AMD's Bulldozer and Bobcat Processors Preview @ [H]

Because we are targeting the server market, not the enthusiast market with Opteron. Overclocking is bad in the server market.
 
JF-AMD said:
Because we are targeting the server market, not the enthusiast market with Opteron. Overclocking is bad in the server market.
Click to expand...

Well ya, but if you were to grab a server CPU and throw it into a consumer motherboard, assuming compatible socket and BOIS, wouldn't the cpu overclock? While I doubt there would be overclocking features in a server board there certainly would be in a consumer board.

I've overclocked opterons and xeons no problem. As long as the motherboard supports the CPU it's really a trivial matter.
 
The problem with that theory is the Server/Workstation chips are on different sockets than the consumer sockets. The same may go for Intel in 2011 as well.
 
This is just pissing the fck out of me. I just recently upgraded to a AM3 board.
 
meh ill just wait and see if the extra spicy bulldozer chips are unlocked like the extra spicy magny-cours are. because i really don't want to switch to Intel just to get an overclockable dual socket system.


hey guess what ASIA911, thats your fault for not doing your research so theres nothing to be pissed about. deal with it. this is technology and it is ever changing. AM3 will still be relevant in the low end system market for another 2-3 years so there's absolutely no reason why you couldn't sell that board near the price you bought it at.
 
fullvietFX said:
I'm excited to jump back on the AMD ship. I hope it beats the i7 by a good 30%, or it's Intel again for me. :D
Click to expand...

As long as it's competitive with i7 I'm on board. Bulldozer doesn't need to whoop the competition, just match it's performance at a better price for it to be a buy for me.
 
Actually if there's so little diffrence beetween 9x0 series chipsets and 8x0 chipsets why AMD doesn't already sell AM3+ mobos ?
 
JF-AMD said:
AMD doesn't manufacture motherboards.
Click to expand...

Nice snide comeback, but you know what he meant. If this is the best we can get in terms of a presence from your company then just give up man. Do you think your boss will look kindly on your soiling the company reputation because you have a chip on your shoulder? Think about it because that's all this post shows.
 
Maybe the reason for the delay of AM3+ boards is not AMD but motherboard manufacturers. And considering that its 4 to 7 months before bulldozer arrives they have some time to create new motherboards.
 
Are the new AM3+ boards getting EFI? Maybe that is causing a delay, not to mention Bulldozer has not been completely spec'd out yet
 
The Pharaoh said:
Nice snide comeback, but you know what he meant. If this is the best we can get in terms of a presence from your company then just give up man. Do you think your boss will look kindly on your soiling the company reputation because you have a chip on your shoulder? Think about it because that's all this post shows.
Click to expand...

Well, his response was litterall. AMD no longer manufactures motherboards. No longer, because I remember a very beautiful 6+2 phase mobo (real phases, not the "8+2" fake stuff we have now that is more like a 4+1) mobo way back.

Besides, JF-AMD is only representative of server solutions :p


In terms of consumer mobos, and AMD partners, I'd guess the 9xx chipsets finally updated their GPU core at least, right? It might of broken the pin compatibility that was present since the 700 series mobos (I think 690, too, but I am not sure).
 
All I am trying to do is point out that we don't control what the motherboard makers do. Too many people think that we have some power over the board makers, but board makers run on their own schedules.
 
The Pharaoh said:
Nice snide comeback, but you know what he meant. If this is the best we can get in terms of a presence from your company then just give up man. Do you think your boss will look kindly on your soiling the company reputation because you have a chip on your shoulder? Think about it because that's all this post shows.
Click to expand...

I know, you know, that he knew, what the other meant.

But I did not take it as a chip on his shoulder, but as someone that has an ability to convey a sense of humour. Mind you, it is hard sometimes to see that when we are not face to face.

ASIA911 said:
This is just pissing the fck out of me. I just recently upgraded to a AM3 board.
Click to expand...

Dont be ASIA911, you got a great mobo\cpu for probably a great price. I know I am happy with my 1055t.

We are talking about sometime in about 6 months. That is an eternity in [H]ardware terms.
 
The Pharaoh said:
Nice snide comeback, but you know what he meant. If this is the best we can get in terms of a presence from your company then just give up man. Do you think your boss will look kindly on your soiling the company reputation because you have a chip on your shoulder? Think about it because that's all this post shows.
Click to expand...
He stated it as it is, simple and to the point. Did you want him to guesstimate answers?
 
JF-AMD said:
there are no consumer mortherboards with C32 or G34 sockets.
Click to expand...

JF-AMD said:
AMD doesn't manufacture motherboards.
Click to expand...

So if I read those 2 posts, then it is a fact that there are consumer mortherboards with C32 or G34 sockets.

But it is also true that you cannot be sure that a manufacturer will not produce a consumer mortherboard with C32 or G34 socket(s).

/dream on.

So potentially EVGA or more likely some other mobo manufacturer, could produce a consumer mobo with dual server cpu sockects that can OC, ala SR-2 ;)

/dream off
 
Some conjecture:

AMD could let manufacturers make consumer "G34" boards (no ECC ram support, re-branded as something else like Intel's 1366), even add a few circuits to prevent cross compatibility), and viola their own performance socket. Then they'd be free to "cripple" and re-brand chips almost directly off their server line to sell to enthusiasts. But, AMD decided the enthusiast segment is not worth pursuing (they're right to do this, it's mostly for bragging rights, we're too small in the market), and they're focusing on delivering mainstream value instead, CPUs that are good enough and GPUs that are good enough. Put two mediocre things together and you get Fusion, perfect for bang for the buck, but we enthusiasts starve.

As enthusiasts, how can we not be disappointed that AMD has given up on pursuing the crown. The recent HD6900 series showed this (much smaller than Fermi, cheaper, efficient, loser). Bulldozer's consumer side will feature 4 modules 8 threads in Q2, and will proudly claim the value market against Intel's socket 2011 in Q3 with 8 cores 16 threads.

The race is over my friends.
 
We've been down the "2P consumer board" path before and it didn't pan out.

I don't have the numbers off the top of my head because I am home right now, but here is basically how it works. Back when the world was single and dual core, there was a need for 4 threaded platforms because at the time, 4 threads was the sweet spot. You saw lots of 2P workstations (the majority) and there were 2 socket enthusiast boards.

However, at the time, the market for 2P enthusiast board was maybe a point or two.

Then the world got to 4 cores, and the market for 2P enthusiast boards went to .8%, then .4%. That was the last that I saw and that was before there were 6-core and soon you'll see 8-core. So that market isn't going to get larger.

The reality is that the 2P enthusiast board may have a "cool" factor, but it does not have a market presence that makes it profitable.

Our server division is not going to support this because it is a real small (and shrinking) market and it takes away resources from our core focus (servers.)

The client guys aren't going to take an Opteron and rebrand it as something else because it just isn't a large enough market to make it financially viable.

The math has been done over and over and there just isn't the case for us to address the consumer market with a server part.

As I said before, we don't tell motherboard makers what to do, and if one of them wanted to make a board, we wouldn't stop them. But we would not support overclocking on Opteron, so the whole thing really does not make sense. That is why there are no consumer boards with G34 and C32, and I would expect that there probably never will be.
 
I know, I know the math won't change in the future, and I think my post earlier mentions that it's not worth AMD's time (rather than simply not caring). Thanks for taking the time to answer, JF. :)

Hopefully in the future when AMD's value based strategy pays off, they're out of debt, and they're profitable enough to afford to waste a bit of cash, we can see a "for pride" performance race. Not before then, it's bad strategy.
 
Last edited:
In reality, the "for pride" things are never really worth doing. To bring a new product to market, even based on an existing die, is ~$3-5M in total resource cost. Doing something like that to sell a few thousand parts is never a good thing.

Plus, we have all of our engineering capacity pegged at the limit. Any time you add one more "pig into the snake" (old Compaq term), you have to not do something else. That is why it is difficult to justify. Even if we wanted to throw away the cash, what we DON'T do is going to be a larger impact.

If, for instance, spending $5M to make a product that will net you $3M in total revenue might seem like a $2M loss on paper. But if a typical program would net you $25M, for instance, off of that same $5M investment, the actual cost of the program is -$22M - the $2M you lose on doing it and the $20M in profit that you lose in scrapping a profitable program to pursue it.

You really don't want us pursuing those things because that just means that we have to charge you more for something else down the road to cover those losses.

You'd be better off with AMD investing that money to squeeze another 5% of performance out of the parts that 99.6% of the market buys than trying to figure out how to lose money selling to .4% of the market.
 
Dang, JF-AMD keeps hammering away at my happy place waiting for a dual socket C32 enthusiast board...

However he is absolutely correct about the opportunity cost to a company if they pursue a niche market that doesn't offer the same return as another investment opportunity.
 
brycejones said:
Dang, JF-AMD keeps hammering away at my happy place waiting for a dual socket C32 enthusiast board...

However he is absolutely correct about the opportunity cost to a company if they pursue a niche market that doesn't offer the same return as another investment opportunity.
Click to expand...

Well, I always questioned the place of C32, since it seems just like half a G34, lol :p So why not get a single G34 and have the future upgradeability? :D
 
JF-AMD said:
C32 has the same upgardeability as G34.
Click to expand...

I had a detailed post, but I felt I was losing the point, so here goes:
C32 is limited to 2 sockets.
G34 can go 4 sockets.

G34, so far, is using 2 ~C32 dies, so each G34 CPU is ~ 2 C32.

So why buy 2 C32, when getting one G34 will put you in the same ballpark?
 
Last edited:
People buy C32 because they need either a low cost or a lower power solution.

If your app doesn't scale beyond 12 threads, then you get a lower cost solution in C32.

As well, C32 can get you 2 procs with really low power.

We did the math and there is clearly a market for it. Mostly in the cloud world. Just got a PO in yesterday for a cloud deal' there were a lot of processors on that order. When the cloud guys buy servers, they do it by the pallet load, and they buy several pallets.
 
I just hope we see something at ces. Getting tired of waiting my current system is getting long in the tooth
 
I wonder is there anything major that would keep AMD from making a lower mainstream with 4 Bobcat cores and 160 to 320 SPs? A chip like that would seem to fit nicely between Zacate and Llano in both notebooks and desktops. I would give performance roughly equal to a midrange Athlon II and use far less power, since the main difference between the TDPs of Ontario and Zacate are clock speeds.

Not that I would want JF to divulge any information on unannounced products, it's just a thought that popped into my head after reading several reviews of Ontario/Zacate.

Edit: Also, when I read Anandtech's G34/Magny-Cours article, I noticed that MC technically has 2 I/O HT links, but one was disabled to save on the number of traces and lands; are the HT links configured as they are when the chip leaves the fab or packaging, or is it something that the motherboard configures?
 
Last edited:
JMccovery said:
I wonder is there anything major that would keep AMD from making a lower mainstream with 4 Bobcat cores and 160 to 320 SPs? A chip like that would seem to fit nicely between Zacate and Llano in both notebooks and desktops. I would give performance roughly equal to a midrange Athlon II and use far less power, since the main difference between the TDPs of Ontario and Zacate are clock speeds.

Not that I would want JF to divulge any information on unannounced products, it's just a thought that popped into my head after reading several reviews of Ontario/Zacate.
Click to expand...

Only real issue would be Socket and TDP. Unless if the existing setup can pass a video feed to the NB via HT :p

EDIT: what I jsut said above is unsubstanciated, and completely a fabrication. Cause the reality is, I have no clue about any AMD CPU products.
 
JMccovery said:
I wonder is there anything major that would keep AMD from making a lower mainstream with 4 Bobcat cores and 160 to 320 SPs? A chip like that would seem to fit nicely between Zacate and Llano in both notebooks and desktops. I would give performance roughly equal to a midrange Athlon II and use far less power, since the main difference between the TDPs of Ontario and Zacate are clock speeds.

Not that I would want JF to divulge any information on unannounced products, it's just a thought that popped into my head after reading several reviews of Ontario/Zacate.

Edit: Also, when I read Anandtech's G34/Magny-Cours article, I noticed that MC technically has 2 I/O HT links, but one was disabled to save on the number of traces and lands; are the HT links configured as they are when the chip leaves the fab or packaging, or is it something that the motherboard configures?
Click to expand...

I don't comment on client.

As to the Anandtech article, Each G34 processor has 3 HT links per die, so 6 total on the package. Four are external to attach to other CPUs or chipsets (allowing a fully connected 4P with 1 hop latencies). One HT link link connects the two dies. The last link is not activated, saving on TDP for the chip. There is no way to activate that in the MB and if you could, all it would do is push up power consumption without giving you any benefit at all.
 
jeremyshaw said:
Only real issue would be Socket and TDP. Unless if the existing setup can pass a video feed to the NB via HT :p

EDIT: what I jsut said above is unsubstanciated, and completely a fabrication. Cause the reality is, I have no clue about any AMD CPU products.
Click to expand...

Bulldozer has no integrated video in 2011.
 
JF-AMD said:
I don't comment on client.

As to the Anandtech article, Each G34 processor has 3 HT links per die, so 6 total on the package. Four are external to attach to other CPUs or chipsets (allowing a fully connected 4P with 1 hop latencies). One HT link link connects the two dies. The last link is not activated, saving on TDP for the chip. There is no way to activate that in the MB and if you could, all it would do is push up power consumption without giving you any benefit at all.
Click to expand...

Thanks for the info! So, technically, it's possible to use more than one chipset. That would be a boon for PCIe SSDs or GPUs for Folding/HPC purposes... 2 SR5690/70 chips split into multiple x16/x8 lanes...

I could make use of a 2S G34 system with 4 x16 or 8 x8 PCIe slots. :D
 
You must log in or register to reply here.
Back
Top