AMD CTO Summit Editorial

FrgMstr

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,773
AMD CTO Summit Editorial - We are not yet back from AMD's CTO Technology Summit held this week in California, but it did make a big impression on us. Here are some thoughts that we want to share with our readers as to the direction of AMD.

After recently attending AMD's CTO Technology Summit I cam away with a much better understanding of its direction in the CPU market as well as the GPU market. While many of the specifics cannot be discussed in detail due to non-disclosure agreements in place we can certainly share our thoughts and feelings about what has been presented to us.
Click to expand...

Please Digg to share!
 
I cant wait , cpus and gpus both seem to be hitting a new high in advancement, mostly cpus anyways =)

i wonder how long it will be before we hit 16core chips
 
The editorial page for me seems to be a bit.... brief (ie, the link appears to be borked)?
 
Interesting how AMD and Intel are heading towards further integration of the CPU.

Kyle thinks the ATI logo will go away with R700 eh? So We might see the AMD RADEON 3900XT, for example?
 
Interesting article. Got me thinking...so it sounds to me like AMD is going to focus more on the mobile market that the desktop? Hmm...Im not entirely sure i like that concept. Sure, they say they wont abandon the high-end graphics market, but i have a feeling it could greatly impact the innovation needed to stay up with nVidia.

Time will tell.
 
jebo_4jc said:
I did not get that from the article.
Click to expand...

I was refering to this by Phil Hester, AMD CTO...

"The CPU/GPU merge model mentioned above simply does not cater to the high-end gaming market, but will likely be comparable to integrated sets of technology we see today. These models obviously appeal to the mobile markets and as AMD merges CPUs and GPUs, we expect to see them move towards a system-on-chip model where AMD will be supplying much of the silicon needs of mobile and business use systems. "

and apparently Bob Drebin, CTO of the graphics product group agrees.
 
Sorry for the brief nature of the article, I was a bit busy yesterday with the party we had in SF but I wanted to get something posted on this. Anand has a deep dive posted that I would suggest reading. We see some things differently, but that is what keeps it fun. :)
 
I'll start being impressed when I can encode a HD movie in 20 mins. I am betting that is several generations aways. Until then C2D should be more than enough for a long while.

EDIT: Hopefully it will really compete with intel though. I really hope AMD steps up.
 
The 'emotion' I have come away with is a depressing one,after reading Anands and Kyle's report.I wonder if Kyle will expand on his impressions of the meet ? or if Anands final words are what we would expect to see in an expanded editorial from Kyle ?

Talk of morale being down,and AMD still being so secretive and closed off to the press,is troublesome at best.Showing off a 45nm SRAM wafer that is a year old,is not at all something they should be proud of.So basically the whole company is being run off someones 'visions' for further mass integration at the cpu level.

The last thing I want as a gamer is more integrated graphics,not to mention how far off it all is (Fusion),this assuming that AMD executes 'flawlessly' as Kyle and Anand told us they would have to do,or what ? What if they dont 'execute flawlessly' over the next two quarters ? Then what ? And what is after Barcelona ? what have they as a back up plan? nevermind what architectures are layed in for the cut come a year or 4 after it ?

I hope the dreams and visions of Phil Hestor make good,and that they do not,turn out to be nightmares instead,financial or otherwise.
 
Digital Viper-X- said:
I cant wait , cpus and gpus both seem to be hitting a new high in advancement, mostly cpus anyways =)

i wonder how long it will be before we hit 16core chips
Click to expand...

16 [identical primary] cores? probably never. There was a great quote in there about "homogenous cores [are not the future]". We'll probably see a cross between Cell and a multicore-X86. Probably 4-8 standard CPUs, attached to an array of cores that are optimized for a lot of threads. If intel gets its way, the array will probably execute x86-SSEn. AMD will probably do the same for safety's sake, but something more Cellish/DX10ish is also a possibility.

The thing to watch here is the application where somebody chained 3 Cells together and got them to produce HDTV(720p) realtime raytracing. If you pull that off (affordable silicon, and full resolution 60fps) you can essentially stop producing GPUs (which is likely a goal of intel's, and seems to be their stradegy since MMX). If a console does this, they can likely get every studio in the world to make games for them, since they won't have to pay for writing all that shader code.

Wumpus

PS. Why all the demands for benchmarks? Kyle has explain what it's like to use the box. If you buy the thing largely for dicksize wars, you will have plenty of time to make your plans.
 
Wumpus said:
snip
Click to expand...

I disagree. Servers will always want more power and in the wide array of what constitutes what a "server" is it would be not be economically feasible to make specialized cores for each application. That said I think we will see many specialized cores come to be, but I also think that we will see a 16 general core cpu in the future. On a side note benchmarks are ALWAYS better than a description for two major reasons:
1) It's objective (assuming full disclosure)
2) "fast" is relative. When I got my 333mhz g3 imac back in the day, I though that was speed.

I find it reassuring that kyle thinks this thing is a beast, but until the NDAs are lifted, I'l hold my reservations.
 
jon67 said:
Having read the anand piece I repeat their request, Show Us The Fucking Benchmarks Bitch. NDAs or not, none of you has obviously seen any benchmarks and any feelings derived from watching realtime encoding just don't cut it.
Click to expand...
isn't the real-world performance, i.e. encoding, much more important than a benchmark score (which could be fudged/optimized)?
 
Darakian said:
2) "fast" is relative. When I got my 333mhz g3 imac back in the day, I though that was speed.
Click to expand...

Of course when Kyle said fast it is relative. He is probably saying it fast compared to what available now, not some obsolete hardware. It's good enough for me if Kyle meant it is faster than Intel QX series.
 
I want to see AMD's secret chips actually do something. I don't want to read about more feel-good PR smoke and daggers.

No offense to Kyle and others, but I think we all have our doubts about if AMD actually can find their way out of a paper bag at this point.
 
rahavsmt said:
Of course when Kyle said fast it is relative. He is probably saying it fast compared to what available now, not some obsolete hardware. It's good enough for me if Kyle meant it is faster than Intel QX series.
Click to expand...

I agree. Kyle knows what he's talking about in relative terms. But, numbers are always better :)
 
coolie_d said:
isn't the real-world performance, i.e. encoding, much more important than a benchmark score (which could be fudged/optimized)?
Click to expand...

An encoding benchmark would pretty much reflect the real-world encoding performance of a system, don't you think? The real-world experience is something different, a dimensionless entity that very much depends on the eyes (and mood) of the beholder. Experience is interesting of course, but I like to see some numbers first.
 
If AMD wants me back as a customer, then it's going to have to stop these constant socket changes.

If AMD's position is that every time they want me to upgrade my CPU, I have to go out an buy a new mobo too, then they can shove that position where the sun doesn't shine. :mad:
 
qdemn7 said:
If AMD wants me back as a customer, then it's going to have to stop these constant socket changes.

If AMD's position is that every time they want me to upgrade my CPU, I have to go out an buy a new mobo too, then they can shove that position where the sun doesn't shine. :mad:
Click to expand...

u have been here just as long as i have and i bet u can remember when p4 went though a bunch of socket changes too while socket A stayed the same for how many years?
 
even aside from the socket changes, intel has had the 865, 915, 925, 945, 955, 965, 975 chipsets for lga775. there were plenty of motherboard upgrades required; first when dual core came along, then for core2duo
 
Anyway, who's saying that a new mobo is required? They're only saying that you'll get better performance from a new mobo. In fact they're keeping their backwards compatibility, which is very nice.
 
Nice brief read. Anand's article does a much better job with information though.
 
mason.kramer said:
Anyway, who's saying that a new mobo is required? They're only saying that you'll get better performance from a new mobo. In fact they're keeping their backwards compatibility, which is very nice.
Click to expand...


Actually, they're really only saying if you want advanced per-core power (voltage) control a la cool'n'quiet, you need a new m/b. There's another thread in here which debates the value of more HT link bandwidth, with the general consensus being its not necessary for single socket systems.
 
mastercheeze said:
Actually, they're really only saying if you want advanced per-core power (voltage) control a la cool'n'quiet, you need a new m/b. There's another thread in here which debates the value of more HT link bandwidth, with the general consensus being its not necessary for single socket systems.
Click to expand...


The power savings and HT 3.0 are surely poised to serve the enterprise market, not the desktop. That is what is great about being able to put these in AM2 socket systems...on the desktop, it is just not going to be mean much if anything.
 
Jack Hammer said:
I want to see AMD's secret chips actually do something. I don't want to read about more feel-good PR smoke and daggers.

No offense to Kyle and others, but I think we all have our doubts about if AMD actually can find their way out of a paper bag at this point.
Click to expand...

QFT

I havent seen this much fluff since the phantom console marketing team. Seriously Im starting to wonder if AMD even knows what a benchmark is.

qdemn7 said:
If AMD wants me back as a customer, then it's going to have to stop these constant socket changes.

If AMD's position is that every time they want me to upgrade my CPU, I have to go out an buy a new mobo too, then they can shove that position where the sun doesn't shine. :mad:
Click to expand...

Because there isnt any other cpu company out there that makes its customers do the exact same thing....oh wait a second....
 
Jack Hammer said:
I want to see AMD's secret chips actually do something. I don't want to read about more feel-good PR smoke and daggers.

No offense to Kyle and others, but I think we all have our doubts about if AMD actually can find their way out of a paper bag at this point.
Click to expand...

I don't care about AMD's marketing: I don't care about what benchmarks are released early. Let's look at the Leaked C2 benchmarks. The vast majority of them were done with the 3.2GHz chip . . . that by the way we are still waiting for. Is this what you want AMD to do?!? That sure seems to be what you are implying. I'll take this a step further for you. The early release of information, while interesting, is nothing but advertisement. Personally, I don't care for advertisements. I don't care about the pretty pictures, big billboard adds and TV commercials spammed across all my channels. This advertising, and that is what it is, does not concern me, and to be honest, I don't think it should concern you either. I think my assertion that early benchmark data is nothing more than an advertisement also gets support from the fact that the chips can't even be purchased yet, and in the case of Intel's games, still cannot be bought! You want these games? You want them to do what Intel does and give "data" for things which cannot be delivered? Or, do you want AMD to give you accurate information as soon as they have it, say, a month before launch when they know final clocks and binning?

How about some more pointed questions: Will AMD's release of better benchmark data make the chip faster at launch? Will it bring in Barcelona's debut? Will AMD being more vocal about it's future somehow, magically, allow it to "find its way our of a paper bag"? Will AMD giving an updated roadmap deep into the future magically make all those things come true? Will verbosity grant them larger caches, higher clocks, or more cores?

The answer to all those questions is a big, fat, "NO!". So then, let's take another look at your post. You imply that you (and others) would have more faith in AMD if they were more vocal. You imply that people are losing faith in AMD with the lack of advertising benchmarks. You do realize that you are directly filling the role of the stereotypical "consumer" here, right? What we have just established is that AMD's verbosity effects nothing but your opinion of them. It seems to me, that all that we have established is that you think you are important, you think that AMD owes you something, and you are doing nothing other than vocalizing this fact.

You are making a logical fallacy here. If you want to use the lack of data to claim that AMD needs to be more vocal to get sales, you may have some merit to your argument. However, after reading the context of your post a number of times, it is clear you are talking about performance and product. I'll tell you what. If you want to choose a product based on how warn and fuzzy the company makes you feel, go for it. If you want to chose the company that makes you feel important by giving you pre-release performance numbers, go for it. I still maintain that pre-release benchmarks are nothing more than advertising, especially with Intel's 3.2GHz bait-and-switch.

I'm not falling for it, but you go right ahead.

I'll pass judgment on AMD when the summer has passed and the products are in hand. That is the form of advertising that raises my confidence: actual peformance of actual product. Pre-release benchmark data is worthless anyways, and I very much respect AMD for the way they choose the represent their company: when the only thing that can be given reliably is "feel-good PR smoke" that is what they do. When there is actual data that will represent actual chips, AMD will tell us all about it. As binning is usually not decided until last minute, I won't be surprised to get this information only a couple of days before launch.

Again, please stop falling for the marketing. Here's another perspective: How long have we known about K8L, quad-core, and the L3? At least 9 months, for some people who follow this stuff even more. How long do you think it has been in the works? At least 3 years or so. So, when AMD is quiet about their 3+ year project a couple of years in, this means, what? You seem to imply that if AMD doesn't babble on about their multi-year project incessantly it means they have a problem? How about it means we have already know about the damn thing for near a year and nothing has changed and it is all on schedule? You seems to be the nervous consumer that needs to constantly "stroked" by PR to keep a positive attitude...

IMHO, a company should be judged by what they DO, not what they say they are going to do.
 
visaris said:
I don't care about AMD's marketing: I don't care about what benchmarks are released early. Let's look at the Leaked C2 benchmarks. The vast majority of them were done with the 3.2GHz chip . . . that by the way we are still waiting for. Is this what you want AMD to do?!? That sure seems to be what you are implying. I'll take this a step further for you. The early release of information, while interesting, is nothing but advertisement. Personally, I don't care for advertisements. I don't care about the pretty pictures, big billboard adds and TV commercials spammed across all my channels. This advertising, and that is what it is, does not concern me, and to be honest, I don't think it should concern you either. I think my assertion that early benchmark data is nothing more than an advertisement also gets support from the fact that the chips can't even be purchased yet, and in the case of Intel's games, still cannot be bought! You want these games? You want them to do what Intel does and give "data" for things which cannot be delivered? Or, do you want AMD to give you accurate information as soon as they have it, say, a month before launch when they know final clocks and binning?

How about some more pointed questions: Will AMD's release of better benchmark data make the chip faster at launch? Will it bring in Barcelona's debut? Will AMD being more vocal about it's future somehow, magically, allow it to "find its way our of a paper bag"? Will AMD giving an updated roadmap deep into the future magically make all those things come true? Will verbosity grant them larger caches, higher clocks, or more cores?

The answer to all those questions is a big, fat, "NO!". So then, let's take another look at your post. You imply that you (and others) would have more faith in AMD if they were more vocal. You imply that people are losing faith in AMD with the lack of advertising benchmarks. You do realize that you are directly filling the role of the stereotypical "consumer" here, right? What we have just established is that AMD's verbosity effects nothing but your opinion of them. It seems to me, that all that we have established is that you think you are important, you think that AMD owes you something, and you are doing nothing other than vocalizing this fact.

You are making a logical fallacy here. If you want to use the lack of data to claim that AMD needs to be more vocal to get sales, you may have some merit to your argument. However, after reading the context of your post a number of times, it is clear you are talking about performance and product. I'll tell you what. If you want to choose a product based on how warn and fuzzy the company makes you feel, go for it. If you want to chose the company that makes you feel important by giving you pre-release performance numbers, go for it. I still maintain that pre-release benchmarks are nothing more than advertising, especially with Intel's 3.2GHz bait-and-switch.

I'm not falling for it, but you go right ahead.

I'll pass judgment on AMD when the summer has passed and the products are in hand. That is the form of advertising that raises my confidence: actual peformance of actual product. Pre-release benchmark data is worthless anyways, and I very much respect AMD for the way they choose the represent their company: when the only thing that can be given reliably is "feel-good PR smoke" that is what they do. When there is actual data that will represent actual chips, AMD will tell us all about it. As binning is usually not decided until last minute, I won't be surprised to get this information only a couple of days before launch.

Again, please stop falling for the marketing. Here's another perspective: How long have we known about K8L, quad-core, and the L3? At least 9 months, for some people who follow this stuff even more. How long do you think it has been in the works? At least 3 years or so. So, when AMD is quiet about their 3+ year project a couple of years in, this means, what? You seem to imply that if AMD doesn't babble on about their multi-year project incessantly it means they have a problem? How about it means we have already know about the damn thing for near a year and nothing has changed and it is all on schedule? You seems to be the nervous consumer that needs to constantly "stroked" by PR to keep a positive attitude...

IMHO, a company should be judged by what they DO, not what they say they are going to do.
Click to expand...

i agree with your 100%.
 
so, amd is going embeded system on a chip huh?

high powered cell phones and pdas
 
Martyr said:
so, amd is going embeded system on a chip huh?

high powered cell phones and pdas
Click to expand...

naw, i'd say cheap workstations for big business


intel and dell didn't get to where they are today selling systems with discrete graphics
 
You must log in or register to reply here.
Back
Top