K10 slower than C2Q in Cinebench, clock for clock!

Quote:

---

The AMD benchmark ran on a single-socket, K10 CPU running at 1.6 GHz on NVIDIA's nForce Professional 3400 chipset. According to the system properties, the AMD system used 4GB of DDR2-667.

The most similar Intel system we could muster up on such short notice was an Intel Xeon 3220. The Xeon X3220 is clocked at 2.4 GHz, and ran on Intel's Garlow workstation chipset (Intel X38). This system property profile stated the system utilized 4GB of DDR2-800.

Cinebench completed the default benchmark in 27 seconds for the 1.6 GHz K10; 17 seconds for the Intel Xeon X3220. The Kentsfield Xeon was 58% faster with a 50% higher clock frequency for Cinebench.

---

http://www.dailytech.com/Quick+and+D...rticle7574.htm

i dont really see the value in that comparison.
as you have no clue what kind of a K10 it was.
no idea how much cache it had and ofc no idea how much of a perf impact cache has on K10 etc.

il see the benches once the CPu is launched and when we know all the info.

OMFG AMD IS DOOMED.
(jk.)

rofl ... the title "slower clock for clock" is rubbish...

not exaclty a good comparison.

Quote:

---

Originally Posted by Origin_Unknown

rofl ... the title "slower clock for clock" is rubbish...

not exaclty a good comparison.

---

agree.

Yes this is a bad comparison.
They shoulld have used equally high clocked cpu's and ram.:down:

Pretty useless as it is.

Quote:

---

Originally Posted by Starscream

i dont really see the value in that comparison.

---

Well, it is the second K10 benchmark, but the first from a third parity bencher. Let's hope we'll see more benchmarks soon.

Quote:

---

as you have no clue what kind of a K10 it was.

---

Read the article, it says the latest silicon revision.

Quote:

---

no idea how much cache it had and ofc no idea how much of a perf impact cache has on K10 etc.

---

K10 has 4x128kB L1, 4x512kB, 1x2048kB L3. K10 is a K8 derivate, so IMO the cache capacity can't make a huge difference in performance.

Quote:

---

il see the benches once the CPu is launched and when we know all the info.

---

I agree. But we heard a lot of promises from AMD about K10 performance and the only 2 benchmarks(POV Ray and Cinebench) are indicating the opposite of what AMD claims.

This is something at least...
Cinebench is using SSE2 codepatch so this alone should increase Barcelona performance compared to old K8 but sadly it's barely faster...
From one of DT posters.

Quote:

---

dual core 3GHz K8 = 829
quad core 1.6GHz K10 = 814

---

For me either 1.6GHz is not correct or Barcelona is not faster that much from K8 on heavy FP load...

doesnt anyone have an e5310 we could do a comparison with? this is the first real benchmark that we could use to gauge the performance of k10. now only if we could see some game benchmarks....

Quote:

---

Originally Posted by triple_A

Yes this is a bad comparison.
They shoulld have used equally high clocked cpu's and ram.:down:

---

The highest clocked RAM that Barcelona will support is DDR2-667.
At lower frequency the CPU has more bandwidth per clock, so it is an advantage for K10 in this case.
Anyway, we still can compare it to 1.6GHz C2Q. We just need someone with Kentsfield to run it underclocked to 1.6GHz(with a little disadvantage of the underclocked FSB to 800Mhz), using DDR2-667.

Quote:

---

Originally Posted by gOJDO

K10 has 4x128kB L1, 4x512kB, 1x2048kB L3. K10 is a K8 derivate, so IMO the cache capacity can't make a huge difference in performance.

---

do u know 100% shure that cache doesnt have a huge impact on k10.
and do u know for shure that this 1.6ghz version has 4x128kB L1, 4x512kB, 1x2048kB L3

Quote:

---

Originally Posted by Starscream

do u know 100% shure that cache doesnt have a huge impact on k10.
and do u know for shure that this 1.6ghz version has 4x128kB L1, 4x512kB, 1x2048kB L3

---

We've known with previous versions of K8 that cache doesn't have a huge impact on performance. Perhaps 2-3% on average between 512KB L2 and 1024KB L2. This is strictly my opinion, but I believe the quad core is going to need good bandwidth and a darn good cache prediction to be effective at keeping all four cores munching on data.

Compare to dual core K8 and you get that core per core this K10 sample would be slower than 4 years old K8 design,which doesn't make sense at all...

Quote:

---

Originally Posted by freeloader

We've known with previous versions of K8 that cache doesn't have a huge impact on performance. Perhaps 2-3% on average between 512KB L2 and 1024KB L2. This is strictly my opinion, but I believe the quad core is going to need good bandwidth and a darn good cache prediction to be effective at keeping all four cores munching on data.

---

i know that this was the case with k8 but u cant be 100% shure if this si the same case with K10 specialy with theaddition of l3.

Quote:

---

Originally Posted by Starscream

do u know 100% shure that cache doesnt have a huge impact on k10.

---

I am not 100% sure, but I think so. It had almost no impact on the K8, because of its ODMC. The K10 will have 2 ODMCs which are going to be more advanced than the one on the K8.

Quote:

---

and do u know for shure that this 1.6ghz version has 4x128kB L1, 4x512kB, 1x2048kB L3

---

All K10 will have 128kB of L1 and 512kB of L2 per core and 2MB of shared L3 cache among all cores.

OK get this....

Until a Barcelona scores higher than C2D in a benchmark nothing is real and nothing is fair!

How difficult is this to understand?

The only real benchmark will be only when the Barcelona win get over it.....

Quote:

---

Originally Posted by Starscream

do u know 100% shure that cache doesnt have a huge impact on k10.
and do u know for shure that this 1.6ghz version has 4x128kB L1, 4x512kB, 1x2048kB L3

---

hey its gODJO he knows it all... and he is very good in calculations and posting preffered when it is bad for AMD and good for INTEL.

its an nfore 3400 btw so that is a socket am2 board !!!! according to all nvidia papers. So how can it be a barcelona cpu? they are demonstrating budapest :shrug:

Quote:

---

Originally Posted by gOJDO

All K10 will have 128kB of L1 and 512kB of L2 per core and 2MB of shared L3 cache among all cores.

---

nope "shangai" is also a K10 on the roadmap and this one will have more cache.

and cache + latency does make a difference performance wise for the k8 check the differences between caches with: 512 on 90nm 1024 on 90nm and the 512 on 65nm

Quote:

---

Originally Posted by Fatal Error

OK get this....

Until a Barcelona scores higher than C2D in a benchmark nothing is real and nothing is fair!

How difficult is this to understand?

The only real benchmark will be only when the Barcelona win get over it.....

---

we dont know if k10 gonna be better than c2d so your comment wasnt all true...but i agree that most of benchmarks are not "fair"

Quote:

---

Originally Posted by duploxxx

hey its gODJO he knows it all... and he is very good in calculations and posting preffered when it is bad for AMD and good for INTEL.

its an nfore 3400 btw so that is a socket am2 board !!!! according to all nvidia papers. So how can it be a barcelona cpu? they are demonstrating budapest :shrug:

---

:rofl: Again you failed to make a point and made a clown of your self.
Please quote the part where I said Barcelona! :slap:

BTW, don't you have anything better to do than picking on me?
I mean, stop tingling my balls. If you have something to say related to the topic which will contribute to the thread and will help others, then go ahead and say it. If you just have to say something, go to MIRC or other chats. Don't waste my bandwidth.

Quote:

---

Originally Posted by informal

Compare to dual core K8 and you get that core per core this K10 sample would be slower than 4 years old K8 design,which doesn't make sense at all...

---

Because of the scaling of Cinebench, you can't compare 4 core vs. 2core. We'll need an underclocked Quad-FX system and C2Q system to do full comparisons....

EDIT: and everyone, stop attacking/bashing gOJDO....final warning, seen it in way too many threads.

Quote:

---

Originally Posted by LOE

some sites are spreading a lot of FUD these days, mainly sites that are generously sponsored by intel

I hope AMD gives answer to all this BS, it's not in their interest people thinking their new chip sucks

Things are pretty simple, you don't put years of development and millions of $$ into making a chip that is as slow as the chip you've been making all the time. There is no way amd release something slower than core2, 1 year after it's launch. You don't redesign a cpu core, eliminate all bottlenecks, add improvements, double the SSE throughoutput and get the same performance. But if there is some major problem and that is true, that means bye bye amd.

I hope we all have our questions answered in the next few weeks.

---

:yepp:

Quote:

---

Originally Posted by gOJDO

All K10 will have 128kB of L1 and 512kB of L2 per core and 2MB of shared L3 cache among all cores.

---

nope
there will be K10 desktop Dual Cores that wont have a L3 at all asfar as ive read.
i know that this is a Quad Core but but this was just so u know that not all K10s will have the cache u mentioned.

Quote:

---

Originally Posted by Starscream

nope
there will be K10 desktop Dual Cores that wont have a L3 at all asfar as ive read.
i know that this is a Quad Core but but this was just so u know that not all K10s will have the cache u mentioned.

---

That wouldn't be a K10. Rana would be a K8 Athlon64 X2.

Quote:

---

Originally Posted by Vapor

Because of the scaling of Cinebench, you can't compare 4 core vs. 2core. We'll need an underclocked Quad-FX system and C2Q system to do full comparisons....

---

Spot on. Where is Movieman w/ his Kentsfield? He does a lot of benching on that machine.

Quote:

---

Originally Posted by gOJDO

:rofl: Again you failed to make a point and made a clown of your self.
Please quote the part where I said Barcelona! :slap:

BTW, don't you have anything better to do than picking on me?
I mean, stop tingling my balls. If you have something to say related to the topic which will contribute to the thread and will help others, then go ahead and say it. If you just have to say something, go to MIRC or other chats. Don't waste my bandwidth.

---

make a point, made a clown of me, i just replied on 2 items of on your comments that are plain wrong

1) cache and latency does make a difference on a k8, you stated it didn't.
2) not all k10 have the same cache size.

again... it can't be barcelona because they used an am2 board, or the wrote down the wrong chipset or it was a budapest placed on an older am2 board.

@Vapor no wonder that we reply on him if you see most of his value added comments.

Quote:

---

Originally Posted by gOJDO

The highest clocked RAM that Barcelona will support is DDR2-667.

---

Where did you get this?

Barcelona will support up to DDR2-1066.

Quote:

---

Originally Posted by duploxxx

1) cache and latency does make a difference on a k8, you stated it didn't.

---

It's less than 1-2% in just about every application. Sure, it's there. But the difference is tiny.


Plus, if we are bit#$ing about 1-2% differences when we are talking about K10 visa via C2Q, there are bigger problems afoot.

Quote:

---

Originally Posted by gOJDO

That wouldn't be a K10. Rana would be a K8 Athlon64 X2.

---

How it can be K8 with 256bit bus to L1 cache, 2x128bit bus to L2, 128bit SSE, and other improvements from K10??

Have you got any prove to your conclusion that Rana is K8??

I think this was the most telling line of the whole article:

Quote:

---

We only had a few minutes to test the chip, but we were able to run a quick Cinebench before we were instructed to leave.

---

You don't scoot people away and try to keep them from running benchmarks when you have something hot to show off. You do that when you don't have all your ducks lined up.

This is not a benchmark I'd expect would change much at all with K10. K8 is allready strong at it, only slightly slower than core. Would certianly expect some improvement though, and to outperform core at least.

What we know of the K10 arch suggests it works on the K8's weaknesses, and not a lot more. Should see some interesting results when we finally start to see more benchmarks, If benches where K8 was weak don't fly with K10, then we know there's a problem.

lol gojdo reminds me of myself first few months at xs :p

And this is how I was dealt with back then ;)

CLOSED

Perkam