AMD Zambezi news, info, fans !

Printable View

Show 100 post(s) from this thread on one page

10-02-2011, 09:42 AM
zoomee

Quote:

Originally Posted by imamage

I guess I can just save the money and wait for Trinity then
wasted the budget to grab Crosshair V :( (Well the board is fun to OC with)

You do realise that benchmark is highly unfair?
SB system has 16Gb of RAM, BD only 4Gb
SB system has 120Gb SSD, BD only has 500Gb HDD...... Anomolies like this usually make me question the authors intent......
10-02-2011, 09:44 AM
DedEmbryonicCe1

Quote:

Originally Posted by MaddMutt

Here I have a Chart..... Sorry it's not BF3. It's Chinebench 10 with my Deneb x4 955 and my Thuban x6 1100T both are OCed to the same speed. You can see that AMD did some under-the-hood tweaking with the x6.

Attachment 120731

I read in here somewhere in a post that one of our guys busted his Thuban by adding to much voltage to the uncore.:shakes::rofl: CHEW loaned/let him have one of his extras to hold him over until BD came out. Here is some testing that I did on this problem using AIDA64. I hope this helps :cool:

Attachment 120725

1.4v is too much for the CPU-NB of Thuban. 1.3875v damaged mine. Around 1.325v max would be safe for 24/7. For comparison I used to run my 940BE with 1.45v on the CPU-NB without any noticeable degradation.
10-02-2011, 10:24 AM
gosh

Quote:

Originally Posted by Mechanical Man

And it is max of four cores now. There is so much dependency between threads it cant extract more than 4 cores worth of parallerism out of threads. So more clock is better than more cores once you have four cores.
That basically means 8-core BD should be in max turbo all the time.

Hyperthreading seems to work well on BF3 and that indicates that there are a lot of cache misses. When the CPU-core waits on memory it kan jump to another thread. Cache misses also indicates that much memory and/or complicated memory patterns are used. Higher frequency is good but avoiding going to ram for data is also important.

If BF3 would have dedicated threads for different tasks, like one render thread, one AI thread etc then hyperthreading is not good.

When work are sliced in smaller jobs it is also important that threads are able to synchronize fast. Don't know if there are improvements there on bulldozer.

F.E.A.R vs BF3 beta
http://gamegpu.ru/Action-/-FPS-/-TPS...-test-GPU.html
http://gamegpu.ru/plugins/content/ju...5wbmd8QlJ8MzA=

http://gamegpu.ru/Action-/-FPS-/-TPS...-test-GPU.html
http://gamegpu.ru/plugins/content/ju...BuZ3xCUnwzMA==
10-02-2011, 10:25 AM
muziqaz

Quote:

Originally Posted by zoomee

If its of any use - My system (stock crossfire 5850 BE's) with 4Ghz on thuban is scoring the following in BF3 - which I am super happy with btw@ 1920x1200

ULTRA/HIGH (As high as it will go) - But no AA - AF set to x16:
Metro - outside 80 > 120FPS, Inside 90>150fps

ULTRA/HIGH 2xAA, AF x16
Metro - outside 35>50FPS, Inside 45>70FPS.

Didn't stress any further,, as I've been trying to play the game - someone kill those darn flashlight flashers!! aaaarggghhhh

Damn, AA gave your cards a run for the money :(
For me AA doesn't even break a sweat for the card :D
10-02-2011, 10:56 AM
zoomee

Quote:

Originally Posted by muziqaz

Damn, AA gave your cards a run for the money :(
For me AA doesn't even break a sweat for the card :D

Yer looks like it - I forgot to mention I used Cat 11.x (older ones) for the first bit of testing and the newer 11.10 BETA drivers for the AA testing - Them 11.10 drivers are well crap - First I got a black screen (could just see team members lol!) - Then I started to get loads of glitching. At least with the older drivers I was completely stable!!

Probably not a fair test - but from what I'm reading it helps to have 1.5Gb+ cards when running this game. Hey ho - Guess I'm still happy I can run it maxed out (without AA of course lol)

If anyone out therre is testing - don't forget to restart the game as video setting changes DEFINATELY don't take place until then.
10-02-2011, 11:15 AM
Mechanical Man

Quote:

Originally Posted by gosh

Hyperthreading seems to work well on BF3 and that indicates that there are a lot of cache misses. When the CPU-core waits on memory it kan jump to another thread. Cache misses also indicates that much memory and/or complicated memory patterns are used. Higher frequency is good but avoiding going to ram for data is also important.

If BF3 would have dedicated threads for different tasks, like one render thread, one AI thread etc then hyperthreading is not good.

When work are sliced in smaller jobs it is also important that threads are able to synchronize fast. Don't know if there are improvements there on bulldozer.
...

Yes I have to agree after running some fast tests on my system, it seems that I have been wrong and BF3 does like more cores over more frequency. I made too fast conclusion on max cpu usage pattern.

I tested with different cpu affinity & different cpu clocks. Most noticeable difference was that with only two cores used @ 3,6GHz was noticeable lag and much of it, but with four cores @ 1,8GHz it ran very smooth.

It was not in anyway scientific, I only used fraps, task manager et AOD. No charts were made. Btw, it run 38+ fps with 6 cores @ 1,8GHz.

GPU used is HD6870 @ stock. Settings: All high, no-aa, aniso 16x, HBAO, 1680*1050, cat 11.10pre.
10-02-2011, 11:41 AM
gosh

Quote:

Originally Posted by Mechanical Man

Yes I have to agree after running some fast tests on my system, it seems that I have been wrong and BF3 does like more cores over more frequency. I made too fast conclusion on max cpu usage pattern.

I tested with different cpu affinity & different cpu clocks. Most noticeable difference was that with only two cores used @ 3,6GHz was noticeable lag and much of it, but with four cores @ 1,8GHz it ran very smooth.

Interesting! BF3 almost acts like serverapplications. Frequency isn't allways that important, more cores and fast memory access is sometimes better.

I know that many gamers buys i5-2500K because they want to be prepared for future games. But if this is the future than that CPU will not be as future safe. Less L3 cache (6MB) and only 12 way set associative. Games using much memory and/or complicated memory patterns and i5-2500K cache is trashed in no time.
10-02-2011, 12:52 PM
MaddMutt

Quote:

Originally Posted by gosh

Interesting! BF3 almost acts like server applications. Frequency isn't always that important, more cores and fast memory access is sometimes better.

I know that many gamers buys i5-2500K because they want to be prepared for future games. But if this is the future than that CPU will not be as future safe. Less L3 cache (6MB) and only 12 way set associative. Games using much memory and/or complicated memory patterns and i5-2500K cache is trashed in no time.

Quote Originally Posted by gosh View Post

Hyperthreading seems to work well on BF3 and that indicates that there are a lot of cache misses. When the CPU-core waits on memory it kan jump to another thread. Cache misses also indicates that much memory and/or complicated memory patterns are used. Higher frequency is good but avoiding going to ram for data is also important.

If BF3 would have dedicated threads for different tasks, like one render thread, one AI thread etc then hyperthreading is not good.

When work are sliced in smaller jobs it is also important that threads are able to synchronize fast. Don't know if there are improvements there on bulldozer.

Question??????? the I5-2500k is a 4 core 4 thread processor???? If so and gamers bought it to play games, then they are not going to be happy as they don't have the HT to help them out.
10-02-2011, 12:59 PM
xdan

Speaking nonsense here. 2500K L3 cache is 2-3X faster than Phenom and Thuban.
Not to put also the IMC much, much faster.
So if it is 6 or 8MB it doesn't coun't.
How that BD with a total of 16MB cache doesn't overall beat 2600K?
If it would be so, than the price vouldn't be 245$.

Quote:

I5-2500k is a 4 core 4 thread processor???? If so and gamers bought it to play games, then they are not going to be happy as they don't have the HT to help them out.

The most stupid thing i ever heard...Until no we even don't have games to use more than 4 cores or threads.
Gamers should take care to buy a strong vga.
Will see about that if BF3 knows more.
And tests at 640*480 is also the bigest nonsense i ever heard.
10-02-2011, 01:09 PM
SBB

Quote:

Originally Posted by xdan

Speaking nonsense here. 2500K L3 cache is 2-3X faster than Phenom and Thuban.
Not to put also the IMC much, much faster.
So if it is 6 or 8MB it doesn't coun't.
How that BD with a total of 16MB cache doesn't overall beat 2600K?
If it would be so, than the price vouldn't be 245$.

The most stupid thing i ever heard...Until no we even don't have games to use more than 4 cores or threads.
Gamers should take care to buy a strong vga.
Will see about that.
And tests at 640*480 is also the bigest nonsense i ever heard.

Testing at 640x480 resolution is an attempt to rule out any GPU bottlenecks and just show the differences between CPUs (as in theory the GPU can easily run this at 200+fps if the CPU is fast enough) , whether it works or not is up for debate though
10-02-2011, 01:12 PM
gosh

Quote:

Originally Posted by xdan

Speaking nonsense here. 2500K L3 cache is 2-3X faster than Phenom and Thuban.

That is also the problem, it is fast partly because they removed some flexibility and size. If cores are waiting on data from ram it doesn't matter if cache is superfast.
10-02-2011, 01:20 PM
CrazyNutz

Quote:

Originally Posted by MaddMutt

Hyperthreading seems to work well on BF3 and that indicates that there are a lot of cache misses.

I would say it's more like BF3 (on ultra quality) all around benefits from more threads/cores.
10-02-2011, 01:23 PM
bamtan2

Quote:

Originally Posted by Olivon

2600K results coming from an OCed CPU for sure (~4500MHz)

thanks for the reality check. if the 2600k blew past bulldozer in math benchmarks by that much, I don't see what place bulldozer would have in servers.

also, even in the misleading benchmarks where bulldozer loses everything, the one thing it wins is floating point math. if bulldozer wins at floating point by such a large margin, that alone may make up for its IPC weaknesses WHEN PLAYING GAMES!
10-02-2011, 01:40 PM
Musho

Quote:

Originally Posted by Musho

Working with the things we know, we can calculate how fast roughly the 8150 is compared to the 1100T in single threaded tests. Looking at Cinebench screenshot comparing the 1100T singlethreaded vs multithreaded, the difference is a 5,261261261 speedup, which is explained by the fact in the single threaded load the 1100T is running at 3,7ghz, while in the multithreaded test it's running at 3,3ghz. Those CPU Passmark benchmarks posted above are very well multithreaded so we'll assume it will show roughly the same scaling going from multithreaded to singlethreaded:

6314/5,261261261 = 1200 for the 1100T @ 3,7ghz

Now we know from movie posted about Bulldozer running Fritz Chess benchmark, that turbo in single threaded workloads can keep Bulldozer running at 4,2ghz with 1-2 cores loaded, while it runs at 3,6ghz with 8 cores loaded. Also, the cores are running at 80% efficiency when both cores in a single module are loaded, so instead of dividing by 8, we'll be dividing by:
8*0,8 = 6,4.
We also shouldn't forget the speed difference in multi and single threaded tests:
4,2/3,6 = 1,166666667

This brings Bulldozers calculated single thread score to:

8681/6,4*1,166666667 = 1582 for 8150 @ 4,2ghz

This gives Bulldozer a 1582/1200 = 1,319 or 31,9% speedup compared to thuban in single threaded workloads at stock speeds with turbo enabled for both.

OR

(1582/4,2)/(1200/3,7) = 1,162 or 16,2% IPC boost

Disclaimer: This is all just some theorycrafting, but it feels right about what to expect. Take these calculations with a grain of salt however :)

Look at this: http://semiaccurate.com/forums/showp...postcount=2208
If this is true, I was _very_ close with my math skillz =)
10-02-2011, 02:51 PM
BeepBeep2

Quote:

Originally Posted by MaddMutt

I read in here somewhere in a post that one of our guys busted his Thuban by adding to much voltage to the uncore.:shakes::rofl: CHEW loaned/let him have one of his extras to hold him over until BD came out. Here is some testing that I did on this problem using AIDA64. I hope this helps :cool:

That's not how Dave's CPU died.
10-02-2011, 03:06 PM
radaja

Quote:

Originally Posted by Musho

Look at this: http://semiaccurate.com/forums/showp...postcount=2208
If this is true, I was _very_ close with my math skillz =)

isnt he saying that BD will get a 22.5%(-5%=17.5%=22.5%) performance increase
from a new Stepping and bios update,his friend says BD is -5% less IPC than thuban?
that would be one of the best stepping/bios improvements ever wouldnt it?
10-02-2011, 05:52 PM
Blacklash

radaja, did you post the below over here yet? I seriously laughed at this.

"well i can tell you it should be bright and sunny, based on what i have been told
from my contacts.....
my brother works at AMD and he said all is superb with BD
my sister works for Global foundries and she said 32nm is 100% perfect
my grandpa works for the Guiness Book of world records and he said BD is the fastest CPU in the world.
my uncle works for Cray and he said Interlagos is the most powerful CPU ever made.
my other sister works for intel and she said everyone is very scared right now at intel and are struggling
to come up with a solution to BD
my mom works for Cyberpower PC's and they are planning on dropping all intel based machines in favor
of AMD's BD for the next 5 years

but this is all just a bunch of BS on my part,nothing i just said is true"
10-02-2011, 06:47 PM
radaja

Quote:

Originally Posted by Blacklash

radaja, did you post the below over here yet? I seriously laughed at this.

no,its not needed here in this BD thread,but it was needed in the one it was posted at:D
10-02-2011, 07:51 PM
imamage

Quote:

Originally Posted by zoomee

Yer looks like it - I forgot to mention I used Cat 11.x (older ones) for the first bit of testing and the newer 11.10 BETA drivers for the AA testing - Them 11.10 drivers are well crap - First I got a black screen (could just see team members lol!) - Then I started to get loads of glitching. At least with the older drivers I was completely stable!!

Probably not a fair test - but from what I'm reading it helps to have 1.5Gb+ cards when running this game. Hey ho - Guess I'm still happy I can run it maxed out (without AA of course lol)

If anyone out therre is testing - don't forget to restart the game as video setting changes DEFINATELY don't take place until then.

I stick with Catalyst 11.9 and it works fine with BF3 Beta
10-02-2011, 11:53 PM
Olivon

http://tof.canardpc.com/view/1486a2c...1ae65bba73.jpg

http://cpc.cx/38W
10-03-2011, 12:14 AM
marten_larsson

AM3+ Socket?

No dates when these slides were made?
10-03-2011, 12:38 AM
maltrabob

Quote:

Originally Posted by marten_larsson

AM3+ Socket?

No dates when these slides were made?

Indeed. Also "Higher Memory Support" of up to 1866 MHz sounds funny with having this support on 1st generation Zambezi FX already.

Edit: obviously not very up-to-date slide, there is no native PCI-E 3.0 or USB 3.0 either
10-03-2011, 12:38 AM
undone

I not only notice the integer performance increase(like Phenom to Llano) but more important is the FMA3 support which is only support on Haswell with Intel plan!:eek:
10-03-2011, 01:02 AM
pumero

Quote:

Originally Posted by maltrabob

Edit: obviously not very up-to-date slide, there is no native PCI-E 3.0 or USB 3.0 either

How should there be PCI-E 3.0 and USB 3.0, when they are still using the AM3+ plattform with the 990FX Chipset?
Technically it might be possible to use the A75 FCH instead of the old SB850/950 to gain USB 3.0 but it doesn't solve the problem of PCI-E 2.0 in the aging Northbridge.

If those slides are legit, they can't be that old, since 2nd Gen Bulldozer was supposed to use socket FM2 and the change to stick to AM3+ was only recently.
10-03-2011, 01:08 AM
-Boris-

Quote:

Originally Posted by Olivon

http://tof.canardpc.com/view/1486a2c...1ae65bba73.jpg

http://cpc.cx/38W

Hope that 10% performance increase is per clock cycle. If it's with frequency i doesn't look to good.

Show 100 post(s) from this thread on one page