Official Phenom Reviews Thread

[JumpingJack]

might be rev 2 of cpus that are ocable?

after all we saw the amd overdrive with quad on 3ghz. so how noone has that kind of cpu ?

These are the rev 2's, I think you mean the B3 stepping.

Now... Intel, AMD or IBM for that matter can dial in higher clocks within process if they want, the problem is that they suffer from a huge fall off in yield or you can cherry pick the best of the best and push up higher that way.

At the end of the line the performance of the body or population of CPUs is normally distributed, see for example this paper:

http://eda.ee.ucla.edu/EE201A-04Spring/GIT-PV.pdf

All the variation from the processing, i.e. within die, die to die, wafer to wafer, even process step to process step ... leads to a normal distribution of performance when the device is finished (Fmax is bell shaped), there is a finite probability of getting die at the very far tail of that distribtion, unforunately that means very few actual die so 'cherry' pick. Unfortunately, you get so few you cannot make enough to go to market but you can make a few to show around to various HW sites and IT journalists to fool people into thinking you have a 3.0 GHz processor.

The other way, the yield killing way, is to specifically tune your lithography to make the gate length narrower... this drives up drive current and results in a higher probability of getting a fast clocking CPU, unfortunately 90% of your die get hosed... so again, you cannot make enough to go to market but you can make enough to show a few HW sites and IT journalists and as such fool people (investors) into thinking you can make 3.0 GHz parts.

I think, in light of the errata, we now understand why AMD would not let anyone benchmark those 3.0 GHz processors ....

Jack

[hecktic]

lol @

But instead of sampling a Core 2 Quad Q9450, the upcoming Penryn replacement to the Q6600, and instead of even further dropping prices to completely ruin the Phenom launch party Intel responded in a way that actually doesn't make much sense: by sampling a $1000+ Extreme CPU, the Core 2 Extreme QX9770.

Thats code for smash AMD on the head

[JumpingJack]

lol @



Thats code for smash AMD on the head

Actually, it's sad ... because it is clear Intel doesn't need to do anything, they are still way in the lead ... this was just a stunt to make AMD feel even lower than they are

[gillll]

...........sad but true.

how did that happened ?

like it's not a small company - they have r&d all over the world .

can't be that intel produce on air 4ghz and above cpu and amd barley 3ghz !
we're their r&d asleep since the k8 intro ?

[awdrifter]

From seeing these results, AMD should've scrapped the native quad core and just MCM two Brisbanes instead of releasing this. The cost of production should be lower, and the performance wouldn't be too much worse, it might be even better because Brisbanes can hit 3ghz easily.

[Shintai]

...........sad but true.

how did that happened ?

like it's not a small company - they have r&d all over the world .

can't be that intel produce on air 4ghz and above cpu and amd barley 3ghz !
we're their r&d asleep since the k8 intro ?

I would say if I had to make an estimate:
1. They should have dropped SOI for bulk (Need to!).
2. Dropped the "native quad".
3. Never bought ATI (duh) so they could have invested in fabs and process,

But you also have to remember...Its still some 4-6x higher R&D in the other end. Its just a continual uphill battle. Its abit of a dream to think AMD can do this each time. With K8 Intel made mistake after mistake. And AMD made none basicly. Now Intel aint making any mistakes. And AMD falls over even the smallest rocks. And thats pretty bad when you are fighting goliath as a chimp.

I also think Hector got conned with ATI. Dave Orton knew were it was going. And he did the most brilliant deal ever for former ATI stockowners. Even the ATI division now is barely a shadow of its former glory. And clearly not a money maker in any terms.

[RPGWiZaRD]

Well all I can say when seeing all these Phenom reviews is that I did a great choise swapping from an AMD S939 setup to a C2D setup (which is my first Intel setup I've ever owned ) in August with the facts that are out now. I didn't have much faith in Barcelona/Phenom and since it was still K8 based I was expecting a tough task by AMD if they would be able to beat Core 2 Duo with only a further modified K8 core. I was expecting it would be able to match it at best so that even if I would go with Intel I had still had nothing to lose especially since my P35 supports upcoming Penryn 45nm as well. But that it's behind C2D this much and only about 20% faster clock for clock than K8 sure was a bit less than I had expected.

Better luck next round AMD. Hope you've aimed your performance goals a lot higher then with ground-up built architecture.

[gillll]

would say if I had to make an estimate:
1. They should have dropped SOI for bulk (Need to!).
2. Dropped the "native quad".
3. Never bought ATI (duh) so they could have invested in fabs and process,

sad but true.........again....the problem as i think is that amd r&d deals with near future architecture and intel has much further future r&d.

[KTE]

The problems are random lockups, in fact, it appears most or a good protion of the CPUs are not stable above 2.5 GHz, one could ask... why are they even launching at all??

The errata concerns problems relating to 1.225V 2.4GHz 9700, which is far below what was supposed to be released right now according to AMD (2.6GHz and even 2.8GHz FX which both were delayed too). That shows problems the day they didn't release 2.5GHz at Barcelona launch. Every company will try their best to release the fastest, no arguments, its a eP thing that wins most rounds. Some will remember, and I know Shintai is one of them who reported it so he will be aware, that Core 2 was hyped to release at 3.33GHz March 06. We didn't see this, not because they couldn't reach it, but no MFG releases products way outside the standard high TDP nowadays. By the time they could've, the marketing prices and demand was no where for 3.3GHz. Penryn top bin was stuck at 3.33GHz quad and 3.4GHz dual for over a year because of this problem again, but by July 07 things started to change for Penryn while K10h was still stuck below 2.3GHz in June 07. That much has always been quite clear.
Instability in 9700 isn't reproducible by us customers, only in AMD labs. For us customers, some of the reviewing guys have been able to get 3GHz stable out of the early B1, BA and B2 Phenom ES 2GHz/2.2GHz samples (not 9700), even the Taiwanese and Chinese early previews ran their 2GHz at 3GHz to simulate 3GHz Phenom, and so has Toms. HH/AT/FS weren't able to do this. But nevertheless, pretty much most of them made it quite clear "ES sample, we can't gurantee its representitive of the final retail". They could be worse or better in overclock headroom. Although it is obvious for a long time that K10h had problems with speed headroom and I stated it quite clear early on; the problem is TDP. 3GHz will probably be reachable on a 9700 but only with voltage, and not without massive power increase which is what AMD is trying to hide; I'm predicting that the ones around need ~1.5V for 3GHz. This is due to their manufacturing process choices, and exactly why they 'aint releasing em until binned much more for lower voltages/TDP. If a company can't make a product below the 130W TDP, it simply translates into process problems.

I don't have factual data on "yields" to comment.

The multi-core bandwidth is very low and latency is very high, worse than an E6750 at 2.4GHz, the memory bandwidth is also lower than a 2.4GHz E6750 at 824 5-6-6-19. The Sandra Arithmetic is again very low, even compared to many dual cores. I don't really know what to make of that yet until I get a retail myself but the tests indicate a v.bad product, worse than optimal expectations would deem, too much PR hype, too many lies and misleads and just failure in all processor departments concerned with a new product. I've already stated AMD reps mislead my uncle who is head at a major government enterprise, and that was all hype so he doesn't go the Penryn route which they knew he was after. I honestly had ordered X4 9700 from a store to test and find out myself and it was set to arrive 2 weeks from now. Many others had too. They will obviously be cancelled now and its a little too late for me to want to get Phenom to test now. For mid-January I made it quite clear: Q9450 please. Because it won't have an equal in price/perfromance/overclock for me.

Post#52 is as bad as it can get TBH.

[Morais]

Well, as I said in another thread, the only thing left to amd is the Quad Crossfire

[Boschwanza]

Is there a review that figured out differences between unganged and ganged memory performance mode?

[naokaji]

yes, in 3d rendering and media encoding the phenom looses against the Q6600... in gaming they are pretty much on par though.

so one has to actually think about what applications are used at all before deciding what to buy...

[JumpingJack]

yes, in 3d rendering and media encoding the phenom looses against the Q6600... in gaming they are pretty much on par though.

so one has to actually think about what applications are used at all before deciding what to buy...

Are we looking at the same data????
http://www.anandtech.com/cpuchipsets...spx?i=3153&p=9
I see nothing but blue at the top. WRT to Q6600, one could argue Crysis is a tie, this is the only site showing that for Crysis though.... Oblivion looks like a tie too.

http://www.hexus.net/content/item.ph...=10427&page=11
No wins there, not even even.

http://www.hothardware.com/articles/...RV670/?page=12

It is unfortunately, hands down, across the board behind Intel.

[Shintai]

in gaming they are pretty much on par though.

so one has to actually think about what applications are used at all before deciding what to buy...

Not really....The phenom can only keep up in 2-3 games, while it loses in the rest.
And those tests are absicly high latency DDR3 vs low latency DDR2.

You would have to buy it for a specific game aswell then.

[LinusTech]

Setting the memory to Ganged yields a slight performance drop in some tests, but significant improvements in others. It's harder to stabilize DDR2-1066 in Ganged mode, but it can be done.

I'd say Ganged yields 5% across the board if pressed to come up with a hard number. The Hardware Canucks review used ganged mode.

[accord99]

yes, in 3d rendering and media encoding the phenom looses against the Q6600... in gaming they are pretty much on par though.

Only if the game settings are pushed to be GPU limited. If not, or if the games are inherently more CPU dependent like RTSes or like Flight Simulator X, the power of the Q6600 shows up like here:

http://www.hardware.fr/articles/694-...enom-9600.html

so one has to actually think about what applications are used at all before deciding what to buy...

The Q6600 is pretty much a no-brainer vs the Phenom 9600 at this point. At worst, the Q6600 is a bit slower; at best it's much faster while using less power.

[JumpingJack]

Only if the game settings are pushed to be GPU limited. If not, or if the games are inherently more CPU dependent like RTSes or like Flight Simulator X, the power of the Q6600 shows up like here:

http://www.hardware.fr/articles/694-...enom-9600.html


The Q6600 is pretty much a no-brainer vs the Phenom 9600 at this point. At worst, the Q6600 is a bit slower; at best it's much faster while using less power.

Do you find this picture funny?

[Morais]

I was looking for an upgrade...Looks like the dark side is tempting me, but I already feel de 45nm prices hurting my wallet

[accord99]

Do you find this picture funny?

Yep, in hindsight I wonder what exactly was going through Henri's mind when he was posing (and smiling) for that picture. I'd imagine AMD by then would have an indication that Kentsfield was going to pose an incredible challenge.

[bedlamite]

I wonder if AMD will be able to survive next year...

[BeardyMan]

This won't stop amd from selling though
The HD series is selling like cakes, and this Phenom wil do the same.

i like this quote
World's First True Quad Core Desktop Processor:

[JumpingJack]

I wonder if AMD will be able to survive next year...

Yeah, the industry won't let it happen. Large OEMs will shove these into low end boxes and sell plenty, as a result AMD will continue with volume.

There are a few critical problems I see for AMD at the moment.

a) The die size is huge and the margins will be small as they must price these to be competitive, they did not regain any pricing leverage.

b) Yields, despite what AMD tells it's investors, are not great -- Barcey is still a no show, and the initial reports is that the availability of Phenom will be scant.

c) Clock speed.... clock speed.... clock speed. While AMD down played it's importance, it is actually still part of the performance equation (Performance = IPC*Clockspeed) and the poor launch speeds is really hurting, the must work to get these up to help boalster the ASPs.

[JumpingJack]

This won't stop amd from selling though
The HD series is selling like cakes, and this Phenom wil do the same.

i like this quote
World's First True Quad Core Desktop Processor:

It sure as heck fire will stop them from making any money, 2008 will be another year in the red for AMD.

[BeardyMan]

It sure as heck fire will stop them from making any money, 2008 will be another year in the red for AMD.

that's why i have to wait for another week because of the out of stock already

Edit*
On a side note is for 60% a paperlaunch as well...

[JVguest]

It's the cache. With only 4x512 meg of L2 cache that actually is fast (the L3 is slow) barc could never match c2q/york. If it had 8 meg L2, I think barc would be 10% faster clock for clock than conroe, and just slightly better than penryn (excepting special cases where the divide latency and shifter really helps penryn). Increasing the L3 won't help much. Since the roadmap doesn't call for increased L2 cache size and we know Barc won't clock well for 2008, AMD will remain well behind Intel performance for all of 2008 (at least) and with Nehalem coming out, all of 2009. The gap isn't large though, and AMD will be fine selling a little cheaper than Intel. Too bad the spent a lot of money going native QC.