Is AMD mounting a successful comeback with Phenom II and others?

[Martimus]

Originally posted by: Idontcare
http://vr-zone.com/articles/am...ift/6273.html?doc=6273

VR-zone has a slide that appears to be an AMD generated slide breaking out the various contributions to a claim of 20% performance improvement in 3GHz AM3 PhII over a 2.6GHz PhI.

(and yes the 15% increase in GHz accounts for 12% in that 20% number)

If this is true then it kinda squashes those earlier hopes of there being a 35% increase in performance with 20% of it being attributable to IPC increases alone.

Actually it shows a 20% increase in performance for the AM2+ PhII over the 9950. The DDR3 boost isn't included in that increase.

These numbers are exactly the same as was reported elsewhere, so this looks like it may be it. (can't remember the place, but I did hear 3% IPC and 5% cache increase a few weeks ago - along with a 5% DDR3 increase) A 25% improvement from a Phenom 9950 with the AM3 Phenom 2 945, with DDR3-1333 memory. It looks very reasonable to me, so I would tend to believe it.

I still want to see official benchmarks though! I also want to see how it does against the Q9550 and Q9650 in applications that I would care about.

Looking at DDR3 prices now, it looks like it will cost me at least $150 for a good 4GB set. Hopefully that goes down by spring, or my chances of buying one of these will be pretty low, unless it beats the afforementioned Q9550 at a similar price with DDR2.

 

[Viditor]

Originally posted by: Phynaz

Originally posted by: Martimus

Originally posted by: Phynaz

Originally posted by: bradley
It's also comparatively a very expensive platform, a certain lowlight in these harsh economic times.

Compared to what?

I can get 20% - 40% better performance for spending 10% - 20% more money on a low end server.

Sounds just like what I need in harsh times.

I didn't realize the i7 servers were available for sale right now.

As available as Shanghai servers

Ummm...Dell, IBM, Microway, and HP all now offer Shanghai servers.

 

[piesquared]

Originally posted by: Ocguy31

Originally posted by: piesquared
lol Don't get you're hopes up Phynaz.

It's pretty serious. Even if enthusiasts pick up on PII because they like the AMD name instead of Yorkie, I dont know that they will pick up any real marketshare. AMD had a better chip for a while, and couldnt break the marketshare plateau. Them coming out and roughly equaling 1+ yr old Intel tech is not going to move mountains. They are getting a cash-infusion from the Arabs and dumping some assets. We'll see how H1 2009 goes.

Its bad news for everyone. Intel would love to charge 1.5X for every chip if there were no competition. AMD would do the same thing.

Again, AMD came very close to break even last quarter with chips widely regarded as trash. Phenom II is not trash. So, where does this leave us? This leaves us with Intel lowering their price on Core2, and selling very few Core i7 due to the higher cost during very bad market conditions. AMD has lived with lower margins for years, while Intel needs much higher margins in comparison.

And above all else, AMD has a >$1,000,000,000,000 fund in Dubai Inc. willing to back them. Evidently they feel confident in a future beyond mid to late Feb. lol
Even the most skeptic of skeptics should realize that an investor of that size has access to roadmaps and future products comming down the pipeline that very few have, and must like what they see.

 

[Idontcare]

Originally posted by: piesquared
Even the most skeptic of skeptics should realize that an investor of that size has access to roadmaps and future products comming down the pipeline that very few have, and must like what they see.

AMD knew of the R600 and bought ATI anyways, knowing of R770 as well of course. We do all get to be monday-morning CEO's, but the truth is as you write it.

There are those in the know, those who like to think they are in the know, and those who are clueless that "the know" even exists.

 

[Cookie Monster]

Originally posted by: Idontcare
... and those who are clueless that "the know" even exists.

You just made me spill my coffee.

:laugh:

 

[Viditor]

Since ATI based products have now been mentioned (I was a bit reluctant), let me also point out some stuff that should probably be in it's own thread...it's just that it does effect this one as well.

Firstly, Fudzilla have finally written a first class article (sorry, should have had you sit first...).
Fudzilla Article

The topic is AMD's new Catalyst drivers with Avivo support.

"The new Catalyst 8.12 drivers are available for download, but don?t expect significant performance increase compared to Catalyst 8.11. No magical code will help HD 4870 in beating GTX 260, but ATI included a sweet deal ? Avivo video converter that will do more good to end users than a couple of measly frames more. All HD 4800 and HD 4600 series owners can put it to good use, and all you have to do is install the latest Catalyst 8.12 drivers, as these include direct support for the aforementioned graphics cards.

AMD calls its technology ATI Stream, so those who own the mentioned cards can expect improved operation in apps that ATI Stream supports. CyberLink, ArcSoft Adobe and even Microsoft see this technology as a means to deliver better quality service. The catch is in integrating AMD?s compute abstraction layer (CAL) into Catalyst drivers, resulting in better software-hardware communication."

The reason that this is relevant in this thread is that it appears to nullify one of the biggest speed increases that i7 brings...namely transcoding video.
As the article shows, Avivo increases transcoding speed to 7.5x faster than CPU based transcoding.
It's also apparently quite easy to use as well...

Interesting article...

 

[JackyP]

Originally posted by: Idontcare

If this is true then it kinda squashes those earlier hopes of there being a 35% increase in performance with 20% of it being attributable to IPC increases alone.

Now that this thread has turned to a discussion about speculation and leaks - which I prefer anyway - I want to add that this recent rumour gives a lot credibility to the good ol' leaks from the far east. They showed 5% IPC gains on early ES months ago (so we could expected some minor improvement on them) and everything pointed towards sub 10% IPC gains for Desktop applications anyway.

Even the most optimistic optimists should realize that AMD is in a dire situation even if Deneb *and* Shanghai perform fairly well. They were lucky with the HD4xxx launch, because somehow they caught Nvidia off-guard, but now the table may turn. Add to that a weak economy (bad luck for AMD) and not even well selling CPUs are going to help.
In this situation with their current cost structure they'd probably need both CPUs & GPUs to perform fairly well or the CPUs to perform extraordinarily, which even if possible will take some time until they can ramp up their output. Where does this leave the ever so often promised break-even? Q3'09 or Q4'09?

With Nvidia's 55nm launch they may make up a lot of lost ground and squash AMD's ASPs and profits.
http://techreport.com/discussions.x/16041

Just wondering if the Arabian investors 'like to think they are in the know' or really are 'in the know'...

 

[JackyP]

doublepost

 

[Idontcare]

Originally posted by: Viditor
Since ATI based products have now been mentioned (I was a bit reluctant), let me also point out some stuff that should probably be in it's own thread...it's just that it does effect this one as well.

Well the Topic title does includes "and others" so I see it as being relevant to first-order

Originally posted by: Viditor
Firstly, Fudzilla have finally written a first class article (sorry, should have had you sit first...).
Fudzilla Article

The topic is AMD's new Catalyst drivers with Avivo support.
.
.
.
The reason that this is relevant in this thread is that it appears to nullify one of the biggest speed increases that i7 brings...namely transcoding video.
As the article shows, Avivo increases transcoding speed to 7.5x faster than CPU based transcoding.
It's also apparently quite easy to use as well...

Interesting article...

Thanks for linking that article, I would have never read it had you not. And you are right, that article actually contained data, and complete sentences! Progress.

My first thoughts were immediately "quality" what is the quality of the transcoding in relation to that of badaboom and the traditional x86-based transcoders. Going by equivalent bit-rate alone is not helpful as that is what distinguishes superior transcoders from the crappy/free/cheap ones.

Francois Piednoel (of self-proclaimed Nehalem inventor fame) made a REALLY big stink over the quality vs. processing time between badaboom and Yorkfield/Nehalem on aceshardware...and his arguments always rang consistent with my experiences so I put some weight behind it as being plausible (that badaboom is nearly worthless, and so too might be avivo).

I pay extra for TMPGenc because it's encoders are generally pretty darn good and produce excellent image quality at same bitrate as other encoders (windows movie maker for example). But TMPGenc is starting to dabble with CUDA (I have not tested it yet myself) so there must be a worthwhile manner in which to extract quality and performance from these GPU's in transcoding. Even if the speedup with avivo-based systems (dragon) comes in at a mere 2x over that of QX9770 once the quality is forced to be there then it would still be a huge step forward on battling with i7 systems.

Originally posted by: JackyP
Now that this thread has turned to a discussion about speculation and leaks - which I prefer anyway - I want to add that this recent rumour gives a lot credibility to the good ol' leaks from the far east.

I considered this thread to always be about speculation as PhII itself is not even released, plus the topic is based on speculating whether or not AMD is mounting a comeback. Speculate away unless the OP clarifies that doing so is offensive to them.

Originally posted by: JackyP
They showed 5% IPC gains on early ES months ago (so we could expected some minor improvement on them) and everything pointed towards sub 10% IPC gains for Desktop applications anyway.

Yeah there were a lot of threads here on the forums about this when it first hit the web and frankly the reason those discussions died down was because AMD themselves came out and made comments that Shanghai would be 35% faster than barcelona for the same TDP...which was later clarified to be a 20% increase on clockspeed (the iso-TDP part) and a 15% increase in IPC (which we didn't parse as cache vs core vs uncore vs ram, etc). This was shanghai/barcelona though, not deneb/agena.

What we are seeing here with this one slide, which again is just rumor slideware for all I can tell as only VR-zone has it, is that for the desktop you can basically cut these AMD numbers in half. We get ~10% increase in performance from the increase in GHz while staying at the same TDP and we get ~8% increase in IPC (all total, not DDR3, just DDR2).

The AM3 part looks to be intriguing as performance is expected to increase another 3-4% while TDP drops considerably (is this solely due to lower Vdimm? I'm at a loss here how the IMC can make a 30W TDP difference).

Originally posted by: JackyP
Even the most optimistic optimists should realize that AMD is in a dire situation even if Deneb *and* Shanghai perform fairly well.

The 45nm ramp timeline definitely paints this picture. Launching 45nm PhII in Jan doesn't mean 100% of CPU sales in Q1/09 will be K10.5 at higher gross margins. I think realistically AMD would be doing an awesome job to be at a 25% 45nm and 75% 65nm product mix for the CPU SKU's by the end of Q1. Which means Q1 financials will continue to be dominated by the gross margins on their 65nm SKUs. That is not good.

 

[Viditor]

Originally posted by: JackyP

Originally posted by: Idontcare

If this is true then it kinda squashes those earlier hopes of there being a 35% increase in performance with 20% of it being attributable to IPC increases alone.

Now that this thread has turned to a discussion about speculation and leaks - which I prefer anyway - I want to add that this recent rumour gives a lot credibility to the good ol' leaks from the far east. They showed 5% IPC gains on early ES months ago (so we could expected some minor improvement on them) and everything pointed towards sub 10% IPC gains for Desktop applications anyway.

Even the most optimistic optimists should realize that AMD is in a dire situation even if Deneb *and* Shanghai perform fairly well. They were lucky with the HD4xxx launch, because somehow they caught Nvidia off-guard, but now the table may turn. Add to that a weak economy (bad luck for AMD) and not even well selling CPUs are going to help.
In this situation with their current cost structure they'd probably need both CPUs & GPUs to perform fairly well or the CPUs to perform extraordinarily, which even if possible will take some time until they can ramp up their output. Where does this leave the ever so often promised break-even? Q3'09 or Q4'09?

With Nvidia's 55nm launch they may make up a lot of lost ground and squash AMD's ASPs and profits.
http://techreport.com/discussions.x/16041

Just wondering if the Arabian investors 'like to think they are in the know' or really are 'in the know'...

Ummm...just to clear up a few things Jacky
1. I think you meant the 38xx launch last year. The 48xx was just an extension of that success.
2. As to Nvidia's 55nm,
"The 55nm parts, internally code named GT206, are finally trickling out like we said they would, with no speed increases, and no power gains. What should have been a simple optical shrink is turning into a totally botched job, with the 'real' 55nm parts unlikely to come out until late January at the earliest following yet another spin""
Text
That will put Nvidia about 1.5 years behind ATI/AMD for 55nm (AMD shipped their first 55nm GPU last year in Sept/Oct).
Also, ATI will be on 40nm in mid 2009, and 32nm in 2010.
Text

The graphics business is certainly the least of their worries these days...

 

[Viditor]

Originally posted by: Idontcare

Thanks for linking that article, I would have never read it had you not. And you are right, that article actually contained data, and complete sentences! Progress.

My first thoughts were immediately "quality" what is the quality of the transcoding in relation to that of badaboom and the traditional x86-based transcoders. Going by equivalent bit-rate alone is not helpful as that is what distinguishes superior transcoders from the crappy/free/cheap ones.

Francois Piednoel (of self-proclaimed Nehalem inventor fame) made a REALLY big stink over the quality vs. processing time between badaboom and Yorkfield/Nehalem on aceshardware...and his arguments always rang consistent with my experiences so I put some weight behind it as being plausible (that badaboom is nearly worthless, and so too might be avivo).

I pay extra for TMPGenc because it's encoders are generally pretty darn good and produce excellent image quality at same bitrate as other encoders (windows movie maker for example). But TMPGenc is starting to dabble with CUDA (I have not tested it yet myself) so there must be a worthwhile manner in which to extract quality and performance from these GPU's in transcoding. Even if the speedup with avivo-based systems (dragon) comes in at a mere 2x over that of QX9770 once the quality is forced to be there then it would still be a huge step forward on battling with i7 systems.

I must admit that I've never compared the quality of the 2 (GPGPU vs x86 transcoding) by eye (I don't think there's any other way in this case, is there?). I think I may have to find a way to do just that (in my copius free time...sigh).
I haven't heard any firsthand problems of the quality issues you mention, but most of the encoders I and my compatriots use are proprietary anyway...

 

[JackyP]

Originally posted by: Viditor
1. I think you meant the 38xx launch last year. The 48xx was just an extension of that success.

Nah, just talking about 48xx, because only those enabled them to actually make a profit. I think Q3'08 was the first quarter AMD/ATI made a profit sinse the obnoxious R600. IIRC the earnings reports. When talking about AMD I'm alwys trying to think in terms of mid- and long-term survival i.e. margins, profits, overall architecture performance etc.

Personally I don't think that Nvidia will be able to magically beat them with their 55nm shrunk cards, but if they are even able to close to gap it could put AMD/ATI in the red once again, because competition + weak economy are particularly bad for GPU sales I'd assume.

What do you think about AMD dropping from the top 10 semi manufacturers? 2007 & 2008 were pretty bad for them, one could expect improvements in the second half of 2009, maybe.
http://i.cmpnet.com/eetimes/ee...2008/chart1_120108.gif

 

[Idontcare]

Originally posted by: JackyP
What do you think about AMD dropping from the top 10 semi manufacturers? 2007 & 2008 were pretty bad for them, one could expect improvements in the second half of 2009, maybe.
http://i.cmpnet.com/eetimes/ee...2008/chart1_120108.gif

Flat revenue from 2007 thru 2008 is pretty dam impressive to me actually. Nvidia fell more than AMD.

I see my alma mater, TI, fell 6% in the same timeframe while their competition (Qualcomm, Broadcom, ST) all substantially increased revenue. Now there's a story worth being concerned about if you are a TI employee or shareholder.

The story with AMD continues to be one of margins, not marketshare or revenue rankings. They lack the economies of scale necessary to generate profitable gross margins. This is a very common story in the memory segment, AMD though continues to be the only guy in MPU marketspace that appears financially indistinguishable from a flash/dram manufacturer. All the profit opportunity of an MPU player, all the realized profits of a memory player.

 

[Idontcare]

Originally posted by: Viditor

Originally posted by: Idontcare

Thanks for linking that article, I would have never read it had you not. And you are right, that article actually contained data, and complete sentences! Progress.

My first thoughts were immediately "quality" what is the quality of the transcoding in relation to that of badaboom and the traditional x86-based transcoders. Going by equivalent bit-rate alone is not helpful as that is what distinguishes superior transcoders from the crappy/free/cheap ones.

Francois Piednoel (of self-proclaimed Nehalem inventor fame) made a REALLY big stink over the quality vs. processing time between badaboom and Yorkfield/Nehalem on aceshardware...and his arguments always rang consistent with my experiences so I put some weight behind it as being plausible (that badaboom is nearly worthless, and so too might be avivo).

I pay extra for TMPGenc because it's encoders are generally pretty darn good and produce excellent image quality at same bitrate as other encoders (windows movie maker for example). But TMPGenc is starting to dabble with CUDA (I have not tested it yet myself) so there must be a worthwhile manner in which to extract quality and performance from these GPU's in transcoding. Even if the speedup with avivo-based systems (dragon) comes in at a mere 2x over that of QX9770 once the quality is forced to be there then it would still be a huge step forward on battling with i7 systems.

I must admit that I've never compared the quality of the 2 (GPGPU vs x86 transcoding) by eye (I don't think there's any other way in this case, is there?). I think I may have to find a way to do just that (in my copius free time...sigh).
I haven't heard any firsthand problems of the quality issues you mention, but most of the encoders I and my compatriots use are proprietary anyway...

Vditor you can see what I was referring to in Anand's latest article on the frontpage - http://www.anandtech.com/video/showdoc.aspx?i=3475&p=1.

As I have not personally used either the Ati or the NV transcode programs I cannot rule out user error in these reportings of low IQ after transcode, but there is a consistency to them in that people tend to report craptastic results when using their GPU to transcode (albeit they get those results super quick like )

 

[SickBeast]

In my experience, the quality output of Badaboom is the same as that of HandBrake.

Badaboom can encode a movie in 1 hour on my 8800GTS, whereas it takes 8-10 hours to do the same thing on my Opteron 165.

 

[Idontcare]

Originally posted by: SickBeast
In my experience, the quality output of Badaboom is the same as that of HandBrake.

Badaboom can encode a movie in 1 hour on my 8800GTS, whereas it takes 8-10 hours to do the same thing on my Opteron 165.

I have no experience with either Badaboom or HandBrake. My only reference points are TMPGenc (which is awesome) and Windows Movie Maker (which sucks).

Is Handbrake a good encoder/transcoder? If the IQ is good then getting the same IQ in 1/8 the time is nice, but if the IQ isn't so good then I really have no interest in how much time it saves me as I won't use the output product for anything.

(speaking only from my personal perspective and value on IQ, not expecting anyone else to value IQ above encoding speed, just elaborating on my current position on what I have heard about these GPU encoders to date)

 

[SickBeast]

Originally posted by: Idontcare

Originally posted by: SickBeast
In my experience, the quality output of Badaboom is the same as that of HandBrake.

Badaboom can encode a movie in 1 hour on my 8800GTS, whereas it takes 8-10 hours to do the same thing on my Opteron 165.

I have no experience with either Badaboom or HandBrake. My only reference points are TMPGenc (which is awesome) and Windows Movie Maker (which sucks).

Is Handbrake a good encoder/transcoder? If the IQ is good then getting the same IQ in 1/8 the time is nice, but if the IQ isn't so good then I really have no interest in how much time it saves me as I won't use the output product for anything.

(speaking only from my personal perspective and value on IQ, not expecting anyone else to value IQ above encoding speed, just elaborating on my current position on what I have heard about these GPU encoders to date)

I am unable to tell the difference between the source DVD and the h.264 backups produced by HandBrake and Badaboom. The quality output is excellent. The one thing Handbrake has going for it over Badaboom is that it can mux the original AC3 audio, whereas Badaboom forces you to use an AAC conversion which leads to quality loss.