Printable View
Whauuuu:) First performance slide!
http://i.imgur.com/udZiT.jpg
Testing in:
- PC MArk Vantage - TV and movies
- Cinebench R11.5
- 3D Mark 06 (CPU část)
Render performance is more than awesome!!!i7 980x will be here beating!
PC Mark:
x6 1100T 6430b
i7 980x 7150b
i7 2600k 7700b
AMD Zambezi 7950b
Cinebench R11.5
x6 1100T 5.9b
i7 980x 8.95b
i7 2600k 6.7b
AMD Zambezi 11b !!!
3D06 CPU
x6 1100t 5990b
i7 980x 7388b
i7 2600k 6600b
AMD Zambezi 8800b!
Listing the source would be appropriate.
Good looking numbers if true.
Also, the image does not work.
What speed was the CPU?
Just found this...
http://www.donanimhaber.com/islemci/...s-dokumani.htm
CPU speed is not important, if it at stock clocks...
where did you find the numbers? i already saw the slide but dont know where the source is for actual numbers
So its probably the fastest chip they'll be releasing at launch then.Quote:
Originally Posted by FlanK3r
Would of really gotten excited if someone was to say,.. its only the 3GHz $199 mid-range part, the 3.8GHz $299 black edition is another 30+% faster! :rofl:
If these are true it still does not indicate single threaded performance.
Most of these apps are multi-threaded and with BD's new design it seems that we already knew that the multithreading would be greatly improved.
If we see Pi runs that match nehalem or DB then we know there was a strong FPU/Single thread increase
Pi? Seriously?
you want to run single threaded x87 app on next gen x64 modular CPU to see it's performance?
with turbo2.0 it will be really tough to say what performance will be since clock rates will be higher, and there is currently no details on clock rates with/without turbo.
Surprisingly yes.Quote:
Originally Posted by muziqaz
Not to measure performance as a whole...but to see single threaded performance. If that has improved, then the architecture compare to K10.5 has been dramatically improved, and may actually catch up/beat intel.
Multithreaded performance is not everything, as some APPs are still coded for single threads.
If the single threaded and multithreaded performance are good then we can see a nice all around chip.
the problem i have with Pi is how old and unoptimized it is
whats a good REAL USE single threaded test for today? besides games that only use 2-4 cores?
Unoptimized and old yes.Quote:
Originally Posted by Manicdan
But I see Phenom II's @ 6.2GHz+ doing 10 seconds under LN2 on 1m
SB is doing 6 seconds at like 5.5.
If the BD design has improved that much, we will at least see some faster pi.
Pi stresses the x87 logic inside the chip.That's is not representative of single thread performance. Single thread integer SIMD workloads are the most representative of single thread "IPC". BD should do well here too.
then use pifast as a comparison.
bottom line good single threaded performance means better over all performance in multithreaded as well since CPC is faster.
A chip for instance that has the same CPC as thuban would juat be thuban with 2 cores added, where as a chip that has much faster CPC would be able to match thuban in multithreaded with just 4 cores........
SuperPi wouldn't show us a thing. This is a completely new uArch that surely has alot of it's optimizations into code and instructions we can only dream that SuperPi has :rofl: Intel nor AMD barely cares about x87 anymore and surely isn't optimizing their newest and hottest for SuperPi.Quote:
Originally Posted by Slappa
You are correct chew* .Don't forget that IPC(or CPC) is only one part of performance equation.There is working frequencies too ;). BD is supposed to be able to clock very high,so 2module/4 core Zambezi should not only be faster than Deneb QC at the same clock ,but should be able to work at much higher clocks from the start(think 4Ghz stock with Turbo on top of that). There are rumors of 3.5-3.8Ghz range for 95W 8 core parts :).
Now that would be one epic cruncher / air gaming chip...Quote:
Originally Posted by informal
Sounds good, atm AMD is tight lipped. I'm really not in the loop anymore for whatever reasons.
To bad we don't have clock sppeds for those numbers. I calculated awhile ago that 8 cores would put us on par with 980X as HT effeciency seems to drop on that platform the more cores they add.
Quote:
Originally Posted by FlanK3r
source?
scores? A logic, some Hans, some i found at the web.
Think, in superpi we will se time about 14s at stock clocks, maybe a bit under 14s
i hope no one hates me for wanting a 2module/4core BD. with overclocking it should still beat my thuban in multithreaded, but demolish it for gaming. i just dont have any reason to use more than 4 cores right now and lower power consumption means i can go cheaper on the board and save a little cash on the electric bill, and alot of cash on the chip. i just feel like i should throw myself off a bridge for upgrading to less cores, lol.
So it's nothing just useless speculation again.:down:Quote:
Originally Posted by FlanK3r
That would be AMD's equivalent to Sandy Bridge. Ivy Bridge is what this should be competing with.Quote:
Originally Posted by Manicdan
An user on SemiAccurate.Quote:
Originally Posted by Oliverda
His (the user) calculations unfortunately based on a fake slide.Quote:
Originally Posted by TOOL_CHILE
Lets say these numbers were true:
PC Mark:
x6 1100T 6430b / 6 =1071 per core
i7 980x 7150b / 6 = 1191 per core
i7 2600k 7700b / 4 = 1925 per core
AMD Zambezi 7950b / 8 = 993 per core < --- slowest single thread
Cinebench R11.5
x6 1100T 5.9b / 6 = .98 per core
i7 980x 8.95b / 6 = 1.5 per core
i7 2600k 6.7b / 4 = 1.675 per core
AMD Zambezi 11b / 8 = 1.375 per core <---behind both intel CPU's
3D06 CPU
x6 1100t 5990b / 6 = 1000 per core
i7 980x 7388b / 6 = 1231 per core
i7 2600k 6600b / 4 = 1650 per core
AMD Zambezi 8800b! / 8 = 1100 per core <----behind both intel CPU's
If these were true (and they aren't) it would fail.
So now the thread can be closed.:ROTF:
i didnt know much about pcmark so i looked around:Quote:
Originally Posted by BeepBeep2
http://benchmarkreviews.com/index.ph...1&limitstart=5
so just divide them all in half i guess for single threaded perfQuote:
The TV and Movies suite concentrates on video playback and transcoding, but only uses two threads at a maximum
beeP.No, its wrong, u can not calculate /8...Efectivity 8 cores is not 8x, but example between 6.7-7.7 example (look at Denebs or Thubans)
4 modules, 8 cores...8 threads. :confused:Quote:
Originally Posted by FlanK3r
:rofl: You make it seem like intel doesn't have HT... :screwy: On the 2600K and the 980, it's more like 8 threads and 12 threads instead of 4 and 6. While undoubtedly, HT is nothing like adding full cores, it's has been shown to add 30-50% performance increases.
Still though, on the topic of single threaded performance not mattering, somebody needs to read up on Amdahl's Law :p. Applications can only be threaded so far until the scaling stops. In which case, the performance on each individual thread is once again the most important aspect. :yepp:
There is a definate reason why I said cores and not threads.Quote:
Originally Posted by ChanceCoats123
can someone close thread, the slides are fake... ><
this is about Zambezi - speculation, infos etc...the same what had some Intel fanboys about their lovely Sandy bit* :-/. Or, do u have some proof about "this is fake or this is not fake"? Think, slide looks as from AMD presentation and scores is possible. Why not? On the other hand, Im not saying that this is also true. But I cant simply say, its bullshi*, its fake, close this thread etc.
Beep: example with x6 1100T, at Cinebench R11.5 1CPU score is about 1.1points, with 6 cores is it about 5.8points. SO efectivity is 5.8/1.1= 5.2x, in R10 its 4.6x for 6 cores.
JF-AMD even mentioned this isnt a amd slide.. :down:Quote:
Originally Posted by FlanK3r
I guess this slide is fake cuz there's some mistakes on it. OTOH I suspect there's sth tricky on the diagram......
hm, JF is server guy, not desktop. Maybe he talked only in style "Im thinkink..."
No, I said I can't verify it. That is completely different.Quote:
Originally Posted by Johnny87au
Headshots all around hahah.
Come on guys take it easy, like Flank3r says, this is the same game like on Sandy bridge threads.
Regards.
Something i came across that i found is on wiki (i know not the most accurate), is that there is something about a dual MCM.
I take its gonna be similar to Core 2 Quad in theory but will it actually happen if the 8 module setup is gonna be reaching the 125watt limit?
If so does anyone know if they are going to go across the HT link or if they are gonna go through a proprietary interconnection like on server processors? If not tehn i think the HT link is gonna become a slight bottleneck in that aspect.
Desktop bulldozer wont use MCM design.Only the server 16 core one.
this is going to be a long couple of months...
Ain't that the truth. :(Quote:
Originally Posted by stangracin3
this is only shows how really eager we all are for the Bull!!!
Let it Loose :D
Beep: look example at this. Cinebench R10
x6 1100T at stock
http://img98.imageshack.us/img98/368...r10default.png
efficiency is not 6x, but 4.48x
And here at i7 980x at stock
http://img442.imageshack.us/img442/3...nebenchr10.png
efficiency is 5.56x (because 12 threads!, not 6 only running)
absolute efficiency doesnt exist, never CPU has efficiency as number of logical cores (threads)
Well, I believe part of the reason AMD's multi-core efficiency is bad is because the cores don't communicate well. What result do you get with the 980X when HyperThreading is off?
In theory it should be similar to when its off if not a little better.
beep, tomorow Il try it.
Because of Turbo ;)Quote:
Originally Posted by BeepBeep2
Single thread - 3.6GHz
Multi thread - 3.2GHz
Efficiency will be low for sure.
Turn off Turbo and efficiency will be higher.
I think those numbers were made up by someone on a different forum as an example of performance increase based off of that graph on the 2nd post in this thread.Quote:
Originally Posted by FlanK3r
I think you have to take into account HT. Cuz those intel numbers are with HT on I would assume, thus giving nonrealistic single thread / core performance.Quote:
Originally Posted by BeepBeep2
So, in order to compensate you should take into account up to 30% benefit from HT and take that much off intels core count.
So, for the i7 980x it would be 8 cores and 2600k would be like 5.25 cores
Why is the 1100T in single channel? We all know intel is much faster, but you're really comparing apples to oranges more so than already. Put both rigs in single channel, and I'm willing to bet the intel efficiency will drop. More than likely, not a whole lot, but the faster ram subsystem on the 1366 platform is a huge part of why those chips score so much better.
I'm with you on that. If the 970 had been available back then I would have never bought my 1090T. The only thing I use more than 4 cores for is converting lossless audio to Ogg vorbis to put on my Rockbox'ed Sansa Fuze. That process is limited more by my HDD speed than anything else at this point so I doubt I'd lose much with 4 cores. That said I may still end up with a 8 core Bulldozer some day if there is a sufficient difference in overclocks versus the cheaper parts because I can never get enough single-threaded performance.Quote:
Originally Posted by Manicdan
Thing is, thubans overclock generally better than deneb, so i dont see a point getting 970 for single core performance :up:.Quote:
Originally Posted by DedEmbryonicCe1
Anyhow, im sure 2 module bulldozer will show up eventually to be replacement for Phenom X2 and Athlons.
Besides, with more potent turbo i dont think there is going to be a problem with single cpu performance.
Intel has turbo too ;)Quote:
Originally Posted by Lightman
what rav said is rightQuote:
Originally Posted by DedEmbryonicCe1
with my 1055T (that i got for 126$) i was at 3.7ghz on all six cores, but 4.25ghz on the 3 turbod cores, which is a really nice OC and really tough to get on all 6 stable because of the really high power draw (it uses 1.52v in turbo)
and HT to make enough difference so that results between AMD and Intel are not comparable ;)Quote:
Originally Posted by BeepBeep2
Just turn off Turbo in AOD and test for yourself, see how Core Speedup changes in Cinebench :yepp:
Ok, here is it Beep: CPU at 4400 MHz without turbo, one with 6/12 and second with 6/6:
5.51x
http://img6.imageshack.us/img6/9638/r10hton.jpg
4.93x HT off
http://img841.imageshack.us/img841/9268/htoff.jpg
PS:tomorow Il try the same with Thuban (OC+without turbo)
Ok, I have here Thuban Cinebench R10 from my friend (he made it earlier than me :) )
http://img204.imageshack.us/img204/3...nchefroxic.jpg
And as u can see, Thuban efectivity is 5.05x and better than Intel Gulftown without HTT.
BRAVO FlanK3r. Very Clearly . The Core usage of AMD better than Intel.
I think Beepbeep dont want to talk anymore
Looks like someones holding a grudge :shakes:Quote:
Originally Posted by delasneo
Flanker and I are forum "friends" I suppose you could say :D
delasneo, I look like an intel fanboy to you? :ROTF:
Hyper threading isn't effective as a real core for Mutli-thread. up to about 85% maybe in some, but not all.
85%?? Are you kidding? that is almost 3X the intel marketing claims (30%) and almost 6X the SPEC INT RATE benchmark delta (14%).
maybe he thought 15% (100-85) for ht :)
in Cinebech was it overall +4000p for next 6 threads (HT on), so on core isnt much-666p= 10.6%
great news about Bulldozer Turbo from JF
http://img130.imageshack.us/img130/7374/15083115.png
from COmputerbase....
When are they coming out with zambezi quads or 6 cores is there an actual date yet?
I haven't seen any dates yet, we probably won't see any before Cebit at least, which is in 1 - 5 March.
The actual launch could be at Computex in 1 - 5 June.
cebit 6 core not sure about the other zambezi chips, havent really herd anything about them..
Throwing this out there too but anyone know what the default mem speed it will be using be? Just thinking about ram purchases for this rig.
I really hope this is good and doesn't make me buy a sandy.
I'd guess 1600 compatible with a possible 1866 divider?Quote:
Originally Posted by Glow9
Either way I believe it will be better than thuban, maybe 2200 would be something most people could get stable.
Score. I just hope the new board support a higher tRC clock. My 790FX only goes to 42 which puts 2000 just out of reach with my current PSC chips. I'll have to see how much better the IMC is cold though. So far, these sticks have been way better than my D9's so they should go a bit farther.
Are you sure tRC is the limiting factor?Quote:
Originally Posted by ChanceCoats123
To be honest, no. It could be the cpu's IMC. I just posted in your other thread about temps, and I think that's the culprit. Like I said, I may get totally different results when cold. ;)
I am highly suspect of these so called performance numbers for Bulldozer. I am all for AMD getting back into the high end as competition is good, but lets not get carried away.
I remember all the fuss about the Radeon 6970 before launch. It's going to destroy nVidia a lot of people said! Then it launches and it's barely faster than an old GTX 480 and slower than the GTX 580. Talk about over-hype deflation.
couldn't be as under wellhiming as phenom I. that was on the same node this is new node and newer techniques.Quote:
Originally Posted by Callsign_Vega
Not cool double post :/ fricken weather here sucks.Quote:
Originally Posted by Callsign_Vega
QFT!! Never hype up a item till its released, lesson learned :up:Quote:
Originally Posted by Callsign_Vega
Priiiiiiiice!
I think, it will be new busspeed (base clock), not 200 MHz, because u cant have dividier 1866 MHz with 200 MHz baseclock. It must be divisible...Quote:
Originally Posted by BeepBeep2
You would think, but they are getting 1066/1333/1666 etc ;-)Quote:
Originally Posted by FlanK3r
Bobcat has not 200 too, but only 100 MHz baseclock :)
1066 is 3:8. 1333 is 3:10. 1600 is 1:4. Nothing out of the ordinary there. :shrug:Quote:
Originally Posted by RaV666
1666 is 6:25 and 1866 is 3:14. A 100 Mhz HT reference clock wouldn't make those fractions any prettier.
That does seem to be right.Quote:
Originally Posted by FlanK3r
Bobcat still runs at 200MHz bus clock, with an x8 multiplier.
Coolest:
Watch this, hm?Im confused now, I seen some CPU-Z with 100 MHz baseclock at stock and some at 200 MHz.
http://images.anandtech.com/gallerie...1268_575px.jpg
Safe to say that one will be the $1000 chip? Curious about the 4/6 cores.
Bulldozer will come under the names A4, A6 and A8 depending on model, these numbers will also be part of AMD's Vision-branding. AMD's goal is to make it easier for consumers with AMD Vision which is why Brazos, Lynx and Zambezi will be part of it instead of having alot of different models people need to remember. So A4 will be a Quad-Core, A6 Hex-core, A8 Octa-core etc...
This is what AMD are planning internally but are a subject of change, what is set in stone is a much bigger focus on AMD Vision. Unfortunately I wasn't the one to see the planned Bulldozer logotypes :shakes: Will post here if I find out about anything else...
FX-A8 or FX-A16. I could dig that. :DQuote:
Originally Posted by Smartidiot89
Or even better for good old memories!Quote:
Originally Posted by TheBlueChanell
AMD FX-A8 7200+ :rofl:
Seeing what AMD "done" with Athlon X2 5000 => Phenom Quad Core FX-5000 ;)Quote:
Originally Posted by Smartidiot89
They surely know how to name it up right for Bulldozer :up:
Very interesting. I guess we need more chips\setups to test with. I only had the chance to test Core Temp on one Bobcat, and it and CPU-z reported 200MHz bus clockQuote:
Originally Posted by FlanK3r
I have last info from Mumak, he said me, baseclock for Bobcat is really 100 MHz.
isn't the bus linked to the gpu part of a pci-express x4 link ?Quote:
Originally Posted by The Coolest
I'm guessing it's depends on which speed they set 1.0 or 2.0 then.
well less stress else where and less heat.Quote:
Originally Posted by FlanK3r
and bobcat is a low end embeded chip ... so it doesnt mean that zambezi will be at a bus speed of 100mhz ...
Holy crap, did I hear a speculated 1000$ for one of them? Jesus, what happened to the 200$ chips like the Phenom II X4 BE?
right, by I think, it will be not 200...Quote:
Originally Posted by Sn0wm@n
There is no way AMD would make a quad core part $1000. Maybe a 12 - 24 core part...Quote:
Originally Posted by ScottALot
I think that the Bulldozer 4 core @ 3.5GHz will be priced about $230ish. That is, assuming they even make a 4 core part. Maybe they will start with 6 core for the same price as 4 core.
The Bulldozer Black edition for unlocked multiplier for another $10-15.
I'm seriously doubting the high speculated prices. They're just speculation. What solid evidence is there to show Amd will move away from their current, affordable pricing?
based on assumed performance relative to SB, a 4 core BD will be around 150-250$, 6 core 200-350$, and 8 core 300-499$ based on clock speeds. (again all my assumptions)