Fudzilla: Bulldozer performance figures are in

[exar333]

As usual , you re jumping so fast to conclusions that you systematically
miss the point...




http://semiaccurate.com/forums/showpost.php?p=132311&postcount=824

That is about as accurate as most of the benches we have seen. I guarantee AMD would never even release/ship a CPU until those issues were well handled. Not likely.

 

[Dresdenboy]

22nm Tri-Gate transistor will allow for 37% more performance at the same power envelope OR 50% lower power consumption at the same frequencies (according to Intel). IB is 8 months away.

37% faster at low voltages, good for low power handheld devices:
http://www.realworldtech.com/page.cfm?ArticleID=RWT050511195446
At higher voltages (as used for high performance processors), the reached speedup is 18%. Notice that the 32nm->planar 22nm shrink curve is still steeper than the Tri-Gate curve. So at voltages of 1.1V or higher they'll be rather close or even match. So esp. overclockers won't see anything of these 37%. What's good is that at the same performance power consumption might drop, which means for example 8 cores instead of 4. Further:

But it is important to realize that the performance gains Intel is citing are not simultaneous. Transistors will not get 37% faster, 50% more efficient and reduce leakage by 10X all at the same time; nor will entire chips see the same gains as the individual transistor level. Intel’s circuit designers will have to pick and choose how to use the newfound advantages throughout each chip to achieve the best overall results, given the product.

No we don't, but the deeper the architecture, the harder it is for it to scale. So if BD has high IPC (i.e., deep pipeline), it will overclock similar to SB. Also, given that they are both manufactured on 32nm, I seriously doubt BD will clock to 5.5-6.0ghz, not at least until the move to 28/22nm.

You see it the wrong way. A deeper pipeline (Northwood, Prescott, Power 6, etc.) means less work per pipeline stage, thus a shorter cycle time is enough to accomplish it (-> higher frequency). In some cases they also had a lower IPC, but that isn't necessarily so. Such CPUs reach higher clock frequencies per se. But providing a clean clock signal at extremely overclocked frequencies might pose a bigger problem than with architectures which started at a low clock.

 

[Riek]

If BD 1 core = 1st gen i7 in IPC, then a 3.6ghz 8 core FX-8150 = 8 Core 3.6ghz Core i5 760, correct? That would be an incredible processor if true for just $300!

But let's look at what you are really saying:

BD 1 core = IPC of 1st generation i5/i7 ==> So that means 2x the performance of an i5 760 @ 3.6ghz for $300 in 8 threaded app?

2500k/2600k is 15% faster on average over i5 760. So that means an i5 760 @ 3.6ghz ~ 3.13g i5 2500k

With the same IPC as a 1st generation i5/i7 AMD would have an 8 core BD 8150 series which is "equivalent" to 8 core 3.13g i5 2500k......and that's without 4.2ghz Turbo Boost on BD?????????? So with Turbo Mode, we would have an even faster CPU than an 8 core 3.13ghz i5 2500k. Ya OK!!!!!!!!!!!!!!!!!

Before you try to attack someone try to atleast gets your numbers and facts straigth will you?

first in the ideal situation with perfect scaling a 3.6GHz 8core would then be as fast as an 8core 3.6GHz nehelam (WITHOUT HT)
lets use your reculculations again:
= 8 *3.13GHz i5 or (-> 3.5GHz n case of i7 2600 without HT 7.1)
add HT -> ~30% -> 5.5 i7 2600 cores with HT. So a little below SB-E.

That is in the perfect unrealistic scenario

In real life you probably get 90% scaling from 4->8 cores at best. so you get in the middle between SB 2600 and SB-E 6core in highly threaded (mostly integer based) applications!!!! Given that everybody assumes an 8core BD performs better in those highly threaded applications than 2600.... would clearly support that number...

The 4.2 turbo is needed to get parity or close to parity on 1-2 threads and have an advantage in 3-4 threads (in case it has an ipc between C2D and nehelam).
You won't reach 4.2GHz above 4 threads... (max turbo only possible when half the modules are active)...

the other things Dresdenboy already handled.

 

[BlueBlazer]

As usual , you re jumping so fast to conclusions that you systematically
miss the point...

AMD is a business, and making money is part of that business (besides making chips). R&D needs big money also. How can AMD sell higher performing CPUs for less than the current CPUs? Especially when introductory prices should higher (not lower), unless AMD is desparate to claw back the server market share that it has been steadily losing ground. Current MagnyCours had been going head to head with Intel 6/8/10-core Xeon counterparts. What point did I miss? :hmm:

http://semiaccurate.com/forums/showpost.php?p=132311&postcount=824

I call that another "runaround". It can be transpired that the results are "a bit too low" because of the lower clocks (compared to the clock speeds published in the recent GigaByte leaks). So, if the "problem is not Bulldozer" then is it the AM3+ boards (which has already been released and sold to many) like what happened to the recent Intel chipset bug issue? Or is it the BIOS is not mature (due to testing with "crippleware ES chips")? Or could it be the 32nm SOI process? I like to see this "might be explained somehow". :\

Hmm.... what's that smell... is it a... paper launch?

You can see the ETA was November 1st, and this is a lowly clocked SKU. This is just the server market, by the way. But I remember JFAMD was saying both server and desktop parts will be launched together?

 

[bridito]

You can see the ETA was November 1st, and this is a lowly clocked SKU. This is just the server market, by the way. But I remember JFAMD was saying both server and desktop parts will be launched together?

If it's Nov. 1, then it was not "60-90 days" as on the famous slide but 130 days! Running right into SB-E within a few days! This one's gonna hurt!

 

[BlueBlazer]

If it's Nov. 1, then it was not "60-90 days" as on the famous slide but 130 days! Running right into SB-E within a few days! This one's gonna hurt!

It could be either hard launch (or paper launch). Like I said earlier, its very hard to comment on actual launch dates. All I can find are snippets here and there. For example >> AMD's First Bulldozer Chips Arriving Sept. 26.....

Advanced Micro Devices will release the first Opteron server processors to feature its new Bulldozer cores on Sept. 26, PCMag has learned. AMD's FX-series Bulldozer chips for high-end desktops are also set for release in either September or October, according to a report from DigiTimes.

AMD is eyeing the Sept. 26 release date for its new Opteron 4200 series, codenamed Valencia, and Opteron 6200 series, codenamed Interlagos, according to a source familiar with the company's product roadmap.

AMD roadmaps have been questionable. Here's another one from HardOCP Chief Editor.....

And I told you months ago that Q2 was going to be missed, then I told you it had been scheduled for early September, and now that has changed again.

Not sure anymore....

 

[LOL_Wut_Axel]

If it's Nov. 1, then it was not "60-90 days" as on the famous slide but 130 days! Running right into SB-E within a few days! This one's gonna hurt!

How exactly? Bulldozer competes with Sandy Bridge, not SB-E.

Since someone will undoubtedly mention CPU costs for the Quad-Core SB-Es, they use a much more expensive platform. Buying a Quad-Core SB-E is nonsensical, anyway.

 

[formulav8]

I just got an email from AMD that says: "hanging out in San Francisco the week of September 12th? Not finding anything interesting?
AMD to the rescue. We'll be making an historic announcement, and want you to be a part of it."

Also says: Be sure to arrive before 7:00pm to hear our big news first hand.

Its a developer email but I can't think of anything else that they would announce?

I will edit this if someone says its not being a BD announcement.

 

[Idontcare]

I just got an email from AMD that says: "hanging out in San Francisco the week of September 12th? Not finding anything interesting?
AMD to the rescue. We'll be making an historic announcement, and want you to be a part of it."

Also says: Be sure to arrive before 7:00pm to hear our big news first hand.

Its a developer email but I can't think of anything else that they would announce?

I will edit this if someone says its not being a BD announcement.

Can't imagine they'd be making a big deal of it if it wasn't good news to be shared. And I agree, given the timing, I think they are setting expectations that this will be a bulldozer related announcement.

 

[Terzo]

I'm still a bit skeptical from the [H] AMD event, although I suppose this is an announcement directly from AMD rather than an [H] publicity move. Although I wouldn't be surprised if it's about the server platform rather than desktop.

Still, I'm hopeful that they'll announce the rumored September 19th release date is for real.

Btw, the email says "week of September 12". Does that mean the announcement can come at any point during the week, or that it will be made on Monday? It mentions to "be sure to arrive before 7 pm" so I imagine they've mentioned the specific day.

 

[ctbaars]

formulav8's post makes little sense. I think he's yanking our chain for fun. So much of the email is left out, so cut up, it could mean anything. As Terzo says, week of, or on 12th? usa says week of 11th. outside usa says week of 12th

 

[formulav8]

formulav8's post makes little sense. I think he's yanking our chain for fun. So much of the email is left out, so cut up, it could mean anything. As Terzo says, week of, or on 12th? usa says week of 11th. outside usa says week of 12th

Huh? Why would I do that? You can look at the image below. :hmm:

I also said i'm not exactly sure bd is the big announcement. But I do agree with what idontcare said. Also its Semptember the 13th to be exact.

 

[AtenRa]

And I agree, given the timing, I think they are setting expectations that this will be a bulldozer related announcement.

no comment () :whiste:

 

[psolord]

Huh? Why would I do that? You can look at the image below. :hmm:

I also said i'm not exactly sure bd is the big announcement. But I do agree with what idontcare said. Also its Semptember the 13th to be exact.

I received the same email as well. Formula tells the truth. Let's hope for something good *sigh*.

 

[nonameo]

How exactly? Bulldozer competes with Sandy Bridge, not SB-E.

Since someone will undoubtedly mention CPU costs for the Quad-Core SB-Es, they use a much more expensive platform. Buying a Quad-Core SB-E is nonsensical, anyway.

Actually, I would contend that bulldozer competes with neither sandy or sb-e. I think that instead, AMD is trying to make itself a niche in the server market, much like they are trying to do with llano. What is the niche? lots of cores for less money.

However, in order for this to work AMD needs to be able to provide a processor that gives significant gains(compared to sandy bridge) in multi threaded applications. It would be nice if it could compete with SB-E at a lower price too.

Personally I think looking for a niche is a good thing for AMD. Intel can't pursue everything, and its not like AMD can compete 1:1 with intel in the same area.

At the same time, I'd expect intel to go for the most profitable areas anyway, so it's not like there's much left for AMD to "niche" into

 

[Edrick]

How exactly? Bulldozer competes with Sandy Bridge, not SB-E.

I love how you keep saying that. Low end SB-E will be around the same price, if not lower, than the high end BD. Just because the platform, in this case the MB, may cost about $50 more. However, you are getting more features on said MB outside of the CPU which accounts for the price increase. And I am quite certain there will be expensive BD MBs released to compete with x79.

 

[Khato]

Who knows exactly what the 'historic announcement' will be, but the timing isn't surprising in the least. AMD pretty much always attempts to do something or another to distract some attention from the Intel Developer Forum, which is the 13th-15th.

 

[LOL_Wut_Axel]

I love how you keep saying that. Low end SB-E will be around the same price, if not lower, than the high end BD. Just because the platform, in this case the MB, may cost about $50 more. However, you are getting more features on said MB outside of the CPU which accounts for the price increase. And I am quite certain there will be expensive BD MBs released to compete with x79.

Irrelevant. Platform costs for Sandy Bridge-E are higher. It's not just the motherboard, but also the memory and for most of the models the CPU as well.

Even if a Quad-Core SB-E model is a bit lower in price, it won't change the story at all. It will not be noticeably faster than the 2600K at the same clock speed.

 

[formulav8]

I received the same email as well. Formula tells the truth. Let's hope for something good *sigh*.

Thankyou for confirming.

 

[Edrick]

Irrelevant. Platform costs for Sandy Bridge-E are higher. It's not just the motherboard, but also the memory and for most of the models the CPU as well.

Even if a Quad-Core SB-E model is a bit lower in price, it won't change the story at all. It will not be noticeably faster than the 2600K at the same clock speed.

How can you say it is irrelevant? And how do you figure that 8GB RAM costs more on a SB-E platform? I have 8GB now on my lga1155 and I will simply move it to a SB-E platform when I buy one. So your arguement there is invalid.

And if I buy a SB-E MB that has 10 SATA 6Gbps ports, with 32 lanes of PCIe 3.0, 4 native USB 3 ports (maybe), quad channel support, and all the other bells and whistles, that costs me $50 more than a MB that does not have all that, how can you tell me that the platform is the reason? In any platform, there are always multiple tiers of MBs prodices at many price points.

I will bet you that I can find a BD MB that costs the same as a SB-E MB when they are both released. Then you last arguement will be invalid as well.

To me, it appears that you are going to simply pick an Intel market that BD will beat, and then just say that is the market it was intended, just so AMD will not lose. At least that is what I gather from the majority of your posts. I hope that AMD hits a home run with BD for the simple fact that competition is good for the consumer. Ever since the release of SB in Jan, this has been a real boring year for CPU releases thus far. I hope that changes. And I feel a lot rides on BD.

The only thing you are correct about is that the quad SB-E will not be that much faster than a 2600K. Maybe a few % points, but not much. But that may be all it needs to best BD. Who really knows now?

 

[Idontcare]

formulav8's post makes little sense. I think he's yanking our chain for fun. So much of the email is left out, so cut up, it could mean anything. As Terzo says, week of, or on 12th? usa says week of 11th. outside usa says week of 12th

says the dude with 2 posts...formulav8 isn't exactly "jerk's your chain" material, his cred is a few rather large steps above that.

Huh? Why would I do that? You can look at the image below. :hmm:

I also said i'm not exactly sure bd is the big announcement. But I do agree with what idontcare said. Also its Semptember the 13th to be exact.

Totally unrelated, I never understood the "an historic" grammar vernacular versus "a historic".

According to this, it is supposed to be "a", not "an", because the "h" in "historic" is not silent.

Exceptions

Use "an" before unsounded "h." Because the "h" hasn't any phonetic representation and has no audible sound, the sound that follows the article is a vowel; consequently, "an" is used.

• an honorable peace
• an honest error

Oh well, just caught my eye in the pic above.

 

[LOL_Wut_Axel]

*snip*

Quad-channel memory costs significantly more than dual-channel and brings with it no real-world performance improvements. Furthermore, you can get PCI-E 3.0 support on some LGA 1155 motherboards now, before SB-E is even available.

Unless you're getting a Six-core CPU, Sandy Bridge-E is completely useless.

Bulldozer is supposed to go against Sandy Bridge. If it loses, it loses.

 

[Terzo]

says the dude with 2 posts...formulav8 isn't exactly "jerk's your chain" material, his cred is a few rather large steps above that.

In all fairness he's been registered for two years, just hasn't posted until today. But yeah, I wasn't questioning the truth of formulav's post, just making some comments.

Totally unrelated, I never understood the "an historic" grammar vernacular versus "a historic".

According to this, it is supposed to be "a", not "an", because the "h" in "historic" is not silent.

Oh well, just caught my eye in the pic above.

link
Seems like it's a holdover from when Arcadio was the norm rather than the exception (not pronouncing the h). I don't see how though, as I've never heard anyone say "istoric", and you're 100% right about the a/an rule.

Quad-channel memory costs significantly more than dual-channel and brings with it no real-world performance improvements. Furthermore, you can get PCI-E 3.0 support on some LGA 1155 motherboards now, before SB-E is even available.

Unless you're getting a Six-core CPU, Sandy Bridge-E is completely useless.

Bulldozer is supposed to go against Sandy Bridge. If it loses, it loses.

But you don't need to run quad channel, do you? Besides, I don't consider memory as big as deal these days since prices are quite low. I'd give more weight to your argument if we were still paying $100 for 4 gigs.

 

[LOL_Wut_Axel]

But you don't need to run quad channel, do you? Besides, I don't consider memory as big as deal these days since prices are quite low. I'd give more weight to your argument if we were still paying $100 for 4 gigs.

Here's your options with SB-E:

Buy 16GB (4x4GB) DDR3: $90
Buy 8GB (4x2GB) DDR3: $50

The option for 8GB is more economical, but you're stuck with 'just' 8GB on an Enthusiast platform; you can't upgrade unless the motherboard has eight DIMM slots, which I doubt they'll have.

For SB you can get 8GB (2x4GB) DDR3 for $45, half the price of what you'd want to put on SB-E. Then we also have the fact that motherboards will be $50-100 more expensive, not to mention the CPU will be mostly a side-grade if you decide on an SB-E Quad-Core. As a whole, the platform is at least $100 more expensive. Bulldozer competes with Sandy Bridge because of overall cost, again.

 

[Terzo]

Can't you just put two 4-gig sticks into SB-E? That way you have 8 gigs and room to expand for the future.