Bulldozers first screens

[qcmadness]

More like 0% in desktop-specific applications, or even -1..3% because of architecture "tweaks" in order to accomodate more cores, the same way Thuban is a bit slower per clock than Deneb.

On server parts sure, Bulldozer will be very nice, up to 16 cores per socket in a decent TDP looks like a winner, but on desktop i doubt it will even get close to SB, unfortunately.

That comes with the use of power gating and TurboCore / TurboBoost.

[slaveondope]

We did a quick test on this, you can find it here, but i'm sure there are more out there.

That would be assuming BD's arch. is the same as denab/thuban, everything indicates its as far from anything x86 we've seen up to this point.

[Micutzu]

While I do agree completely, I was just trying to point out that more cores do not mean higher average performance on desktop CPU's due to the usage model.

That comes with the use of power gating and TurboCore / TurboBoost.

Turbo Core on Thuban, TurboBoost and power gating on Intel CPU's . (next gen AMD will also have power gating, but that's another story)

[slaveondope]

While I do agree completely, I was just trying to point out that more cores do not mean higher average performance on desktop CPU's due to the usage model.

Your completely right, people do need to take note of your results when we run into more pasting of cores to the same arch.

[Solus Corvus]

While I do agree completely, I was just trying to point out that more cores do not mean higher average performance on desktop CPU's due to the usage model.

Not everyone has the same usage model. Single core isn't everything, even on the desktop. Many people run multiple apps at once, it's called multitasking and has been around for a while.

It is quite likely that we won't have a simple X processor is better than Y processor scenario with BD. It hasn't been that way in the past and nobody should expect it now. Even A64 vs P4 there were plenty of specific applications where a P4 was the better choice.

People will have to choose which processor to buy based on what kind of apps they use and how they use them. If you run a single threaded app and you run only one app at a time, then SB might be the correct choice. If you run multithreaded apps, or multiple apps then BD might be the better choice. Either way, we can be sure that the reviews won't match any particular user's usage model very closely (except benchers'). You will have to use your brains to figure out how the numbers really apply to how you use your computer.

[NEOAethyr]

The only reason I think the newer cores are slower then each last gen core is because of the timings they use in the romsip.
They do this to make it so it clocks higher and stays stable.
But in reality they don't put much time into it, there's over 2x the performance still in there yet to be used but we can't access these timings because the rom that contains them is inside the cpu core it's self.

If they went back to using tables that someone like me could see and mod, or if they switched to a dynamic table that any user could adjust inside the bios would we see a dramatic increase in performance.

I know it's a fact because of the pathetic ram bandwith we get on all of the newer amd's.

And when a user like me signs up for the amd dev program, and asks about it by email to the dev's, the dev's are dumb founded.
So I gave up trying to find the info, and haven't tried since.
I'de gladly sign an nda if I could get the info I needed to make our systems faster and more stable.

AMD Service Notice:{ticketno:[....]}

From:
"TECH.SUPPORT@AMD.COM" <TECH.SUPPORT@AMD.COM>
Add sender to Contacts
To:
....@yahoo.com

Dear Customer,

Your service request : SR #{ticketno:[.....]} has been reviewed and updated.

Response and Service Request History:

It´s a very complex issue, so we recommend you to contact your technical center and require for the proper information, please access the website below.

http://support.amd.com/us/contacts/P...alSupport.aspx

Regards,

In order to update this service request, please respond, leaving the service request reference intact.

Best regards,

AMD Global Customer Care

__________________________________________________ ___________________________________________

This email is a direct result of your contact with AMD Global Customer Care and not part of a campaign. There is no need to unsubscribe to this email as you will only be contacted again if you directly request another service from AMD Global Customer Care.

The contents of this message are provided for informational purposes only. AMD makes no representation or warranties with respect to the accuracy of the contents of the information provided, and reserves the right to change such information at any time, with or without notice.

__________________________________________________ ___________________________________________

Original Text
From: ....@yahoo.com
To: TECH.SUPPORT@AMD.COM
CC:
Sent: 05/12/09 21:12:51
Subject: SIP, BUI, SROM

ello. I'm a programmer in asm, I dev bios'es and intergrated operating systems as a hobby. I need as much info as possible on the SIP protocol, BUI and SROM. I'm willing to sign a nda if need be, if I need to do that then ask me and I'll send my addressing info's via email. To give you a better idea of exactly what info I need: db 0x69, 0x41, 0x24, 0x00, 0x00, 0xED, 0x15, 0x19 ;11X db 0x69, 0x41, 0x25, 0x00, 0x00, 0xED, 0x15, 0x19 ;11.5X db 0x69, 0x41, 0x25, 0x00, 0x00, 0xED, 0x15, 0x19 ;12X db 0x69, 0x41, 0x26, 0x00, 0x00, 0xED, 0x15, 0x11 ;12.5X+ db 0x69, 0x41, 0x23, 0x00, 0x00, 0xFD, 0x26, 0x22 ;5X db 0x69, 0x41, 0x24, 0x00, 0x00, 0xED, 0x26, 0x21 ;5.5X db 0x69, 0x41, 0x25, 0x00, 0x00, 0xED, 0x26, 0x19 ;6X db 0x69,0x41, 0x26, 0x00, 0x00, 0xED, 0x26, 0x19 ;6.5X db 0x69, 0x41, 0x20, 0x00, 0x00, 0xED, 0x16, 0x19 ;7X db 0x69, 0x41, 0x21, 0x00, 0x00, 0xED, 0x16, 0x19 ;7.5X db 0x69, 0x41, 0x21, 0x00, 0x00, 0xED, 0x16, 0x19 ;8X db 0x69, 0x41, 0x22, 0x00, 0x00, 0xED, 0x16, 0x19 ;8.5X db 0x69, 0x41, 0x22, 0x00, 0x00, 0xED, 0x15, 0x19 ;9X db 0x69, 0x41, 0x23, 0x00, 0x00, 0xED, 0x15, 0x19 ;9.5X db 0x69, 0x41, 0x23, 0x00, 0x00, 0xED, 0x15, 0x19 ;10X db 0x69, 0x41, 0x24, 0x00, 0x00, 0xED, 0x15, 0x19 ;10.5X Those are the mulipliers, the actual sip. I have a hard time tweaking them to perfection. I can adjust the amount of sip tables. I'm having a hard time making then adjustable on the fly though... The main prob I have is when I adjust one multiplier in one fsb table, another fsb table needs adjustaing afterwards. It's a real pain. So I really need as much info as possible on the actual data within. Bit's and bytes, I need the mappings. I'm working with a nvidia nfii chipset that has it'sbui right next to each table. It's unlikely you'll know anything about that so no biggy. Also there is a few more things cpu wise that I wanna know. some systems, plus all newer systems the sip is no longer static in the bios image. I would like toknow how to read the default sip that's built into the cpu it's self. Also howto redo it on such system, an example in assembly would be great... If I could adjust it on the 790fx platforms I could increase bandwith quite a bit... Especially for those that run tCR of 2t. I go under the nick of NEO, and I make custom bios'es for diff board's once in a while. I don't disclose code usally, if I do it's small pieces to explain this or that to someone. Like I said before, I'm willing to sign an nda if need be. I have the ability to assemble my own bios from scratch for the most part. And to recover it, take measurements and etc. So any kind of examples, or docs or whatever, just know that it won't end up killing a system of mine lol, it's no prob . Pretty please .... Looking for maybe tech doc # 21902. At the very least... Anything that can actually explain the mapping of the sip table of the athlon -> the athlon 64. Thanks

[marten_larsson]

That review only prooves that clock per clock per core performance has gone down, not that the performance per clock has gone down. Why don't you have one bar with all cores activated to see performance? (I don't read romanian and am too lazy to get google to do a bad translation for me.)

[FlanK3r]

otherwise, IMC at Thuban is more stronger. Performance increase with 3 GHz NB at Thuban is really easy (u need only 1.27-1.35V at CPU-NB) and for rendering etc is very nice impact.

[eXa]

But it has less performance on the same clock as denebs cpu nb.

[Micutzu]

Not everyone has the same usage model. Single core isn't everything, even on the desktop. Many people run multiple apps at once, it's called multitasking and has been around for a while.
...

The term "usage model" generally defines the main type of applications most of the users run. If 90% of people use the PC for watching youtube videos, 40% to play games and 0.2% run on 3DStudioMAX and Sony Vegas at the same time (made up figures), the weight of multitasking performance in the usage model will be almost zero; Intel Clarkdale is the perfect example for this, while it wasn't fast not even at launch and it sucked in heavy multithreading (memory performance was bad), it was a great solution for the mainstream market because it focused on multimedia content and poorly threaded apps so it sold very well even though it was fighting the best AMD platform in the last 5 years.

Sure, some users need more than that, but that's why there are niche markets and products. For a CPU/platform to be a success, it needs to shine in the bulk applications the users will run, not to hunt rare scenarios where it beats the competition just to claim a victory.

[-Boris-]

The term "usage model" generally defines the main type of applications most of the users run. If 90% of people use the PC for watching youtube videos, 40% to play games and 0.2% run on 3DStudioMAX and Sony Vegas at the same time (made up figures), the weight of multitasking performance in the usage model will be almost zero; Intel Clarkdale is the perfect example for this, while it wasn't fast not even at launch and it sucked in heavy multithreading (memory performance was bad), it was a great solution for the mainstream market because it focused on multimedia content and poorly threaded apps so it sold very well even though it was fighting the best AMD platform in the last 5 years.

Sure, some users need more than that, but that's why there are niche markets and products. For a CPU/platform to be a success, it needs to shine in the bulk applications the users will run, not to hunt rare scenarios where it beats the competition just to claim a victory.

Bulldozer isn't made for people who are playing flash games and are watching youtube. Llano is aimed for them. Bulldozer is made for people who actually pushes their computer a bit. And you seem to be talking about the sales of the product. At this forum we don't care to much about sales, we care about perforamnce. Of course sales might be nice if we are wishing a company well, but it's hardly our main interests right?

Do YOU want a computer capable of more than Youtube and a few tabs at the same time?
Personally I run a minecraft server, have tons of browser windows with even more tabs, might be doing some multithredaded calculations that can take days, and maybe want to play GTA IV once in a while. I bring my quadcore to its knees on a daily basis.

EDIT: Can you name a few single threaded programs where you suspect Bulldozer to be to slow for the average users you are talking about? I don't know if there are any commonly used single threaded programs left where performance in modern CPUs actually makes a difference anymore. Theoretical but unnoticeable differences in Microsoft Office isn't really a deal breaker.

[FlanK3r]

But it has less performance on the same clock as denebs cpu nb.

I dont think so....yes, its slower in superpi 32M with the same stings (Chew* tested it), but not in normally mutlithread programs. R10 is not so good for comparsion, because every run with R10 u can get +300points up and sometimes down! R11.5 is more better optimalized.

[Solus Corvus]

The term "usage model" generally defines the main type of applications most of the users run. If 90% of people use the PC for watching youtube videos, 40% to play games and 0.2% run on 3DStudioMAX and Sony Vegas at the same time (made up figures), the weight of multitasking performance in the usage model will be almost zero;

Yes, those are made up figures. But working with that, I'd suggest that the 90% figure probably wouldn't even notice the difference between something like high-end SB or BD and something lower like llano/lowend SB. Most gaming is casual gaming, and CPUs are a wash there too.

I am simply pointing out that lower single core performance isn't necessarily the penultimate indicator of useful desktop performance. Who multitasks? Almost everyone, I'd wager. Just some more than others. Will more cores be more useful than more IPC/clocks? That depends on you. Unless you somehow fancy yourself as the perfectly average computer user (to whom BD doesn't matter anyway), how average users use their computers isn't really the point.

[Micutzu]

...Will more cores be more useful than more IPC/clocks? That depends on you.
...

Just think about it this way: IPC/clocks can be used by much more apps (and more common ones too) than more cores.

...

EDIT: Can you name a few single threaded programs where you suspect Bulldozer to be to slow for the average users you are talking about? I don't know if there are any commonly used single threaded programs left where performance in modern CPUs actually makes a difference anymore. Theoretical differences in Microsoft Office isn't really a deal breaker.

I sense you are trying (maybe uncounsciously) to rewrite what I said just to fit your optics better. "Games" turned into "flash games", "poorly threaded" turned into "single threaded" etc. Gaming performance will be a scenario that is pretty important for a CPU and wich would not benefit at all from more than 4 cores but scales ok with better thread performance.

[NEOAethyr]

I don't know of any games that need more then an x6 and the same goes for emulation.
But when it comes to video post processing, the more cores and more mhz the better.
I've seen a couple avi synth stuffs that cripples the x6.

[Solus Corvus]

Just think about it this way: IPC/clocks can be used by much more apps (and more common ones too) than more cores.

Which apps are those though? Many apps are single threaded simply because CPU performance isn't critical to their use. And the apps that can use more cores will get a more significant boost out of that than out of extra ipc/clocks.

I don't know of any games that need more then an x6 and the same goes for emulation.
But when it comes to video post processing, the more cores and more mhz the better.
I've seen a couple avi synth stuffs that cripples the x6.

This goes along with my point. Do you game? Do you encode? Do you do both? If you do both which do you do more and which matters most to you? Etc.

Arguments about market share and what average computer users are doing doesn't necessarily have anything to do with me or you as individuals.

[-Boris-]

I sense you are trying (maybe uncounsciously) to rewrite what I said just to fit your optics better. "Games" turned into "flash games", "poorly threaded" turned into "single threaded" etc. Gaming performance will be a scenario that is pretty important for a CPU and wich would not benefit at all from more than 4 cores but scales ok with better thread performance.

No, I didn't rewrite games to flash games. I added flash games to the youtube kind of people. I didn't mention gamers specifically since I didn't think they were essential to the discussion, I put them in the category of people who as I said "pushes their computer a bit". I thought you talked about the youtube guys mainly. And Bulldozer isn't aimed for them.

But we can break it down to these main groups.
1. Ordinary users, using Office, youtube, flash games and simple browsing and occasional games.
2. Gamers.
3. Advanced users.

Bulldozer isn't aimed for number 1, Llano is, and might be a much better option than SB here. So these people aren't a problem. For the guys at number 2 Bulldozer has the potential of boosting performance with its decoupled FPU. So even if the games won't use all 8 integer cores they might be boosted by all 8 FPUs. Combined with very aggressive turbo I don't think you should count Bulldozer out as an alterantive for these guys. For the guys that fit in number 3 Bulldozer seems to be a perfect fit. 8 cores performance when needed, or fewer boosted cores when the situation changes. And if you want fewer threads there will be quad and hexa core Bulldozers as well. Most likely with higher frequencies and much lower price. I don't doubt these will compete very well at their price point.

So I honestly don't see the problem here. For all we know BD will have more cores, higher IPC and might even have a decent frequency advantage over Phenom II.

[Motiv]

But we can break it down to these main groups.
1. Ordinary users, using Office, youtube, flash games and simple browsing and occasional games.
2. Gamers.
3. Advanced users.

Bulldozer isn't aimed for number 1

Sorry, you are very wrong here. Bulldozer is aimed at servers. If for any reason it's multicore strength is useful for desktop, then that is where it will be aimed on the desktop; but that's not the primary goal of bulldozer.

[Olivon]

Ket: srry, but my 1100T Thuban beating your 5 GHz 2500k in rendering (R11.5, 3DsMax,POV Ray, practical Cinema 4D)...Dont forgot at it. In render is stock Thuban very good comparsion too with 2600k (without R10, it is not good optimized for x-cores)

Your Thuban is raped by the 2600K ...

Have you really tried SB platform

[Micutzu]

I lost interest in this discussion when I got to the texbook marketing press release vocabulary:

...

8 cores performance when needed, or fewer boosted cores when the situation changes. And if you want fewer threads there will be quad and hexa core Bulldozers as well. Most likely with higher frequencies and much lower price.
...

Edit, @FlanK3r : Thuban is on par with Lynnfield in those apps and WAY behind Sandy Bridge.

[-Boris-]

Sorry, you are very wrong here. Bulldozer is aimed at servers. If for any reason it's multicore strength is useful for desktop, then that is where it will be aimed on the desktop; but that's not the primary goal of bulldozer.

We are talking Zambezi, you should know that from reading the thread. I don't think there will be much AM3+ servers out there.
Besides, Bulldozer is made with both desktop and server in mind. AMD don't have the resources or sales to develop chips made exclusively for servers. BD will be the base for all AMD desktop chips for maybe a decade, Llanos successor will be Bulldozer based. They can't make any tradeoffs in desktop performance. Of course AMD has designd a chip that will work good in both desktop and servers, but they still had desktop in mind when they were designing it. Or do you actually believe they would go with the K7-based processors another decade or that they have another completely new design around the corner? Was K8 server or desktop?

[saaya]

Which apps are those though? Many apps are single threaded simply because CPU performance isn't critical to their use. And the apps that can use more cores will get a more significant boost out of that than out of extra ipc/clocks.

you make it sound as if most apps were well threaded and its rare to have apps that dont scale with more cores :P

[AbortRetryFail?]

Change thread title to "Intel fanboys comment on first Bulldozer screenshots" please.

[-Boris-]

I lost interest in this discussion when I got to the texbook marketing press release vocabulary:

Your argument makes perfectly sense! I see you are right, BD is useless for desktop and has no selling point.

Or... Can you please stop attacking my choice of words and present some arguments for your point of view?
Could you point out where I am wrong? Your arguments against Bulldozer works perfectly fine at a "traditional" 8-core. You seem to think that just because it got 8 integer cores it isn't good for poorly threaded applications and only is an alternative when you are running lots of threads all day long. Do you say the same about 4C/8T i7s? The difference isn't all that big you know. We are not talking 8 complete cores dragging down frequency and low thread performance as you may think. We are talking 4 modules with beefed up FPUs and extra integers. With advanced energy saving and Turbo these few percent extra die space won't hold the cores back much in frequency in low threaded situations.
The whole point with Bulldozer is that it's flexible and with just little extra diespace can run twice the amount of threads. And in the same time can utilize almost everything when running few threads.

Is 4C/8T i7 just hunting rare scenarios? Or is it simply a good quad core with some extra die space made to make it more capable in the increasing amount of multithreded applications that is out there? Would your argumentations against Bulldozer sound the same if it was marketed as an quadcore with extended HT?

[Postmodum]

I love these discussions ... no one as the single clue about what BD is going to be ... and that creates such a frustration on users that you guys end just "shooting bullets in the dark" and being "agressive" with each others ... All this, because you are just speculating about something you are completely "on the dark" about how it will perform ...

I must say that you keep comparing Thuban's blablabla ... this is a new architecture ... and it's kept on the secret of the gods for some reason ( good, bad or just beucase they decided to keep it secret no matter how good or bad it is ) ...

So yes... you can continue to feed a bottomless discussion where no one has really good arguments because, at this point, there AREN'T any ...

Or you can really create a good discussion not based on SB vs BD ... with architecture analysis ... and theoretic arguments ...

Oh well ...