Actual deneb review/comparison to Intel

[JAYDEEJOHN]

I would like your posts to my comments/links on perf. Ive never done alot on the perf end, as we simply didnt know. Show me. As to the ocing, its all been done, and is being done. Show me the perf links thatre so far off please?
As for Kentsfield, since P2 is closer to it than Yorkfield, thats the only reason I bring it up. As to the fast enough thing, kinda goes against alllllllllllllllllll my posts regarding ocing doesnt it? Dont paint me out as a fanboy.
As for unlocked multis and OEMs, I as you, simply dont know, and itd be a crime if this were true, but some (like the XS systems from Dell) do allow it.

QAgain, show me the links to these questionable results, as theyre the same as Anand and others have read and followed. They were done by respected people, from coolaler to K|ngP|n, and have been reproduced by others since, so again, which ones?

 

[JAYDEEJOHN]

There was 1 link I posted where Agena was compared to the Q6600 in real world apps that someone did on their own. That was in response to someones elses doing the same, yet mine is no more proof or questionable than the other. At the time, its all we had, and again, someone was trying to shoot P2 down using something that isnt as varifiable as say Anands, and why would they do this? And include negative comments along the way?
Ive been misunderstood for bringing out the bright spots of P2, and no, Ive not got my head in the sand. I would loved to have seen better perf, and even higher ocees. Being thats its only a shrink, its done well, and being that i7 is a whole new arch, and the cost etc, and its showing with current gpus, but hopefully a better showing with future ones, I was somewhat disappointed.

Ive been accused of being a fanboy for expecting more from a whole new arch, while singing the praises of a cpu that almost outdoes its predessessor as much as a whole new arch which was already ahead does.And it was a simple die shrink, and is alot cheaper.Thats reality

 

[gpippas]

Personally I have my preferences. I like AMD and ATI over Intel and Nvidia. But I will always choose the best price/performance option for me at the time. Its like all the idiots who argue that there Playstation is better than the xbox and so on. Who cares. Buy the one with the games you want to play.

 

[rangers]

dont really care one way or another, ill put in a cpu from ether company, what ever cpu is the fastest (at my price point) thats the one ill go for

 

[jed]

Ive been accused of being a fanboy for expecting more from a whole new arch, while singing the praises of a cpu that almost outdoes its predessessor as much as a whole new arch which was already ahead does.And it was a simple die shrink, and is alot cheaper.Thats reality

That's what make pretty much full of s%$#.
Intel was ahead so their new chip was suppose to
blow away their predessessor.
You must have forgotten that before phenom it
was the FX series which at that time was beating
intel, so AMD was in the lead at the time.
now when phonom came out it didn't blow away
the FX chip at all, and that was a new arch.
Native Quad in all,so wheres the dissapointment
for the phenoms.

 

[fazers_on_stun]

Yo, folks - just a suggestion but let's give JDJ a break here and not pile on so much. I know I did some bashing in posts above, but I have to agree with him - I never saw him post anything on P2 performance other than between Kentsfield and Yorkie, and frankly looking at all the benches on all the sites, that seems to be where it is. Yeah there's reviews showing it lower than a Q6600 and probably others showing it beating a Cray V, but in the end, I think the AM2 owners will find it a nice upgrade and really that's all that matters.

Maybe JDJ got a bit enthusiastic with the OC stuff but he's alright in my book (although his avatar could use a dose of bleach to get all that green out ). He's a far sight better than the infamous Mrs. Bee-yotch or 9-inch or MMM. I just wish he could learn to tipe and spelle gooder .

 

[gpippas]

You forgot Barron.

 

[spoonboy]

Well now you want to debate the quality of the review sites?? I could point to a few you listed - techspot for one - that seem a bit on the bogus side, as in using Superpi for a bench. At least [H] gives you sufficient details on the test setup

It sounds to me like you just want to limit the gaming reviews to those that "prove" your point. Lessee - low res with max eye candy, but no SLI or tri-SLI because that makes P2 look bad.

Even the AMD-favoring Inquirer says

The good: Faster than low end Penryn chips in CPU-specific tasks

The bad: Intel chips remain best for gaming. Phenom II has slightly higher power draw than Penryn chips.

No really dont get me wrong im not saying that cos i think hardocp is 'teh suck' cos it dont support what im saying, it just really strikes me as a very odd/poor gaming performance benchmark. 1 title with tri-sli?!? wtf??? when does that tell you anything??? If u dont think its an odd way to guage gaming performance, well, hmmmmmmmm...

your not dissing 'the spot' are you?

Some sites are still wedded to some old benchmarks, and techspot does lean towards the oc community a bit, certainly more than tom's does. Its a dual english/spanish site, so maybe the spanish/south american enthusiasts are into the superpi thang. Anyway its a good site dont be a snob.

 

[spoonboy]

Sorry, but "the masses" buy PCs from places like Dell which have generally use a locked bios and don't allow overclocking, even with unlocked chips. Maybe the software-based overclocking utilities work on them, but I haven't found those to be great long-term overclocking solutions.

My posts, my links werent done to persuade anyone here. It was simply done for the excitement about a decent product . Theres been more negative, refuting, obstinate bellyacheing going on here, for what? Is Kentsfield now crap? If Penryn is so great, why arent Kent owners flocking to it? If i7s perf is so great over anything else, then why worry about its price? Is it only Kent owners who are being obstinate here by not going to Penryn or i7? Is it only AMD fanboys saying i7 costs too much?

While I don't really fault you for being excited about the launch you did link to a whole lot of questionable web sites that showed some unrealistic results.

By the way, "Is Kentfield now crap?" reminds me a lot of the "it's fast enough for me" quotes back when Phenom was originally released. No Kentsfield isn't crap; it's a decent product but it's a lot cheaper than the Phenom II 920 and two generations out of date.

Its done what early posts said it can do, regardless the naysayers.

This is simply not true. Several of your links in the other posts indicated that the Phenom II 940 could beat out the i7 920 and in some cases even keep up with the i7 965. From what I've seen it can't even come close to the i7 920 in most benchmarks.

Now, the thing is... I do think the Phenom II is a decent chip. It performs well when compared to its closest competitors. I wasn't expecting a whole lot from Deneb considering it's pretty much just a die shrink with additional cache, but it does perform as well as I could have hoped. I don't see it living up to the crazy hype that some of the early "previews" had suggested but it does give AMD a decent competitor to intel in its price range.

way to go splitting hairs. Its perfectly obvious he meant for the 'masses' of the enthusiast community. You knew what he meant. Its like replying to posts to tell someone about some spelling mistakes.

 

[epsilon84]

As for Kentsfield, since P2 is closer to it than Yorkfield, thats the only reason I bring it up.

But is it really? You keep insisting Deneb is faster than Kentsfield and almost at Yorkfield levels, but I'm not so sure:

This is a standadised score based on numerous reviews, with a PII 940 being the mid point at 100. Just going by points divided by clockspeed:

1) Q9550: 104 / 2830MHz = 0.0367
2) Q6600: 86 / 2400MHz = 0.0358
3) PII 920: 95 / 2800MHz = 0.0339
4) PII 940: 100 / 3000MHz = 0.333

Might wanna rethink your 2 - 3% faster estimate there JD... if anything its the other way around IMO.

A minor caveat that I think should be mentioned for the sake of fairness is that many of the C2Q results were benched using DDR3 which does have a slight performance advantage compared to DDR2. Since most Core2 users are still using DDR2 due to cost reasons, it should be noted that the performance advantage to Core 2 being shown here is a 'best case' scenario and Kentsfield would be just about equal with Phenom II if using DDR2.

 

[WR]

From the first page of this thread:

My guestimate was 2-3% higher than Conroe clock per clock overall. Its looking like this may be true. If anyone downs this cpu, theyre also downing the Q6600s et al as well. We need more info still, and of course pricing and power usage numbers would help, as well as overall ocing abilities, which I see as 3.8 to 4.2, but again time will tell

The point is not whether you claimed as fact that it would OC to stated values - even if you mistakenly wrote that far, nothing but insider knowledge would assure you of that - but that you repeatedly and strongly suggested, or heavily entertained, the thought that it would put a lot of pressure on Yorkfields by matching or exceeding the average frequency headroom, based on what looks to be entirely AMD's show (their ES CPUs, their live demos, no mass sampling). And you got caught in AMD's bluff again, though it is not as severe and obvious of a bluffing attempt as with Phenom I.

This time, there was light bluffing with the cherry picks (e.g., having audience members randomly select CPUs out of an already 25-50% cherry-picked versus initial retail tray), but the more important problem was how they validated the overclocks. Just like you should take everything claimed in an infomercial with a grain of salt, when all they demo is Crysis, and all they require to claim an overclock is 2 rounds of some Crysis demo on an open air bench, you should strongly suspect that actual respectable overclocks on the very same chips would be lower because generally accepted tests of stability are a higher standard.

When it came time for the independent review sites to overclock, none of them claimed stability on the basis of gaming alone, but used the much more robust (and time-efficient) Prime95 test. And the results came out in the 3.5-3.9 GHz range for air, quite a ways off from your hopeful 3.8-4.2, and from the cherry picked "consistent 3.9" observation at the PII overclocking competition. In fact, of more than 30 leaked samples with reputable reviewers, none actually hit 4 GHz on air with Prime95 stability.

You made a good point that AMD does tend to improve the process over time. Precedent is a very strong argument. People and companies have a nasty habit of repeating what they usually do. Even Hector only changed seats; AMD is being run by the same people. AMD's initial 65nm was an unusually poor showing - a fluke early release; this 45nm is looking more like the typical 90/130nm that the later-65nm node only approached in maturity. This typical level of maturity ends up close to Intel's but is chronologically behind by a bit over a year and brought down to within a year in node-to-node milestone comparison by earlier releases.

You also deserve credit for not believing the AMD-released slides showing PII well ahead of higher Yorkfields in gaming and various other tasks. Again, their viral marketing has a reputation for cherry picking and partial disclosure, something well publicized by the PI debacle.

As for the performance guesstimates, the reason the previews were spot-on (in between Kentsfield and Yorkfield) is that it's hard to influence performance when the whole CPU, near its final stepping, is subjected to independent benchmarking. You never see dramatic performance changes right after the very last ES stepping because the #1 priority before a retail release for any reputable CPU manufacturer is to iron out all the critical bugs. Those do far more damage than minor comparative benchmark changes that can be compensated by small price adjustments.

singing the praises of a cpu that almost outdoes its predessessor as much as a whole new arch which was already ahead does.

This is confusing to read.

When people say the Phenom 65nm was architecturally ahead of its time or process node, that's a bad design comment. Good processor design allows a chip to be manufactured equally well on the current and the next future node (which is hopefully well under way), and it also keeps optimum balance between performance features and degradation among the two process nodes. When you overdesign, the chip ends up weak at the current process node, strong on the next, and weak again on the one after that (because you simply don't know the process details that far in advance). This makes the architecture shine for only one process generation instead of two.

From a design standpoint, it would be easiest to architect for one process node only. However, this would mean changing both the architecture and process node simultaneously, which would be a debugging nightmare from a production standpoint, or recycling a mature process, which would put your product a year or more behind the best that you could achieve with overlapping architecture and process.

Because the Phenom was especially weak at 65nm from overdesign, it is only natural that at 45nm, for which it was also designed, it would be comparatively much stronger. But this also bodes poorly for 32nm. The Phenom could not have been designed accurately for 32nm, so they'll be forced to change it, and even then it won't be optimal, perhaps resembling the Brisbane at 65nm over 90nm, or the Athlon XP at 130nm over 180nm.

This is a standadised score based on numerous reviews, with a PII 940 being the mid point at 100. Just going by points divided by clockspeed:

1) Q9550: 104 / 2830MHz = 0.0367
2) Q6600: 86 / 2400MHz = 0.0358
3) PII 920: 95 / 2800MHz = 0.0339
4) PII 940: 100 / 3000MHz = 0.333

I wouldn't divide benchmark index scores by clock speed. Benchmarks scale very imperfectly. These are certainly not all pure core tests. Most the reviews that adjust clocks for comparison do find the PII averaging right between the Kentsfield and Yorkfield - that is pretty precise because there isn't much performance difference between Conroe and Penryn.

 

[Just_An_Engineer]

In fact, of more than 30 leaked samples with reputable reviewers, none actually hit 4 GHz on air with Prime95 stability.

I'd suggest waiting a couple of weeks before making any definitive judgments on the overclockability of the Phenom II's. You may remember that when the Phenom I's first came out none of the reviewers were able to get more than 100 - 200 MHz overclock on them and they were immediately dubbed as "unoverclockable". However, only a couple of weeks later some reviewers were getting 300 - 400 MHz out of them (still not very good but a heck of a lot better) after figuring out some of the quirks of the chips. (this was before the advent of SB750 and ACC)

Given that the Phenom II's seem to be a different beast entirely from the Phenom I's I'd suspect that it might just take reviewers having a little more time to play with them and figure out the quirks before we see some better overclocks. I'd also note that several of the reviewers were reporting that they couldn't get the AMD Overdrive utility to work properly with the Phenom II's. This utility was listed as being quite helpful in overclocking the Phenom I's so perhaps an updated version with better Phenom II support might help things along as well.

 

[epsilon84]

I wouldn't divide benchmark index scores by clock speed. Benchmarks scale very imperfectly. These are certainly not all pure core tests. Most the reviews that adjust clocks for comparison do find the PII averaging right between the Kentsfield and Yorkfield - that is pretty precise because there isn't much performance difference between Conroe and Penryn.

Fair point - lets leave Kentsfield out of the equation for a second then and focus on Yorkfield vs PII in terms of percentages:

Q9550 @ 2.83GHz = 104
PII 940 @ 3.00GHz = 100
PII 920 @ 2.80GHz = 95

Thats a ~9% advantage to Yorkfield in general. Now its generally accepted that Yorkfield is ~5% faster per clock than Kentsfield. Thus Kentsfield is still slightly ahead of PII in terms of IPC.

 

[WR]

A few % difference in performance can be attributed to the mix of benchmarks used. I mean, ideally you do not want to include synthetics (Sandra/Everest) because the extent of meaningful real performance improvement is already found in non-synthetics. And you don't want to include GPU-limited gaming because that artificially evens out the scores without telling you anything you don't already know.

If you include the control with a similarly clocked Kentsfield, I think you'd find in that benchmark mix the Yorkfield comes out more than 5% better. The synthetics were sampled overly often because they are easy for underfunded review sites to run.

 

[epsilon84]

Of course it depends on the mix of benchmarks, there is no universally accepted way to 'measure' IPC so the best we can do is get an aggregate from a number of reviews.

As for Yorkfield vs Kentsfield, in most reviews the margin invariably hovers around the 5% mark, for example:

http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3069&p=3

Of course a few % differences means very little in the real world and I would happily stick with PII if I had a compatible motherboard. Its a marked improvement over PI which is pleasing for existing AMD users looking for an upgrade.

If I were to buy a new system though I'd think very carefully about investing in PII at this stage as Yorkfield price cuts are around the corner according to Anandtech, plus the Q6600 is still around and at ~3.6GHz it is still a match for an overclocked PII.

 

[Reynod]

Got one for the quads Epsilon ?

Nice little graph there ... good post.

I feel the PII is about on par with the Kenty too ... seems fair.

Plus now that the Phenoms overclock well this will be a nice upgrade for AM2+ owners ...

I am looking at whether to get an i7 system next but it doesn't seem enough of an upgrade from an overclocked Kenty for me.

 

[fazers_on_stun]

You forgot Barron.

BaronMatrix, at least in his later incarnation, had toned the ad hominem attacks way down and mostly stuck to the facts - at least the facts that favored AMD . The others I mentioned added little content value to the discussion and mostly just flamed. I believe all of those got perma-banned eventually.

Then you have the fly-by posters like Thunderman (and previously Sharikou), who make little sense and are so ridiculous they have entertainment value .

Of course the Intel side has a similar range of posters but I'll leave it to the AMD fanbois to point them out.

So in short, JDJ adds content to the discussion - some of us Intel-favoring folks may not like what he posts but then that's what a discussion is supposed to be - exchange of ideas, yadda yadda.

 

[fazers_on_stun]

No really dont get me wrong im not saying that cos i think hardocp is 'teh suck' cos it dont support what im saying, it just really strikes me as a very odd/poor gaming performance benchmark. 1 title with tri-sli?!? wtf??? when does that tell you anything??? If u dont think its an odd way to guage gaming performance, well, hmmmmmmmm...

your not dissing 'the spot' are you?

Some sites are still wedded to some old benchmarks, and techspot does lean towards the oc community a bit, certainly more than tom's does. Its a dual english/spanish site, so maybe the spanish/south american enthusiasts are into the superpi thang. Anyway its a good site dont be a snob.

I'm not defending [H] - I'd really like to see some heavy-duty investigation as to why i7 does spectacularly well in some games on multi-GPU (well, OK, mainly SLI). Something interesting going on there, clearly. So yes the review could have been much better, but then you could say that about almost any review.

As for Techspot, frankly it seems behind the times and amateurish - not even a drop-down navigation menu to skip over the BS and get to the bench results . I've already seen the mobos and the chip die plenty of other places, sheesh.

 

[fazers_on_stun]

I'd suggest waiting a couple of weeks before making any definitive judgments on the overclockability of the Phenom II's. You may remember that when the Phenom I's first came out none of the reviewers were able to get more than 100 - 200 MHz overclock on them and they were immediately dubbed as "unoverclockable". However, only a couple of weeks later some reviewers were getting 300 - 400 MHz out of them (still not very good but a heck of a lot better) after figuring out some of the quirks of the chips. (this was before the advent of SB750 and ACC)

Given that the Phenom II's seem to be a different beast entirely from the Phenom I's I'd suspect that it might just take reviewers having a little more time to play with them and figure out the quirks before we see some better overclocks. I'd also note that several of the reviewers were reporting that they couldn't get the AMD Overdrive utility to work properly with the Phenom II's. This utility was listed as being quite helpful in overclocking the Phenom I's so perhaps an updated version with better Phenom II support might help things along as well.

Okay - how does this square with the "easy" overclocks that AMD was spouting at their demoes? Easy for whom - AMD engineers?

Frankly, I blame AMD for this PR mess, once again, even though it is much less of a mess than the P1 disaster. You'd think they would have learned by now. Obviously it's only the former ATI side that has learned that it's much better to keep quiet and then release a killer product, than to spout stuff that gets everybody excited and then under-deliver.

 

[JAYDEEJOHN]

Ive seen some great and truly honest responses here,TY. The benches are all over the place because as Epsilon said, theyre using tooo much DDR3, didnt speed values on the ram etc, and have P2 pretty much limited to 1 basic setup, tho there are exceptions there as well, which just confuses the overall picture as well. Im looking for a few other reviews and a few followups for the dust to settle so to speak.
My comment about last gen/die shrink goes like this. P2 was almost as good an improvement overall in whats available at stock as going from Penryn to i7, while ones a new arch, the other is but a die shrink. Now, the starting points arent the same, and Intel was already ahead, but for a simple die shrink a overall 25% increase at stock plus ipc etc is good to see, and if it were carried into the gaming segment for i7, I would be quite impressed and alot happier, but it doesnt do as well there like in other areas, where its awesome, but only so so in gaming, and thats why I was somewhat disappointed in it, tho, as Ive been saying, it may show progress yet in newer gen gpus. Im not talking multi card setups, as thats too exclusive, and fail, as scaling just doesnt get it done over costs for me.
As to the negatives for P2, where theres always someone, or alot of someones to bring those out, and you havnt seen me defending P2 too much regarding those things, as some are legit, and some are simple fanboy ravings.
To me, its good to have another potential cpu out, one with a different flavor of operation, different tweaking to do for ocing etc but most of all its a simple thing. AMD needs this, we need this. Im happy for yipsl and others, and wondering just whats going on over at "the zone" heheh right now. I dont go there, but Im sure Burger King is doing well there, as Im sure theres tons of whoppers about heheh

 

[JAYDEEJOHN]

Those easy ocees were done by major oceers, and they were given 30minutes, as was said in every article

 

[JAYDEEJOHN]

Think of it this way. How many of the "reviewers" were savvy enough, and had the time to do as just-a-engineer said? Also, having to learn the nuances of a chip can be fun and intriguing as well, it adds flavor, as opposed to just using sliders. Sorta like the difference between popping a game in a console vs a pc, and all its differences. Some may like this, some may not. To each his own

 

[Just_An_Engineer]

Okay - how does this square with the "easy" overclocks that AMD was spouting at their demoes? Easy for whom - AMD engineers?

Frankly, I blame AMD for this PR mess, once again, even though it is much less of a mess than the P1 disaster. You'd think they would have learned by now. Obviously it's only the former ATI side that has learned that it's much better to keep quiet and then release a killer product, than to spout stuff that gets everybody excited and then under-deliver.

I wouldn't exactly call it a PR mess as most of the reviewers seem to have been happy with the overclocking abilities of the processor and I doubt that many people will be complaining about being able to hit only 3.7-3.8 GHz on air.

I would still say that there is probably more headroom in the chips than has been seen so far as some of the reviewers stated that they were able to undervolt the processors fairly significantly and still have them be stable at the stock speed. This suggests to me that it shouldn't be necessary to throw as much voltage into the chips as many of the reviewers did in order to get decent overclocks. Like I said before, give it a few weeks and we will likely start hearing about people getting better overclocks just as we did with the Phenom I's.

 

[roofus]

to add to the maybe not as great as expected (while still good) overclocking results of the p2, initially not all mainstream sites did a great job with the i7 overclocking either. if one were inclined to see what end users are able to overclock theirs to, you see varied degrees of success and i am sure some are attributed to skill and some luck.
the results over at xtreme shouldn't be expectations outside of the people who do it all the time. if they have a personal bias, they aren't going to tell you they are running a phase change unit if they are intent on putting egg in someones face. its a stretch but some people are irrational when it comes to this whole AMD vs Intel crap.
don't too quickly dismiss TH or Anand's skills in overclocking. they probably have more experience in OC than most people who will buy the p2's which brings it closer to reality. most will go out of the box, on to the board and never get overclocked anyways. its nice to see what it can do in a drag race but overclocking is an art. not everyone who attempts it would be able to match the semi-pros results.

 

[fazers_on_stun]

I wouldn't exactly call it a PR mess as most of the reviewers seem to have been happy with the overclocking abilities of the processor and I doubt that many people will be complaining about being able to hit only 3.7-3.8 GHz on air.

I would still say that there is probably more headroom in the chips than has been seen so far as some of the reviewers stated that they were able to undervolt the processors fairly significantly and still have them be stable at the stock speed. This suggests to me that it shouldn't be necessary to throw as much voltage into the chips as many of the reviewers did in order to get decent overclocks. Like I said before, give it a few weeks and we will likely start hearing about people getting better overclocks just as we did with the Phenom I's.

I would guess those people expecting 4+GHz on air would be complaining if they only got up to 3.6GHz stable, as in the overclocked Intel vs. AMD review posted today here on Tom's.

My point was that AMD goes out with these demoes (as in hands-off, wierd location demoes with P1, and now some exotic LN2 oc demoes with P2) and raised everyone's expectations, and then the actual product comes out and underperforms. That's bad PR. AMD seems incapable of learning that simple fact. And as for the oc voltage, the Tom's review showed P2 as having very high voltages and power consumption exceeding that of the i7. All in all, the i7 oc'ed higher and easier than the P2, not to mention soundly thrashing P2 in just about all the benchies except for the GPU bottlenecked gaming tests.