Lonbjerg
Diamond Member
- Dec 6, 2009
- 4,419
- 0
- 0
- Status
It reminds me of the stupid MHz race we had years back
As this product is made for the server cost/benefit market; What we can hope for for desktop and mobile is a product that have good power profile down to a single module 2012, and that AMD can ramp production quickly so it will be cheap
As this product is made for the server cost/benefit market; What we can hope for for desktop and mobile is a product that have good power profile down to a single module 2012, and that AMD can ramp production quickly so it will be cheap
Fairly old quote. Is it still accurate. Also how does it compare the two? Is it comparing against the first MC parts that were released or the ones with clock speed bumps that have been released most recently?
We don't know what kind of clock speeds we're going to see either so we don't know how they compare on a clock per clock basis. If the IPC stays the same, but the clock rates jump to around 3 GHz there's going to be a sizable performance gain from that.
I can understand that they don't want to release any performance data. If it's really good they're going to see a big drop in the sales of their current offerings and either way it tips their hand to Intel.
You'll find no disagreement from me, but I will point out that this is where the burden of adding caveat upon caveat to one's statements falls to the individual (or business entity) making the statements in the first place.
Its really not our place to presume caveats on John's behalf, that rarely helps clarify the situation. If John wants to add caveats then he is free to do so. He asked a question, I answered it. Where the conversation goes from there is up to him.
Sadly it's probably not going to go anywhere if he even comments on it at all. I'm sure we'd all love for further confirmations of performance or a generous spilling of other details, but that's just not going to happen.
It doesn't help that the best information about have about the release is Q2 (which can be anywhere from about a month away to almost midyear) with nary a reputable leak in sight.
It doesn't help that the best information about have about the release is Q2 (which can be anywhere from about a month away to almost midyear) with nary a reputable leak in sight.
Idontcare, just doing some rough estimation based on the leaked AMD slide you linked to along with JFAMD's quote, it appears as though Bulldozer will be about 35-55% faster overall compared to the current Phenom II chips when comparing what each "core" is capable of.
I find it hard to believe, personally.
I find it hard to believe, personally.
The slide isn't leaked.
It is AMD's presentation to their analysts from Nov 2009, the link I gave is to the AMD server that hosts the presentation for public access.
Presumably the information presented there is accurate, accurate enough to be presented to investors as a matter of SEC concerns one would hope.
35-55% better than PhII is great. Could all be from higher clockspeeds though, don't assume everything is clock normalized.
It is AMD's presentation to their analysts from Nov 2009, the link I gave is to the AMD server that hosts the presentation for public access.
Presumably the information presented there is accurate, accurate enough to be presented to investors as a matter of SEC concerns one would hope.
35-55% better than PhII is great. Could all be from higher clockspeeds though, don't assume everything is clock normalized.
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
How young are you?
I meant dominant from a performance perspective, not market share. Clearly everyone except you understood that.
Why are you so concerned about my age?
edit: @mopetar we can infer a few things. 1. BD is not imminent or they would have been singing its praises to the high heavens after intel's sata difficulties. 2. 2q typically means "just before q3 starts". If they wanted to launch in mid-may we would have updated guidance most likely by this point, something like "early-mid" q2. Lack of news implies rather strongly that late june is the best (and possibly even most likely) case with a statistically significant chance that it will end up in q3 at some point, instead.
Last edited:
Lonbjerg
Diamond Member
- Dec 6, 2009
- 4,419
- 0
- 0
Because AMD had the lead once.(perfomance, AMD never lead in market share)..for about two years...the rest of the time (before and after) Intel has reigned supreme...and we are talking a lot of years.
And I often hear "younger" people talk like AMD and Intel have traded places a lot.
People who's first CPU usually was an Athlon and have no idea of wha happend before that and think it was the norm.
1) IPC of the bulldozers will be fine.
2) we know they ll come in 3.5Ghz+ topbin speeds, with 500mhz turbo core boost speeds.
So Im guessing.... take any 4 threaded Phenom, overclock it to 4ghz+, and run some 4threaded benchmark, and your probably not that far off from what the bulldozer (stock speeds) will be like in 4 threaded stuff.
stuff that can make use of ~16 threads, bulldozer will naturally be much much faster than the phenom IIs... or even the 12core magny cores. I believe JF-AMD said atleast +50% performance on the 12core magny cores, they currently have.
Its bound to be a big step up from the current Phenom II's... which is a good thing.
Last edited:
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
my first "cpu" was a commodore 64 that my parents got when I was in college. Intel dominated sales for a lot of years, but didn't amd actually take the lead when tbirds were out? And after that there seemed a lot of give and take for 7-8 years, with intel generally faster but amd highly competitive, before the athlon 3200 + came out and amd was mostly faster for around 4 years from 2003-2006? honestly I think that the past 5 years has been one of intel's strongest periods in relative performance ever, in fact probably #1 in consumer desktop space at least. Hopefully BD at least starts up a conversation that doesn't start with "from a value standpoint"...
Yeah, things fall out of context over time. That was almost 2 years ago. I would have to go back to my notes and lord knows where those are but I explained it in the past. the key on that quote is that it was a progression of 3 performance statements: what istanbul was, what Magny Cours would be over Istanbul and what Bulldozer would be over Magny Cours.
The problem (if you can call it that) was that the numbers we were tossing out for Magny Cours over Istanbul were actually low, so we go more upside on Magny Cours. That made the gap between the two larger, but Bulldozer was a completely different design. That meant that the actual delta to MC became smaller because MC came in higher than expected.
That is the problem with trying to make performance estimates keying off of a product that is not in final silicon- too many variables.
But the performance expectation holds from back then. I can't recall which site I did the math on (probably xtremesystems.org) but somewhere I have the math that shows the deltas, the difference in performance and how it all holds together.
If Istanbul, MC and BD were all in the same family and the same design, upside in MC probably would have pushed over to BD, but because they were different architectures, that doesn't really hold.
Spoiler alert - the 50% number was based on the 6176 SE, which was the fastest 12-core at the time the statement was made. We just released a 6180 SE, so you would need to subtract that upside from the 50% to keep consistent. It's a small amount, but some people nit pick.
Thanks John :thumbsup:
I figured you'd help us understand what we were looking at, a bit of the numbers behind the statements perspective.
I hope no one nit picks you over the information. It is generous of you to take the time to help us see the forest for the trees.
I figured you'd help us understand what we were looking at, a bit of the numbers behind the statements perspective.
I hope no one nit picks you over the information. It is generous of you to take the time to help us see the forest for the trees.
Irrespective of everything else, I'm just curious how well anyone can estimate performance several years out when it's unlikely that physical chips from the process you plan on using even exist. I'm sure there's good estimates based on veteran chip designer knowledge and if nothing else, you can always simulate things, but that doesn't taken into account the fabrication process and all of the little devils in those details.
AMD's chip designers obviously know what they're doing, or AMD wouldn't even be here, but as the saying goes: In theory everything should work out in practice, but in practice things rarely work out according to theory. If those developers thought they could build a better chip, they probably would have, but I'm interested in how this architecture has changed and evolved over time. It's probably late to the game, at least more so than AMD would have wanted, but how has the design evolved or improved over those periods of delay?
I know that's not something you can talk about right now, but it would be interesting to hear about months down the road. Bobcat certainly upped the ante so I'd like to believe that Bulldozer has just as much in store. Hopefully Anand will do a nice writeup on the development process as much as he does about the technical merits of the architecture.
Well, I could probably say that it is all about guard rails. If you have ever done a road trip, you know how to play the "GPS game". You get in the car and it says new orleans is 9.5 hours from Austin. But every hour you get closer to your destination, you hone in on a more accurate time and at 5 hours into the trip you are going to be there in 8.5 hours. Then, at 7 hours the trip should only take 8 hours.
It's kinda like that. Generally speaking, on a common architecture, the projections are close. To a scary degree.
On new architectures the guard rails drop down with every spin of the silicon as you narrow in on the target. Think of it as things coming into view.
It's kinda like that. Generally speaking, on a common architecture, the projections are close. To a scary degree.
On new architectures the guard rails drop down with every spin of the silicon as you narrow in on the target. Think of it as things coming into view.
Though it is possible that the author mixed that up we have no indication for this. Especially as the whole ISSCC presentation was about Bulldozer and it would be rather strange if they mixed Bobcat information in that presentation.
Anyone access/link to the ISSCC presentation slides?
However this 60% performance improvement would rather fit to other statements/rumors. E.g. the one of this thread: "Bulldozer 50% faster than Core i7 and Phenom II".
It is unclear if Turbo Mode adds something on Top or was included in the ISSCC statement, but it is possible that it adds something on Top. In that case a single BD core could again become faster than a Phenom core because of higher clock and then the performance statements (ISSCC, JFAMD) would be as well in sync.
Nemesis 1
Lifer
- Dec 30, 2006
- 11,366
- 2
- 0
I not sure . But it would seem that some people are not being honest here .
Ya I understand that AMD modules are 2 core 2 threads.
But whats not being said is Intels modules and thats what intel calls them are 2 cores 4 threads . Now AMD may or may not end up with more modules than intel on 1 die. But intels single module is 2 cores 4 threads. Ya can break it up any way ya want . But thats the way it is . so if we take a 2 module BD and go up against 2 module SB . this is what we have .
AMD 2 modules= 4 cores 4 threads . Intels 2modules= 4 cores 8 threads .
AMD may choose to go with more modules per die but than its not apples to apples.
Lets take SB 2 modules that we can buy right now . JFAMD seems to be saying that a 2 module BD will be faster than present SB. I don't believe it for a second. hype is what its called. Now I use the words hype. But after BD is released I will call it by what it is . Good info from JFAMD or pure out and out ----!
Ya I understand that AMD modules are 2 core 2 threads.
But whats not being said is Intels modules and thats what intel calls them are 2 cores 4 threads . Now AMD may or may not end up with more modules than intel on 1 die. But intels single module is 2 cores 4 threads. Ya can break it up any way ya want . But thats the way it is . so if we take a 2 module BD and go up against 2 module SB . this is what we have .
AMD 2 modules= 4 cores 4 threads . Intels 2modules= 4 cores 8 threads .
AMD may choose to go with more modules per die but than its not apples to apples.
Lets take SB 2 modules that we can buy right now . JFAMD seems to be saying that a 2 module BD will be faster than present SB. I don't believe it for a second. hype is what its called. Now I use the words hype. But after BD is released I will call it by what it is . Good info from JFAMD or pure out and out ----!
Let's just say that performance per socket will be the best way to measure performance between two competing architectures.
"Per core" metrics always have an agenda and are always biased. Intel will point to per core performance and I will point to per core power and per core price. But people don't buy cores, they buy processors. And that is why per socket is the only way to compare them.
"Per core" metrics always have an agenda and are always biased. Intel will point to per core performance and I will point to per core power and per core price. But people don't buy cores, they buy processors. And that is why per socket is the only way to compare them.
No that is not was JFAMD is saying. In fact JFAMD says nothing regarding SB.
From all we know we will see comparable products on thread count, means you have to compare a Bulldozer 8C/8T using CoreMultiThreading to get 8 threads with a Sandy Bridge 4C/8T using SynchronousMultiThreading to get 8 threads.
That is also compareable from the die sizes:
AMD Bulldozer 2C/2T = 18.0 mm² (on 32 nm process)
Intel Sandy Bridge 1C/2T = 18.4 mm² (on 32 nm process)
So you have to compare that not only fair but also how they will be sold. And about that JFAMD gave clear statements ("cores are marketed") and initial Zambezi 8C/8T with initial Sandy Bridge 4C/8T will both be the top consumer parts.
In addition some information about Intel's compiler with cripple AMD "function":
http://www.osnews.com/story/22683/Intel_Forced_to_Remove_Cripple_AMD_Function_from_Compiler_
Last edited:
So you completely ignore the actual architecture differences, apply your own logic and say everyone else is wrong...... , there is a clear difference is data flow between intels HT design and AMD CMT design.
based on your logic i now declare that the Indian ocean and the Mississippi river are the exact same thing, after all they both contain H20. lets just ignore the amount, the temp, the salinity, the acidity etc.
can both HT threads get scheduled/executed at the same time 100% of the time, no. can the CMT threads get scheduled/executed at the same time 100% of the time, yes. Does each HT thread get its own 128bit FP MUL and add, no, does each CMT thead get its own 128bit FP MUL and add, yes.
Then lets look at core die size, a SB core + 2mb L3 cache is almost the exact same size as a bulldozer module with 2mb of L2 cache. Now you are basically saying that intel needs 2 times the amount of die space to compete, ie two of AMDs thread take the same amount of space as two of intels threads.
AMD's design is about converging the front end to increace efficenty, Intels design is about single threaded thoughput and then having a mechanisum there to increase overall processor thoughput because otherwise there much wider core doesn't get very good utilisation. So in the end, Intels backend is bigger and AMD front end is bigger, they both come out around the same size.
( based on SOC the L3 cache will make a module bigger then a SB core + cache).
Your comparison is nothing close to an “apples to apples comparison. Even if bulldozer can't get the performance crown its just made every non HT intel CPU look alot worse in a performance per mm perspective.
Nemesis 1
Lifer
- Dec 30, 2006
- 11,366
- 2
- 0
Ya see John this is my problem . You want to discuss sockets. Which is fine . But Why sockets. Ya I assume a 8 module BD will have OK performance. But will it be faster than 8core SB. Or 6 core IB. How is AMD going to market its products based on socket or modules per die. If they go by socket I am all in on AMD bandwagon. I can't imagine turning down a $200 BD 4 module socketed PC.
No JFAMD . I won't be looking at sockets . It will be performance per module that determines the successs of BD and not its socket. By the way I want BD to be faster than apple to apple SB . I want AMD to raise its pricies . Intel won't lower theirs.
I grow tired of the bang for the buck nonsense that has infested hardware forums. Atom zacata tegra innovation Cell phones all sellable items. But LOL at performance . People are still saying that zacata is as fast as HD3000 which is 2x faster than zacata. I look at politics and hardware sites with the same basic view . They sold us out.
Look at Atom Oak trail a 3 watt part . When AT finely does a real article about it the final conclusion should be based on power usage and performance . I can't wait for the reviews myself. But the only thing I will buy is razors product when its released. That only because my grand kids will enjoy it.
Nemesis 1
Lifer
- Dec 30, 2006
- 11,366
- 2
- 0
I don't care about HT or CMT . You missed the whole point! 1 module AMD = 2 cores, 1 module SB = 2 cores. BP 1 module = 2 core 2 threads SB 1 module = 2 cores 4 threads
You say I am twisting to suite my end . LOL! I am using the same metric for BD and SB as whats been traditional in hardware forums . It is AMD who is hyping a new method of performance based on their end. That even fails . We have JFAMD saying BD in 2 nd qt 2011 for BD yet yesterday he said its hard to make guesstaments based on NON final silly. Its going to be enen harder for AMD to have BD in 2nd qt if they haven't final silly at this date. It also tells me that the lastest waffers sucked bigtime with the management repositioning going on at AMD.
Actually, the price/performance metric is a pretty important metric, and it's not used only on anand, but pretty much every forum of discussion that involves spending some of your hard earned money to buy stuff.
Nemesis 1
Lifer
- Dec 30, 2006
- 11,366
- 2
- 0
You did read the topic title did you not? Which is a joke to start with. BD is 50% faster than core i7 and PH II. I didn't know phII and i7 were = I thought i7 was 25%+ faster than PH II . Hell core II i7 slaps magna cores around thats only 2 intel modules using HT.
- Status
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter