AMD 45nm Deneb Pictures, CPU-Z and Super Pi Results

[saveus222]

emm.. noob question.. but what the heck is DVT

[informal]

@gallag

It's called C0 DVT sample.D V T sample.Not production model.TDP will not be the same as C1.
Question i asked in Nehalem thread had nothing to do with power consumption of Nehalem(since it's also not the retail but ES),it had to do with results of the tests.Also Nehalem is on a platform that is not quite finished yet(single channel etc.) so the results could be even better than what we saw.

No you didn't hijack the thread and yes i had fun time .

[Mathos]

But, Nehalem isn't a whole new uArch. It's built off the Core2 Micro arch with a modular design that allows them to change certain parts of the die as needed. Biggest step forward is the IMC, which is mainly going to be a boost on memory intensive tasks. Or in the server arena tasks that were limited by FSB. The other big step is the reintroduction of hyperthreading. The other thing they did was make their cache system inclusive instead of exclusive like AMD. But they did that to get around the smaller L2 cache size, to better utilize the L3 cache.

Really what it will do is allow Intel to catch up on the server side, where AMD has a performance lead. If Shanghai manages to pull enough of an improvement over Barcelona, then AMD will still be in the game in that market. And, AMD only needs to improve about 10-15% clock per clock to compete with Penryn. Phenom Agena is only about 9% slower clock per clock than Conroe on average.

[Asgard_thor]

Yes yes we know that intel does better at superpi than AMD, they always have. But superpi is not a good indication of performance between the two. The only thing it is good for IMO is getting a little bit of an idea how a CPU compares to its predecessors, in the same company. Cross company comparison is not valid. Great example is P4 vs K8, P4 was faster in superpi but lost overall compared to K8.

didn't know that sorry

[KeZzZu]

dunno if those temp readings are right, 37c' idling @ 3.4 ghz + is pretty impressive with air cooling.

[BrowncoatGR]

I thought I read about DVT samples being available a few months ago?

I'm quite sure those where EVT samples. Way to early for DVT.

someone willing to explain the differences between the prefetchers of AMD and Intel?

Prefetching is tied to the cache structure so they are quite different. One important difference is that K10 also prefetches directly to the L1. This is a very detailed article on Phenoms memory subsystem. http://www.digit-life.com/articles3/...ma-phenom.html

emm.. noob question.. but what the heck is DVT

Design Validation Test. Mostly bug free samples used to validate the platform.

dunno if those temp readings are right, 37c' idling @ 3.4 ghz + is pretty impressive with air cooling.

I think that's unlikely. We don't even get accurate temps from Agena procs afaik.

[spursindonesia]

If the retail version can reach 4 GHz under water reliably, i think i'll comeback to AMD camp, unless Intel makes a serious price cut over Yorkfield high multi chip.

[IvanAndreevich]

If the retail version can reach 4 GHz under water reliably, i think i'll comeback to AMD camp, unless Intel makes a serious price cut over Yorkfield high multi chip.

Not going to happen, sorry. Best you can hope for is around 3.5 GHz.

[Bandwidth]

3.4GHz ES sample not bad

Kentsfield is on average about 10% faster than Phenom clock-for-clock. Yorkefield adds like 6% more performance to that, so around 16%. The new 45nm Deneb is supposed to be 15% faster than Phenom, so it should be a little faster than Kentsfield and about equal in performance to Yorkfield.

Power consumption will not match that of Intel 45nm CPUs, because Intel uses HighK-MG while AMD doesn't (yet). AMD expects power consumption to be 15% lower on Deneb, which is not that great but still helps.

Phenom is in many cases cache starved, 2MB of shared L3 cache is too small, and Deneb should do much better with 3x more L3 cache. Now I really hope AMD will be able to clock the NB higher than 2GHz with Deneb, as every 200MHz increase should give a 2-3% increase in performance. Doesn't seem like much, but when you have a Deneb running at 3.2GHz for both CPU and NB, the difference will be very noticeable

[Donnie27]

You think this is bad, You want to check out some of the "not happy" comments in the nehlam thread. A lot worse.

Here are a few, This one in particular is funny considering you are jizzing over AMD pi scores.

Hypocrite much? Perhaps instead of wondering why people are not jumping up and down with excitement and posting only positive things about 45nm phenom you should ask yourself why you where so negative in the nehlem thread?

Sweet! You made my day!!

[Wiggy McShades]

3.4GHz ES sample not bad

Kentsfield is on average about 10% faster than Phenom clock-for-clock. Yorkefield adds like 6% more performance to that, so around 16%. The new 45nm Deneb is supposed to be 15% faster than Phenom, so it should be a little faster than Kentsfield and about equal in performance to Yorkfield.

Power consumption will not match that of Intel 45nm CPUs, because Intel uses HighK-MG while AMD doesn't (yet). AMD expects power consumption to be 15% lower on Deneb, which is not that great but still helps.

Phenom is in many cases cache starved, 2MB of shared L3 cache is too small, and Deneb should do much better with 3x more L3 cache. Now I really hope AMD will be able to clock the NB higher than 2GHz with Deneb, as every 200MHz increase should give a 2-3% increase in performance. Doesn't seem like much, but when you have a Deneb running at 3.2GHz for both CPU and NB, the difference will be very noticeable

I don't think the nb/L3 clocking is really that important I've run them 1:1 on my phenom and its no faster than when its not. The higher the cpu speed gets the nb and l3 speed need to increase also but it seems like as long as you keep no larger than a 400mhz disparity in clock speed between the two then you aren't missing out on performance(doesn't totally work that way in memory benches. I'm speaking in respect to the Crysis benchmark and 3dmark). What we don't know is what if the three times larger L3 cache will have a larger effect from clocking it higher.

[Sparky]

I have the perfect smiley for this stuff that gallag brought up:
If you have a personal beef with someone, take it to PM, not turn a thread into a personal arguement.

Hipno650, glad your q6600 hits 3.9GHz (curious,is that 100% stable?), I wish mine did. But IIRC that isn't a common, guaranteed OC. Mine is only 100% stable at 3.45GHz for example. Roughly a 1GHz overclock. I got that same 1GHz boost out of my old opteron 165. Also, on the volts and power, remember this is not the final chip but a test sample. I'm not saying I know what the final specs will be, I don't, but I would imagine that they will be different than we see here as this is a first run.

[Bandwidth]

I don't think the nb/L3 clocking is really that important I've run them 1:1 on my phenom and its no faster than when its not. The higher the cpu speed gets the nb and l3 speed need to increase also but it seems like as long as you keep no larger than a 400mhz disparity in clock speed between the two then you aren't missing out on performance(doesn't totally work that way in memory benches. I'm speaking in respect to the Crysis benchmark and 3dmark). What we don't know is what if the three times larger L3 cache will have a larger effect from clocking it higher.

If you check the xbitlabs link here They tested with 1.8GHz and 2GHz NB, the gain was 1% on average, only 0.7% for Crysis, but 2.8% for Winrar.

Here we are talking about 2MB L3 cache, and gaining 1% performance per 200MHz increase in NB speed. So maybe as you said for 6MB cache the difference boost will be higher, that's how I assumed 2-3% gain in Deneb's case.

You know something just does not seem right with the Phenom NB, whatever it is, I hope it gets tweaked/fixed in Deneb.

[clayton]

I hope they get rid of that silly phenom name.
because that name isn't doing them anything good

AMD Carnage 10000.

[Trunks]

AMD Carnage 10000.

I'll buy it at high prize

[perkam]

"No No No, AMD does not have the first native 45nm Quad, we do. It's right here ahh.......Errr....ummm...Look !! There's a Nehalem !!"

This is good stuff. But probability of cherry-picking is high without other corroborating results.

Perkam

[annihilat0r]

AMD Carnage 10000.

How about DEVASTATOR 3500

I'd buy it regardless of performance of price. It would be great to have a processor with such a brutal name.

[Rammsteiner]

Damn, I go away a few hours and mis these stupid posts.

TBH dead simple, I hardly can spot a negative part in these posts from informal... There's still that difference between 'AMD sucks' or 'my Intel brainfarts this way' posting in AMD threads are just being a little bit disappointed by the results thus far from an overhyped wannabe expensive platform.

I wouldnt mind people saying they're disappointed by Deneb's performance, although it has to be fundamented well enough. But since K10.5 > K8 pretty much, there's hardly anything to complain about unless AMD will charge 1K$ for their top CPU again.

Also the posts before stating that AMD wouldnt even reach a 10% IPC on Deneb while it shows 12%+ already on an ES. Yerrr...

More I hang out, more I spot unproven statements posted in such a way to make the poster look smart and very well informed. Oh well, saw enough of that to know what people not to take seriously at all.

[Hornet331]

"No No No, AMD does not have the first native 45nm Quad, we do. It's right here ahh.......Errr....ummm...Look !! There's a Nehalem !!"

This is good stuff. But probability of cherry-picking is high without other corroborating results.

Perkam

lol still bringing this "native" marketing crap.

[_Lone_Wolf_]

Also the posts before stating that AMD wouldnt even reach a 10% IPC on Deneb while it shows 12%+ already on an ES. Yerrr...

With all due respect achieving such a speedup in a canned benchmark rarely translates to real world performance, a point I'm sure you appreciate. As terrace215 suggests the increase could purely be due to the larger cache, taking what amounts to pre-lunch hype at face value usually leads to dissapointment. The superpi figures are interesting but without more system level test results, proclaiming AMD IS BACK!!! at this point is premature.

[JumpingJack]

3.4GHz ES sample not bad

Kentsfield is on average about 10% faster than Phenom clock-for-clock. Yorkefield adds like 6% more performance to that, so around 16%. The new 45nm Deneb is supposed to be 15% faster than Phenom, so it should be a little faster than Kentsfield and about equal in performance to Yorkfield.

Not quite convinced this will be true ... I ran this just a bit and posted at Aces' so will put it here, a 2.5 GHz Yorky:



Nonetheless, deneb is looking much better overall ... SP1M is not a good indicator of the relative between the two, and this 'leaked' info is much better than what we saw pre-Barcey/Agena where nothing in the wild got shown.

Overall positive results for AMD...

jack

[demonkevy666]

Damn, I go away a few hours and mis these stupid posts.

TBH dead simple, I hardly can spot a negative part in these posts from informal... There's still that difference between 'AMD sucks' or 'my Intel brainfarts this way' posting in AMD threads are just being a little bit disappointed by the results thus far from an overhyped wannabe expensive platform.

I wouldnt mind people saying they're disappointed by Deneb's performance, although it has to be fundamented well enough. But since K10.5 > K8 pretty much, there's hardly anything to complain about unless AMD will charge 1K$ for their top CPU again.

Also the posts before stating that AMD wouldnt even reach a 10% IPC on Deneb while it shows 12%+ already on an ES. Yerrr...

More I hang out, more I spot unproven statements posted in such a way to make the poster look smart and very well informed. Oh well, saw enough of that to know what people not to take seriously at all.

I think thats great about 12.5%, means they could get more possible since it's only C0 stepping.

[terrace215]

With all due respect achieving such a speedup in a canned benchmark rarely translates to real world performance, a point I'm sure you appreciate. As terrace215 suggests the increase could purely be due to the larger cache, taking what amounts to pre-lunch hype at face value usually leads to dissapointment. The superpi figures are interesting but without more system level test results, proclaiming AMD IS BACK!!! at this point is premature.

Yes, Dirk himself was asked if there were any major core improvements, and said no. A few tweaks, and a larger L3. So if you choose a benchmark that is quite sensitive to cache size, you're going to see some improvement, but you can't extrapolate this improvement to the "average application". I'd guess the larger L3 & other tweaks might bring 5-7% clock/clock improvement on avg, but we'll have to wait for more benchmarks to see.

[LightSpeed]

I think thats great about 12.5%, means they could get more possible since it's only C0 stepping.

12.5% on superPi is outstanding, since amd sucks at it anyways

im sure we'l see greater performance increases in other apps in which amd usually scales better

and abt relative performance compared to yorkies, i assume they will be within 5% the performance of em..

which is great news, but if only amd can clock these up to 3.2 which is seemingly improbable at the start

im sure this was cherry picked however, hve seen lots of the cherry picked 65nm k10s clocking great and raising my hopes and then seeing the retail oc`s makes me disappointed again. i hope not tho

[Epsilon84]

With all due respect achieving such a speedup in a canned benchmark rarely translates to real world performance, a point I'm sure you appreciate. As terrace215 suggests the increase could purely be due to the larger cache, taking what amounts to pre-lunch hype at face value usually leads to dissapointment. The superpi figures are interesting but without more system level test results, proclaiming AMD IS BACK!!! at this point is premature.

I'd agree. If it was a 12.5% jump in something other than SuperPi, I would have been really impressed. However, SuperPi is quite sensitive to cache size, and a 12.5% boost in SuperPi generally does not correlate to same increase in most real world applications, unless the application was also very cache dependant. For example, when comparing C2Ds of different cache sizes we get:

( Source )
E6600 (4MB L2) @ 3.5GHz ~14.7 secs
E6400 (2MB L2) @ 3.5GHz ~16 secs

An E6600 is ~9% faster than an E6400 per clock in SuperPi, but the real world difference between the two is actually only ~3.5%.

Now considering we are comparing a Phenom w/2MB L3 with a Deneb w/6MB L3, could it be possible that the majority of the performance increase is down to the much larger cache?