IBM published Barcelona SpecCPU results.

[kl0012]

SPEC CPU2006
Benchmark
x3455 – Quad-Core AMD Opteron Model 2347 Processor
(1.9GHz, 512KB L2 Cache per Core)

SPECint2006 11.3
SPECfp2006 11.2

ftp://ftp.software.ibm.com/eserver/b...cpu_091007.pdf

For example the best results of Xeon 1.86GHz - 14.2/12.5 for SpecInt/SpecFP

[awdrifter]

What would a similarly clocked K8 do on this?

[kl0012]

What would a similarly clocked K8 do on this?

2.2-2.4 GHz Opteron

[madcho]

rate is not bad

[w0mbat]

@kl0012: A similar clocked K8 would do 2,2-2,4GHz? What do u mean?

[kl0012]

@kl0012: A similar clocked K8 would do 2,2-2,4GHz? What do u mean?

No. I mean 2.2-2.4 GHz Opteron can match these scores. Ther are no published results for 1.9 GHz Opteron (K8).

[red]

SPECint2006 11.3
....
http://www.spec.org/cpu2006/results/...802-01605.html
10.7 specint 2ghz k8 (cores don't matter for single threaded nonrate benches)
http://www.spec.org/cpu2006/results/...820-01837.html
11.5 specint 2.2ghz k8

SPECfp2006 11.2
...
http://www.spec.org/cpu2006/results/...802-01603.html
2ghz k8 scores 10.9
http://www.spec.org/cpu2006/results/...814-01803.html
2.2 scores 11.6
===
SPECint_rate_base2006 score of 72.8 is in between a pair of 4 socket 2,2.2ghz chips 8 cores http://www.spec.org/cpu2006/results/...806-01678.html and http://www.spec.org/cpu2006/results/...806-01674.html

or almost double 4 2ghz k8 cores http://www.spec.org/cpu2006/results/...802-01602.html
----
2 socket 8 k10 1.9ghz are comparable to 4 socket 8 k8 2ghz in SPECfp_rate_base2006 http://www.spec.org/cpu2006/results/...806-01680.html
or 70% better than 2 socket 4 k8 2ghz (but recent result, optimized) or almost 100% than 2 socket 4 1.9ghz k8 http://www.spec.org/cpu2006/results/...416-00875.html

I don't see much improvement per core...

[v_rr]

I don't see much improvement per core...

If that new is true....
Or we will see tomorrow

[awdrifter]

Wow, that's pretty small improvement from the K8 (2.7-5.6&#37. They can probably get those results from the K8 if they increased the L2 cache to 2-4mb per core.

[Ubermann]

If that new is true....
Or we will see tomorrow

So Dell is wrong and now IBM their close "partner" is wrong and cooaler is wrong.
Only one that is right is a dude at a forum so far (cuz he had something good to say) ?
It does not sound to good about K10 and i fear thats a fact tomorrow.
I guess some people in here will abandon XS for a while tomorrow =)

[red]

yah, but clock for clock performance/improvement is irrelevant. 4 1.9ghz quads is likely cooler than 2 3ghz k8 and at least in server, cheap. speccpu does not test for realistic desktop usage (or any specific one for that matter) so we need workstation apps to test SSE improvements (non broken runs of Cinebench, 3dsmax etc)

SPECint_rate_base2006 score of 72.8 beats 4 3.2ghz k8 by 30% http://www.spec.org/cpu2006/results/...723-01515.html and SPECfp_rate_base2006 of 68.5 beats 4 3ghz k8 by 38% http://www.spec.org/cpu2006/results/...703-01354.htmlExcept for rate, you must be using exorbitant amounts of bandwidth which is not likely.

And in single thread INT, http://www.spec.org/cpu2006/results/...604-01202.html k8 is 42% faster (3ghz k8 vs 1.9ghz k10, btw 3/1.9 does not mean 57% more performance, shown above similarly clocked k8s)

[Jimmer411]

Wow, that's pretty small improvement from the K8 (2.7-5.6%). They can probably get those results from the K8 if they increased the L2 cache to 2-4mb per core.

Cache has a totally different effect on Intels cpus than it does on AMD ones. There is virtually no noticable difference in performance between 256, 512 and 1mb per core cache on A64s. where the difference between 2mb and 4mb is quite noticeable on C2D.


Im interested in seeing how the power functions of the new processors does. Im selling my laptop off since alot of the features look very nice for notebooks.

[red]

http://www.realworldtech.com/page.cf...WT060407195417
next mobile amd chip is k8 with k10 northbridge or something like that...

[hollo]

they should have skipped the l3 cache and released it at the start of the year with decent clockspeeds

if i ever meet hector ruiz i'm going to stab him in the eye with a pen

[Ubermann]

Good luck but i think hector has twice the amount of bodyguards tommorow.
You need a spear..

[Hornet331]

Good luck but i think hector has twice the amount of bodyguards tommorow.
You need a spear..

and a good spartan warrior.

[AliG]

dammit amd, why did you do this to yourself? They knew what intel was capable of doing once they saw conroe, and even though they were losing millions of dollars and even a few billion now, they still haven't gotten their act together and only pretty much tweaked k8's performance, k10 seems more like a core update than an entirely new marchitecture on it's own

[mstp2009]

dammit amd, why did you do this to yourself? They knew what intel was capable of doing once they saw conroe, and even though they were losing millions of dollars and even a few billion now, they still haven't gotten their act together and only pretty much tweaked k8's performance, k10 seems more like a core update than an entirely new marchitecture on it's own

FYI, the design for K10 was pretty much set in stone before C2D was released. There was little AMD could do to tweak the design so late in the game. I re-design would have taken about 18 months (longer), not counting debugging.

[AliG]

one thing I'm wondering about is if a lot of this could deal with the cache? Amd hasn't messed with l3 caches in a while, and their ln2 caches are way smaller than intels (and I also believe slower too). I really wonder how much of a performance impact could that have? Because in 2008 k10.5 comes out with a bigger l3 cache, and in 2009 the individual cores get bigger caches too, not to mention full sse4 and sse5 at that point. Assuming amd stays alive, and they learn from their mistakes and tweak their ipc to have more instructions for single threaded tasks, k10.5 and k11 could possibly be serious nehalem competitors

[AliG]

FYI, the design for K10 was pretty much set in stone before C2D was released. There was little AMD could do to tweak the design so late in the game. I re-design would have taken about 18 months (longer), not counting debugging.

If that's the case, then why only make such a small upgrade from the start? Common, your competition is bound to wow you at some point and time, and since they can't afford to keep designing new architectures at the rate intel can, k10 should have been a huge upgrade that at least doubled k8's performance as it was planned to last for multiple years. We can now see that amd plans to do a k10.5 and k11 very rapidly after k10's launch and hopefully they can improve their ipc and give the cpus more cache+higher clockspeeds, but still, that should have been done from the start, same way 65nm should have been down over a year ago for them, not try to fix it at the last moment, and that's why they got caught with their pants down

[dexman]

i'm eager to see what excuse will amd use tomorrow if this holds true.

[Jacky]

If that's the case, then why only make such a small upgrade from the start?

All x86 cpus have predecessor, on which they are based on (e. g. core 2 is a revamped core). Every company has to start to desgin their chip from a certain base, there is no processor "designed from scratch".. there are no revolutionary, new designs. All x86 cpus have a LOT in common.
It's all about tweaking old designs.

You can only revamp a core that much, otherwise if you choose a design that is too ambitious and new you risk even more severe problems (implementation of new features is really hard).

The Barcelona project was already VERY ambitious. Thus it's quite late.. and it probably was only designed to take on core 2 duo but not penryn (that's why it seems kinda "slow").

It seems that AMD concentrated on throughput, multi-core scaling and HPC. I don't know if they intended to do it from the start or it was the best they could do with their transistor budget.
However the Barc. arch falls short of Conroe in single threaded applications that is OBVIOUS. Intels huge cache and agressive prefetch is great for single threaded aps..

Intel and AMD engineers are DAMN smart, and all their cores are heavily tweaked, and they have to keep to a transistor budget on a certain node - the chip can't be to big and to hot for the node it is produced on - and every chip is unbelievably complex. To extract even more performance out of such a complex design is quite a task (the individual barcelona cores can't be much bigger than 65nm K8, because then the die would be simply too big for 65nm... so they had to use some clever tricks to get more single threaded AND mutli threaded performance out of the design)

That's why there are no 5-issue, 128 bit, octa cores with 16mb cache. It's simply not feasible when you have to keep to a certain die size.

Tweaking anything is an extremely difficult task (because almost everything is already heavily optimised!) - that's why I always said that barcelona SIMPLY CANT BE 100% faster in common real world tasks (excluding IMC advantage in throughput) than conroe (e.g. when there were those stupid theinq lies about 30k 3dmark...).

If everything was more or less equal (architecture) than Intel would have to be better, because they have a much more advanced process.. however the architectures are very different in some aspects, but the point is, it's much easier for Intel to win, because of their sophisticated 45nm process.

Imho barcelona is a very specialised core, too specialised, which will be unmatched in throughput.. but that won't sell, they'll have to price their processor according to overall performance and the weak int performance.

[JVguest]

If that's the case, then why only make such a small upgrade from the start? Common, your competition is bound to wow you at some point and time, and since they can't afford to keep designing new architectures at the rate intel can, k10 should have been a huge upgrade that at least doubled k8's performance as it was planned to last for multiple years. We can now see that amd plans to do a k10.5 and k11 very rapidly after k10's launch and hopefully they can improve their ipc and give the cpus more cache+higher clockspeeds, but still, that should have been done from the start, same way 65nm should have been down over a year ago for them, not try to fix it at the last moment, and that's why they got caught with their pants down

K10 is a major upgrade, and very similar to core 2 but with 3 complex decoders instead of Core 2's 3 simple + 1 complex, has slightly weaker OOO and smaller L2+L3, but retains a larger L1 cache, better instruction fetch bandwidth and the IMC advantage. What K10 needs is clockspeed, there's nothing remotely mediocre about the architecture. The Spec numbers don't tell much at all, as K8 performance is similar in spec to core 2, but in real world heavy SSE2 use core 2 blows away K8 and typically outperforms it in other typical applications. Tomorrow we'll know how K10 performs (hopefully) but I expect very good performance (clock for clock) particularly in video encoding/decoding, games and pure number crunch applications. Too bad the launch clock speeds will be low. Looks like reliable motherboard availability will be low as well for some time too. So for now, Intel remains in the lead till early Q1 2008.

[generics_user]

K10 is a major upgrade, and very similar to core 2 but with 3 complex decoders instead of Core 2's 3 simple + 1 complex, has slightly weaker OOO and smaller L2+L3, but retains a larger L1 cache, better instruction fetch bandwidth and the IMC advantage. What K10 needs is clockspeed, there's nothing remotely mediocre about the architecture. The Spec numbers don't tell much at all, as K8 performance is similar in spec to core 2, but in real world heavy SSE2 use core 2 blows away K8 and typically outperforms it in other typical applications. Tomorrow we'll know how K10 performs (hopefully) but I expect very good performance (clock for clock) particularly in video encoding/decoding, games and pure number crunch applications. Too bad the launch clock speeds will be low. Looks like reliable motherboard availability will be low as well for some time too. So for now, Intel remains in the lead till early Q1 2008.

IF manufacturers bring bios updates with the phenom launch there will be more than enough motherboards

[JumpingJack]

K10 is a major upgrade, and very similar to core 2 but with 3 complex decoders instead of Core 2's 3 simple + 1 complex, has slightly weaker OOO and smaller L2+L3, but retains a larger L1 cache, better instruction fetch bandwidth and the IMC advantage. What K10 needs is clockspeed, there's nothing remotely mediocre about the architecture. The Spec numbers don't tell much at all, as K8 performance is similar in spec to core 2, but in real world heavy SSE2 use core 2 blows away K8 and typically outperforms it in other typical applications. Tomorrow we'll know how K10 performs (hopefully) but I expect very good performance (clock for clock) particularly in video encoding/decoding, games and pure number crunch applications. Too bad the launch clock speeds will be low. Looks like reliable motherboard availability will be low as well for some time too. So for now, Intel remains in the lead till early Q1 2008.

Yep... on throughput, and large working set workloads, this is a great core... well suited for high data rate FPU and HPC applications, this is the excess bandwidth AMD put emphasis on during design. Even the 2.0 Ghz will be shining in a few places over Intel's top CPU I suspect.

The IBM non-rate data is a glimpse into the actual core (BW not a factor) performance relative to core microarch. It catches up but does not surpass if this data is indeed accurate. This would mean on desktop, C2 will likely retain the performance crown, and AMD's shot at moving up will depend on how quickly they can ramp clockspeeds.