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

[Rammsteiner]

I know, still I think it's not impossible. It's not only just larger L3 cache mind you. The L3 cache is now 48-way set associative instead of 32-way and there are a few more little tweaks.

We saw a 12% increase in SuperPI although that shows us in the end all. Although Phenom scales badly in SuperPI so it's interesting at least.

Of course Deneb wont be >= Yorkfield in everything, no way. But I do think in quite a bit more programs than Agena >= Yorkfield is right now to such extend it's possible to say Deneb => Yorkfield.

[Helmore]

For Deneb to be >= to Yorkfield, it'll need to be 15 - 20% faster than 65nm Phenoms. This seems a bit unrealistic, when was the last die shrink (even with core enhancements + cache increases) that yielded a 15 - 20% increase in performance? Northwood was about 10% faster than Willamette, thats the biggest I've seen from a die shrink. In more recent times, Penryn is some 5 - 6% faster than Conroe. Off the top of my head, I don't remember any AMD die shrinks that have resulted in significantly improved performance, in fact performance went down slightly in the last K8 shrink from 90nm to 65nm.

Now obviously Deneb will increase performance some, the larger L3 alone guarantees that, but all I'm saying is that 15 - 20% is a LOT to ask for. K8 to K10 improved by about that margin per clock, and we're talking an architectural overhaul. You're asking for the same increase from a die shrink + some enhancements, it seems a bit optimistic IMO but time will tell. I personally think Deneb will be 5 - 10% faster per clock, putting it around Kentsfield levels but slightly slower than Yorkfield. Hopefully I am wrong and Deneb lives up to the hype you guys are heaping on it, but I've sort of lost confidence in AMD's ability to deliver, especially on the CPU side of things.

Yeah I also believe that a 5-10% increase in IPC is more realistic for Deneb, although the average will probably be closer to that 5% increase and DDR3 could increase that a little.
About that K8 shrink from 90 to 65 nm., AMD did not want to invest anything into that shrink and they only did the necessary things to get it working on 65 nm. and no optimizations at all. I believe they could have made their 65 nm. part on par of higher performance per clock compared to their 90nm. part but this was probably not worth the hassle. Getting a mature 65 nm. process up and running for Phenom was probably more important, though even there they haven't gone to the same extent as they did with their 90 nm. process and are now primarily focusing on their upcoming 45 nm. process.

@Rammsteiner: Is there any word on the clocks of the L3 cache on Deneb? Is there still a clock deficit between the L3 cache and the cores? I believe they could get a nice boost by clocking them at the same speeds, nothing ground braking probably and a downside might be a decrease (over)clocking potential. But these are all changes to the overall architecture of the chip, the cores them self could use a healthy boost and I don't think that adding some SSE instructions will do so right now.

[Epsilon84]

I know, still I think it's not impossible. It's not only just larger L3 cache mind you. The L3 cache is now 48-way set associative instead of 32-way and there are a few more little tweaks.

We saw a 12% increase in SuperPI although that shows us in the end all. Although Phenom scales badly in SuperPI so it's interesting at least.

Of course Deneb wont be >= Yorkfield in everything, no way. But I do think in quite a bit more programs than Agena >= Yorkfield is right now to such extend it's possible to say Deneb => Yorkfield.

Cache associativity normally increases with larger caches anyway, at least thats what I've noticed on Intel CPUs. I think its needed to help the cache 'scale', I'm no expert on this matter though, I'm sure someone like JumpingJack can give a much better technical explanation.

I agree the SuperPi result doesn't tell much about real world performance (when does it ever? ), which is why I'm hoping someone leaks results for more 'useful' benchmarks.

Its quite typical of an AMD pre-launch though, they are always so secretive! Sometimes I wish they would be a bit more open like Intel, they allowed Anandtech to preview Nehalem a quarter prior to release, and they did the same with Conroe too.

[cky2k6]

i guess intel can afford to be more open. though at this point, amd might as well spill it cause they don't have much to lose...

[Donnie27]

i guess intel can afford to be more open. though at this point, amd might as well spill it cause they don't have much to lose...

Just as I said before Barcelona launched, if AMD has something worth a damned, it would leak it as well. Sure it'd be on or featured on a biased AMD site like the AMD zone Intel can afford to not be so secretive because just like Conroe and Penryn, they have another in the words of MaxPC, "Kick @ss" product.

Kind of Funny how lies are not thread crapping then folks wonder why so many and are posted.

[Rammsteiner]

Just as I said before Barcelona launched, if AMD has something worth a damned, it would leak it as well. Sure it'd be on or featured on a biased AMD site like the AMD zone Intel can afford to not be so secretive because just like Conroe and Penryn, they have another in the words of MaxPC, "Kick @ss" product.

Kind of Funny how lies are not thread crapping then folks wonder why so many and are posted.

Actually the only '' and '' I see is from your FTW, QFT! posts. Maybe you might start to think that not the whole world is wrong, AMD zone visitor and a liar but the problem is somewhere around you.

Also it might help to tell us where the lies are instead of just stating that people lie.

Besides that, AMD hardly leaks stuff in the first place. SB750 isn't anything to be shamefull about and it's not leaked. Only Anandtech got a way around the NDA and showed some results which actually prove my earlier statement (a lie?) that all Phenoms, at least B3, can do 3Ghz.

And if you read reviews which reward products with 'Kick @ss'... It gives me more and more the kind of idea who you are

Also, nVidia did leak things hear and there and you see where they're now with the GTX2x0 price drops. No offence, but leaks or not, it's no guarantee at all for the results of the actual product.

[BrowncoatGR]

Just as I said before Barcelona launched, if AMD has something worth a damned, it would leak it as well. Sure it'd be on or featured on a biased AMD site like the AMD zone Intel can afford to not be so secretive because just like Conroe and Penryn, they have another in the words of MaxPC, "Kick @ss" product.

Kind of Funny how lies are not thread crapping then folks wonder why so many and are posted.

AMD does not leak stuff. Just because Intel started previewing new tech early does not mean that all other companies do, nor that Intel has always done this.

[The Coolest]

Source

Looks promising if true. Just for the record, if this image is legit, this is a newer revision than the one in the first post.

[JumpingJack]

AMD does not leak stuff. Just because Intel started previewing new tech early does not mean that all other companies do, nor that Intel has always done this.

Actually, this behavior of Intel is relatively new ... up until Conroe they were always very tight-lipped about up coming products.

EDIT: If I may speculate, as I recall this type of marketing/preadvertisement started essentially after Otellini stepped up. Otellini, as I recall, has expertise in the marketing field, not engineering .... engineers tend to hunker down and talk quietly about their stuff until it is ready, Otellini as a marketer may embrace the idea that getting some buzz started early is a powerful deal. A theory anyway.

Jack

[bowman]

Source

Looks promising if true. Just for the record, if this image is legit, this is a newer revision than the one in the first post.

If the voltages are accurate and the shot is legit, that's incredibly good. Monolithic quad die, 1.1V, 4GHz? Maybe enthusiasts might look at AMD once again..

[STaRGaZeR]

If the voltages are accurate and the shot is legit, that's incredibly good. Monolithic quad die, 1.1V, 4GHz? Maybe enthusiasts might look at AMD once again..

Look at Overdrive voltages, 1.475v

Not so good in my book, but an improvement over Agena I guess...

[The Coolest]

If the voltages are accurate and the shot is legit, that's incredibly good. Monolithic quad die, 1.1V, 4GHz? Maybe enthusiasts might look at AMD once again..

1.475v - look at the AOD window in the back.

Edit: STaRGaZeR was first

[insurgent]

Source

Looks promising if true. Just for the record, if this image is legit, this is a newer revision than the one in the first post.

The numbers on the next 3 frequency boxes look funny, the zeroes are big enough to touch the boundary and looking at other overdrive screenshots, I don't think that's how they should look like.

[foch3 -USA-]

I wish I could believe..... That pic just can't be real, there is no way they have come that far in such a short time.

[ryboto]

The numbers on the next 3 frequency boxes look funny, the zeroes are big enough to touch the boundary and looking at other overdrive screenshots, I don't think that's how they should look like.

In the AMD sub-forum, users posted AOD screenshots that show the same small-font behavior in the first frequency box. People even zoomed in the CPUZ shot and tried to disprove the screen shot due to a lack of pixel distortion, but that theory was disproved to.

[ryboto]

I wish I could believe..... That pic just can't be real, there is no way they have come that far in such a short time.

On an SB750 board, the 9950 has been reviewed by Anand as reaching 3.5ghz stable, why would 4ghz be such a reach? no one said it was stable.

[JCornell]

I wish I could believe..... That pic just can't be real, there is no way they have come that far in such a short time.

That's rude for saying fake with no actual prove
Lucky me no one mention it was fake at Nehalem thread

[bowman]

Look at Overdrive voltages, 1.475v

Not so good in my book, but an improvement over Agena I guess...

Ah well. Too good to be true. Unless they've discovered a magical 45nm process that can somehow take 1.45 and live to tell the tale..

[informal]

Ah well. Too good to be true. Unless they've discovered a magical 45nm process that can somehow take 1.45 and live to tell the tale..

Actually AMD's SOI can take much more voltage than intel's sensitive high-k bulk process,at 45nm.I wouldn't be surprised to see AMD chips run at higher than 1.4V for OCing purposes.

[KeZzZu]

Look at Overdrive voltages, 1.475v

Not so good in my book, but an improvement over Agena I guess...

well my celeron 440 ( conroe-L ) required 1.7vcore to get 4 ghz, so good result at least in my eyes.

[insurgent]

In the AMD sub-forum, users posted AOD screenshots that show the same small-font behavior in the first frequency box. People even zoomed in the CPUZ shot and tried to disprove the screen shot due to a lack of pixel distortion, but that theory was disproved to.

I'll check that, thanks. I just googled AOD images earlier and all the images I saw had similarly sized fonts, I can't account for the version numbers though.

[STaRGaZeR]

well my celeron 440 ( conroe-L ) required 1.7vcore to get 4 ghz, so good result at least in my eyes.

Yeah and my E8400, which is a bad clocker, does 4GHz at 1.33v. Tell that to the folks with an E8600 E0 stepping, 4GHz at 1.21v. Nice comparison, new shinny high end 45nm AMD processor vs very low end, 65nm Intel processor

This is only a good result if you compare it to Agena. If you start talking about Intel it's bad, very bad. And you know it, seriously I don't know why are you trying to compare apples to cars. Because almost every Intel user here can show you an apple being faster than the car

[FlanK3r]

I think its true, on MB with SB750...Agenas are now with default voltage on 3GHz easily and with 1.4-1.5 about 3.5-3.6GHz. And this is ES Deneb, final products will about 3.5-3.8GHz max OC with air..

[bowman]

Actually AMD's SOI can take much more voltage than intel's sensitive high-k bulk process,at 45nm.I wouldn't be surprised to see AMD chips run at higher than 1.4V for OCing purposes.

Well, there's OCing and OCing, I'm thinking of 24/7 clocks. Anything else is completely uninteresting to me.

[Rammsteiner]

@Rammsteiner: Is there any word on the clocks of the L3 cache on Deneb? Is there still a clock deficit between the L3 cache and the cores? I believe they could get a nice boost by clocking them at the same speeds, nothing ground braking probably and a downside might be a decrease (over)clocking potential. But these are all changes to the overall architecture of the chip, the cores them self could use a healthy boost and I don't think that adding some SSE instructions will do so right now.

Sorry, missed this part. Thus far it seems they clock the Northbridge still at 1.8Ghz, so excluding any eventual tweaks it would be just as fast. Deneb BE/FX might be clocked at 2Ghz again or higher.

Yeah and my E8400, which is a bad clocker, does 4GHz at 1.33v. Tell that to the folks with an E8600 E0 stepping, 4GHz at 1.21v. Nice comparison, new shinny high end 45nm AMD processor vs very low end, 65nm Intel processor

Funny to see how lots of people post their Yorkfield results in AMD threads to say Agena is crap, but if an Intel user does the opposite it's bad?

Anyway, it's bad anyway, just find it funny this is the first time someone critizes it not being an AMD user. However, you do the same with your C2D remark:

4Ghz at 1.33V for a Dual Core with High/k and metal gates that is which had quite some improvements/steppings already. This Voltage on an ES 45nm Quad without High/k and metal gates, I think it's very impressive.

It's not superb, but if you remember where we're coming from, B2 Phenoms, B3 Phenoms + SB600... Basicly all B3's (if not B2's as well) can clock well beyond 3Ghz now. I do hope they get rid of the few flaws in the NB though cause atm it's even slower clock for clock with Agena.