rate2006=178 [1]
- Xeon 7340 (2.40 GHz, 8 MB L2): SPECint_rate2006=192 [2]

A jump from 6 MB to 8 MB increases the score by 8%. Here is a breakdown of the scores obtained by each of the 12 benchmarks composing SPECint_rate2006:

1.187 bzip2 - Compression
1.178 hmmer - Search Gene Sequence
1.100 omnetpp - Discrete Event Simulation
1.097 mcf - Combinatorial Optimization
1.094 xalancbmk - XML Processing
1.089 libquantum - Physics / Quantum Computing
1.083 gcc - C Compiler
1.043 astar - Path-finding Algorithms
1.041 h264ref - Video Compression
1.019 perlbench - Programming Language
1.008 gobmk - Artificial Intelligence: Go
1.004 sjeng - Artificial Intelligence: Chess

(For example: "1.187 bzip2" means that the 7340 scores 18.7% better than the 7330 in the bzip2 benchmark: 146/123=1.187 according to [1] and [2].)

The increased cache size benefits mostly 2 of the SPECint_rate2006 benchmarks by almost 20%: bzip2 and hmmer.

Of particular importance is also the negligible impact of the cache size on h264ref. It suggests Tigerton's big L2 cache will almost be of no help when competing against K10 on video encoding tasks. I am sure power users will notice ;-)

Despite its modest L2/L3 cache sizes and given the generally impressive results K10 obtains at equal clock speeds against Core2, I can't wait to see what the 3.0 GHz Phenom monster is capable of.

[1] http://www.spec.org/cpu2006/results/res2007q3/cpu2006-20070821-01882.html
[2] http://www.spec.org/cpu2006/results/res2007q3/cpu2006-20070821-01881.html
///////////////////////////////END///////////////////////////////

6 Comments:

Blogger Scott said...

Yes. Powerusers will notice the platformance of K10 while they megatask.

Perhaps I should run this by AMD before posting, I wouldn't want to upset them. I had to sign an NDA when I first logged onto AMD.com, so I can't say much more.

I already got my K10. I can't say much about that either. The box is actually sealed with an NDA that I had to sign before taking the processor out.

3:35 PM, September 08, 2007  
Blogger Chuckula said...

Well I've had a K10 for the past 3 years and it allows me to communicate with aliens!

Some guy on some site said that Sharikou has an unhealthy relationship with goats! That must be true too!

4:57 PM, September 08, 2007  
Blogger Unknown said...

AMD goats you mean? I hear they have 4 cores !!! :P

8:42 PM, September 08, 2007  
Blogger terry said...

amd barcelona processor will kicks intel penryn processor into the shit

11:51 PM, September 08, 2007  
Blogger Al-Saqer said...

If cache is the only trick, what is wrong with it when it actually works?!

8:44 AM, September 09, 2007  
Blogger netrama said...

Al-Saqer said...

If cache is the only trick, what is wrong with it when it actually works?!


The problem is it makes the C2D look good in benchmarks , than an actual user experience

8:41 AM, September 10, 2007  

Post a Comment

<< Home