AMD Zambezi news, info, fans !

[TheBreezyBB]

CPUz doesn't support BD in version 1.57.1 only in 1.58. So these so called BD benchmarks are probably fake.

[muziqaz]

Is that QR Code on the CPU?
If so, any1 here have QR Code Reader Software on their mobile so we can decode it?

Doubt that. At least my phone doesn't pick up anything

[demonkevy666]

CPUz doesn't support BD in version 1.57.1 only in 1.58. So these so called BD benchmarks are probably fake.

that's the first thing I noticed.

they believe their is a problem with B1's L3 cache.

if you look you will see it's write cache that is the slowest part here not just L3 cache all the way threw write cache is too slow.

[Manicdan]

why in the world would they run cb10 and not cb11.5

[Particle]

The 2D code on AMD CPUs is semacode (datamatrix), not QR. You can get them read online from images if you're interested. That's why I always chuckle when people blur numbers or barcodes but leave semacodes untouched.

Take this Thuban for instance, which is my old 1055T.



Isolate the code and feed it into something like this: http://www.2dtg.com/decode.html

9316895C00278_HDT55TFBK6DGR

They contain the serial number and product codes. I took a look at the one in the German article, but the code is badly scratched. I can't get a clean decode. Revision: I'm almost certain the semacode has been purposely sabotaged. It contains intentional formatting mistakes that can't be caused by scratches. Clever.

[sergiojr]

they believe their is a problem with B1's L3 cache.
if you look you will see it's write cache that is the slowest part here not just L3 cache all the way threw write cache is too slow.

It's already mentioned in optimization manual that was released in April. Also AMD promises to fix this in BD version 2.
http://support.amd.com/us/Processor_TechDocs/47414.pdf

The following performance caveats apply when using streaming stores on AMD Family 15h cores.
• When writing out a single stream of data sequentially, performance of AMD Family 15h
processors is comparable to previous generations of AMD processors.
• When writing out two streams of data, AMD Family 15h version 1 processors can be up to three
times slower than previous-generation AMD processors. AMD Family 15h version 2 processor
performance is approximately 1.5 times slower than previous AMD processors.
• When writing out four non-temporal streams, AMD Family 15h version 1 can be up to three
times slower than previous AMD processors. AMD Family 15h version 2 processor performance
is comparable to previous AMD processors.
• Using non-temporal stores but not writing out an entire cacheline may cause performance to be up
to six times slower than previous AMD processors.

[undone]

Those who only post benchmark but with masked 2D code on ES chips are obviously either fake or initial ES, no doubt.

[FlanK3r]

turbo not worked good, I thinking...This superpi is near 3600-3700 MHz max. OBRs results with ASUS board are better (and the same CPU)

[undone]

OK, I suddenly realized, that Cinebench R10 score is similar to that OBR posted before, higher than Thuban 1090T about 50% and 980x about 15%. IF this is real perhaps those other benchmark like superpi & R11.5 cause some bug during test and result in crappy score. Wait & see some result about mature ES chips or retail chips.

[FlanK3r]

no, its not simillary. OBR had 27 700 in R10 with UD7 and the same chip, superpi 15.4s there is only 24 400 and 19.5s

[Olivon]

no, its not simillary. OBR had 27 700 in R10 with UD7 and the same chip, superpi 15.4s there is only 24 400 and 19.5s

And what clocks OBR used during this test with UD7 ? Stock ?

[Macadamia]

At stock, BD is nearly the fastest x264 CPU there is.
At 1st pass, 980x and 2600k do about 95-105 fps, whilst BD does 135 most likely because this pass has less TLP and allows for more turbo.

2nd pass where things get more threaded, 2600k does 36 or so, 980x does 48, and BD scores 45 here.
Probably using all threads and can't consistently turbo, and pass 2 favors Intel a bit more.

EDIT:
Also wins in Fritz.

2600k - 13,017 // 12834 (another source)
2600k at 5.2 - 19288
870 - 11,995 // 875k does 12450
980x - 12,733
1100T - 11,219

BD - 14197

[DedEmbryonicCe1]

The 2D code on AMD CPUs is semacode (datamatrix), not QR. You can get them read online from images if you're interested. That's why I always chuckle when people blur numbers or barcodes but leave semacodes untouched.

Take this Thuban for instance, which is my old 1055T.



Isolate the code and feed it into something like this: http://www.2dtg.com/decode.html

9316895C00278_HDT55TFBK6DGR

They contain the serial number and product codes. I took a look at the one in the German article, but the code is badly scratched. I can't get a clean decode. Revision: I'm almost certain the semacode has been purposely sabotaged. It contains intentional formatting mistakes that can't be caused by scratches. Clever.

I tried as well. I wasted quite a bit of time trying to recreate the code from the donanimhaber pictures on the last page but the combination of photoshopping and strategically placed watermarks was too much. Even accounting for the obvious errors like there not being a solid line along the left side (filled those in, a thin dark line is visible along the left side where some dots were erased) I have too many errors for a decode. That is not too surprising as I can see other spots with a faint smudge (one just above middle/center) where a dot was either erased or it is an artifact around the edges from the copy->pasting they did. If it weren't for the watermarks the other photos could be used but they are no help other than the obvious differences in the lower left corner.

[undone]

Also wins in Fritz.

2600k - 13,017 // 12834 (another source)
2600k at 5.2 - 19288
870 - 11,995 // 875k does 12450
980x - 12,733
1100T - 11,219

BD - 14197

Is Fritz multi-threaded? 980x little slower than 2600k, seems Fritz doesn't use more than 4 threads.

[FlanK3r]

Olivion: the same clocks as donanimhaber

[Macadamia]

Fritz uses >4 threads. i5 750 scaled to 870 clock speeds = 9 355; with Hyperthreading it's 12k or so.

[Olivon]

Olivon: the same clocks as donanimhaber

Good newz ! Hope that's true ...

[muziqaz]

At stock, BD is nearly the fastest x264 CPU there is.
At 1st pass, 980x and 2600k do about 95-105 fps, whilst BD does 135 most likely because this pass has less TLP and allows for more turbo.

2nd pass where things get more threaded, 2600k does 36 or so, 980x does 48, and BD scores 45 here.
Probably using all threads and can't consistently turbo, and pass 2 favors Intel a bit more.

Do you have a source for x264 bench?
I remember doing extensive benching on that benchmark some time ago with my x4. And I found that CPU NB clock OC gives biggest boost to that benchmark.
In 1st pass my x4@3.8ghz was kicking hard i7-920@3.8ghz. Wasn't winning but close. 2nd pass was a bit different story, thou still not far from i7 because of my tight timings and high NB clocks.

[informal]

It's the same DH "preview".Go to gallery ,there are 12 images there,one is x264 benchmark.

[XRL8]

CPUz doesn't support BD in version 1.57.1 only in 1.58. So these so called BD benchmarks are probably fake.

People keep saying that.Thats just not true.
FULL BD support was brought with 1.58.It supported BD earlier.
http://wccftech.com/official-bulldoz...aurthor/20986/

Thats CPUZ creator showing a screenshot of his BD with 1.56.Also 186W isnt sign of a "fake" .Its just what cpuz shows for now.

Thats B1 stepping they got there.Probably clocks are not f-up anymore.There isnt previous confirmed B1 if i recall correctly (OBR MAY have one tho, he hid it well)

[muziqaz]

It's the same DH "preview".Go to gallery ,there are 12 images there,one is x264 benchmark.

oh, thanks

[JumpingJack]

At stock, BD is nearly the fastest x264 CPU there is.
At 1st pass, 980x and 2600k do about 95-105 fps, whilst BD does 135 most likely because this pass has less TLP and allows for more turbo.

2nd pass where things get more threaded, 2600k does 36 or so, 980x does 48, and BD scores 45 here.
Probably using all threads and can't consistently turbo, and pass 2 favors Intel a bit more.

EDIT:
Also wins in Fritz.

2600k - 13,017 // 12834 (another source)
2600k at 5.2 - 19288
870 - 11,995 // 875k does 12450
980x - 12,733
1100T - 11,219

BD - 14197

http://www.tomshardware.com/reviews/...0x,2584-8.html

FriTz for a 980 is is > 18000... looks like you took your score from some site that was doing a thread for thread comparison.

[kaktus1907]

It's the same DH "preview".Go to gallery ,there are 12 images there,one is x264 benchmark.

ah my bad i miss that.. DH says more tests coming..

reference x264 results..
http://www.forum-3dcenter.org/vbulle...d.php?t=508846

[muziqaz]

Here is thuban 3.8ghz

Results for x264.exe r1913
==========================

Pass 1
------
encoded 1442 frames, 139.77 fps, 3913.31 kb/s
encoded 1442 frames, 138.15 fps, 3913.31 kb/s
encoded 1442 frames, 138.37 fps, 3913.31 kb/s
encoded 1442 frames, 138.55 fps, 3913.31 kb/s

Pass 2
------
encoded 1442 frames, 36.72 fps, 3960.06 kb/s
encoded 1442 frames, 36.94 fps, 3958.73 kb/s
encoded 1442 frames, 36.51 fps, 3959.39 kb/s
encoded 1442 frames, 36.87 fps, 3959.50 kb/s

In pass 2 it uses 96% of CPU, and pass 1 leaves one core out.

[demonkevy666]

It's already mentioned in optimization manual that was released in April. Also AMD promises to fix this in BD version 2.
http://support.amd.com/us/Processor_TechDocs/47414.pdf

The chips stepping is right in AIDA, they blocked it on cpu-z lol at that failure OR-B1

turbo not worked good, I thinking...This superpi is near 3600-3700 MHz max. OBRs results with ASUS board are better (and the same CPU)

why does/would turbo be use on all 8 cores ?
I though it was just for single thread and up to 4 core maybe 6.

At stock, BD is nearly the fastest x264 CPU there is.
At 1st pass, 980x and 2600k do about 95-105 fps, whilst BD does 135 most likely because this pass has less TLP and allows for more turbo.

2nd pass where things get more threaded, 2600k does 36 or so, 980x does 48, and BD scores 45 here.
Probably using all threads and can't consistently turbo, and pass 2 favors Intel a bit more.

EDIT:
Also wins in Fritz.

2600k - 13,017 // 12834 (another source)
2600k at 5.2 - 19288
870 - 11,995 // 875k does 12450
980x - 12,733
1100T - 11,219

BD - 14197

Fritz uses >4 threads. i5 750 scaled to 870 clock speeds = 9 355; with Hyperthreading it's 12k or so.

fritz does use more then 4 threads I test it on My thuban at 4.0ghz and 3.0ghz Nb with 2000mhz ram I get around 13,000 relative speed is about 27.5

http://www.tomshardware.com/reviews/...0x,2584-8.html

FriTz for a 980 is is > 18000... looks like you took your score from some site that was doing a thread for thread comparison.

scaling isn't linear that's all.