AMD Bulldozer Eng. Sample leaked, benched

[renq]

I agree that BD has to equal SB's performance as a total but not core2core. And that's because from C2D to present Intel was 15-20% ahead in c2c performance and with SB it's another 10%. So, even if BD is 10% back in c2c but wins overall in multithreaded programs, it's very nice a product.

Lets leave CLK-2-CLK performance aside. First 8-core BD gaming tests show it to be equal to 6-core (12 thread) Gulftown in gaming tests. Of course, we don't know, how much BD is/was capped, but nevertheless, shouldn't expect any major improvements in the ready-to-market revision of Bulldozer. Now, faster Sandys, especially the unlocked K-versions, are equal or better than Gulftown in gaming tests. Thus 8-core BD ~= 4-core HT Sandy.

Now lets look into the future, Enhanced BD (Komodo) will have, IIRC, up to 10 cores, which with probable architectural advancements and possibly a bit higher clocks, should yield about 30-35% performance improvements over 8-core BD (in gaming, that is, provided games will be even more multicore/-thread happy). Now, Sandy-E will bring, IIRC also, up to 6-core HT-d CPU-s. Now, on paper at least, it should bring at least 50% increase in performance provided linear scaling (lets leave aside the ~1K+ USD pricetag it'll prolly have).

So, even if somehow BD takes the throne of the fastest (gaming) CPU this fall/year, AMD will have tough competition from Sandy-E next year...

 

[Pestilence]

Lets leave CLK-2-CLK performance aside. First 8-core BD gaming tests show it to be equal to 6-core (12 thread) Gulftown in gaming tests. Of course, we don't know, how much BD is/was capped, but nevertheless, shouldn't expect any major improvements in the ready-to-market revision of Bulldozer. Now, faster Sandys, especially the unlocked K-versions, are equal or better than Gulftown in gaming tests. Thus 8-core BD ~= 4-core HT Sandy.

Now lets look into the future, Enhanced BD (Komodo) will have, IIRC, up to 10 cores, which with probable architectural advancements and possibly a bit higher clocks, should yield about 30-35% performance improvements over 8-core BD (in gaming, that is, provided games will be even more multicore/-thread happy). Now, Sandy-E will bring, IIRC also, up to 6-core HT-d CPU-s. Now, on paper at least, it should bring at least 50% increase in performance provided linear scaling (lets leave aside the ~1K+ USD pricetag it'll prolly have).

So, even if somehow BD takes the throne of the fastest (gaming) CPU this fall/year, AMD will have tough competition from Sandy-E next year...

Has Amd come out and said how much better memory bandwidth will be with BD?

 

[TheMailMan78]

Has Amd come out and said how much better memory bandwidth will be with BD?

1866 with "wider channels".

 

[cadaveca]

With the new Bulldozer-based Opterons, which are set for release in the third quarter, AMD is introducing TDP Power Cap, which will give enterprises the ability to set the TDP (thermal design power) of their processors, essentially customizing their chips to meet power and workload demands. Using various knobs in the BIOS, businesses will be able to reduce the overall TDP of the chip—they won't be able to increase it beyond the maximum level set by AMD—which will help in power consumption, and then tweak the frequency of the cores as needed to get the maximum amount of performance allowed under the TDP setting, Kerby said.

"While you set the [TDP] cap, you can still operate at a high frequency," he said.

In addition, businesses can keep the TDP at the level set by AMD, and change the frequencies of the processors to add power, while keeping the overall power use under the TDP.

http://www.eweek.com/c/a/Desktops-a...s-Will-Feature-TDP-Capping-Technology-834387/


uh....

 

[TheMailMan78]

http://www.eweek.com/c/a/Desktops-a...s-Will-Feature-TDP-Capping-Technology-834387/


uh....

Thats for the server chips. Not the FX.

 

[cadaveca]

But hte power gating shoud be in ALL Bulldozer chips, AFAIK.

Might also explain performance problems with ES chips. If they are gated to only allow so much power consumption, it's possible that while benching clocks are dropped, unnoticed by whoever is running the benches.

 

[TheMailMan78]

But hte power gating shoud be in ALL Bulldozer chips, AFAIK.

Might also explain performance problems with ES chips. If they are gated to only allow so much power consumption, it's possible that while benching clocks are dropped, unnoticed by whoever is running the benches.

If thats true we have some fail on our hands.

 

[Thatguy]

If thats true we have some fail on our hands.

Fial on the part of the people benchmarking engineering validation samples ? absolutely.

 

[repman244]

But hte power gating shoud be in ALL Bulldozer chips, AFAIK.

Correct, it will be able to shut down and entire module which is not in use.




I don't think the desktop version of Bulldozer will have that TDP cap (well basic version will be - Turbo core stuff, and even if it does have you can control it if I read the post about it correctly), and maybe the ES have the feature (but the question is, who the hell needs that on ES for desktop part). So I'm guessing that's enterprise only.

 

[trt740]

Gigabyte pisses me off there is no way that this board couldn't handle the Bulldozer look at it's power system 8+2 and 140 watt last Gigabyte board I will buy.

http://www.gigabyte.us/products/product-page.aspx?pid=3010#ov

 

[Mussels]

Gigabyte pisses me off there is no way that this board couldn't handle the Bulldozer look at it's power system 8+2 and 140 watt last gigabyte board I will buy.

http://www.gigabyte.us/products/product-page.aspx?pid=3010#ov

the pins have changed in the socket, remember...

 

[trt740]

the pins have changed in the socket, remember...

No it fits in AM3 boards it's just a bios flash thats why it fits in a 770 board or a Crosshair and this shit will support it.http://www.gigabyte.us/products/product-page.aspx?pid=3807#ov The difference in this socket is the color of the plastic on this socket. It is a AM3 socket with black plastic and a bios flash. Atleast that how it was explained to me but we will see. Please feel free to correct me if I am wrong.

 

[Deleted member 67555]

Revision 3.1 supports it...the others do not

 

[trt740]

Revision 3.1 supports it...the others do not

revision 3.1? explain

 

[Mussels]

revision 3.1? explain

they keep releasing new revisions of the mobos with the black sockets, and claiming those ones are BD compatible. I think this is where people are getting confused, because you can have a revision (whatever) version of the board with the black socket and the beta BIOS are for THEM, and people are thinking those BIOS are for older revisions as well, when they arent.

 

[Wile E]

Not trying to start an argument, but your wrong.

http://tecinfozone.blogspot.com/2011/03/asus-am3-motherboards-get-beta-bios.html

Here's the info from Asus themselves.

http://event.asus.com/2011/mb/AM3_PLUS_Ready/

Hmmm, I stand corrected. I wonder why AMD is trying to force the move to the AM3+ socket? Sounds a lot like the same stuff people bitch at Intel for.

if your cooler uses the stock clip mechanism, they're still compatible. if you're replacing it completely, then yeah the screw holes are different.

The discussion was about water blocks, and erocker said mounting holes specifically. That's all I was commenting on. I do believe you are correct about all clip on heatsinks tho.

the pins have changed in the socket, remember...

The links Smooth posted above seem to suggest otherwise.

 

[Damn_Smooth]

Hmmm, I stand corrected. I wonder why AMD is trying to force the move to the AM3+ socket? Sounds a lot like the same stuff people bitch at Intel for.

From what I've read, Bulldozer will run on AM3 but not all of the power gating features will work. I don't think turbocore, or cool n quiet will work on AM3.

That's not a big deal to us because we would shut the shit off anyway, but I can see why AMD won't officialy support it. I'm not forgiving Gigabyte though.

 

[Pestilence]

Hmmm, I stand corrected. I wonder why AMD is trying to force the move to the AM3+ socket? Sounds a lot like the same stuff people bitch at Intel for.

Intel changes sockets because staying with the same socket leaves performance on the table. Look at Lynnfield vs Sandy Bridge. The swap to a new mb was WELL worth the gains. I had a 4.4Ghz 760 and i still swapped to SB because i like to waste money.

 

[repman244]

Intel changes sockets because staying with the same socket leaves performance on the table. Look at Lynnfield vs Sandy Bridge. The swap to a new mb was WELL worth the gains. I had a 4.4Ghz 760 and i still swapped to SB because i like to waste money.

Sorry but IMHO they just removed the pin to force people to buy motherboards, I imagine the gains did not come from the mobo but from CPU.
Same thing we have here with AM3 to AM3+, I bet if "AMD officially supported" AM3 for Bulldozer, everything would work just fine. I don't see any huge differences between the two (there are even boards with 890FX chipset and AM3+ socket) and it's not a weak VRM issues here, boards like CHIV and such can provide more than enough power IMHO.

 

[seronx]

5.1GHz on Air....OBR says the proof video will come soon

He's trolling so trollin'

Look, this is amazing! Fully stable in Cinebench R11 @ 4,85 GHz! SuperPi 1M - 5,1 GHz on AIR ... proof video:

 

[repman244]

I really don't see the point in those "leaks" that he does.
He is just showing some frequencies, but we all know frequency is just part of the story.

And I bet the video will feature CPU-Z showing the CPU and blocked scores of benchmarks...boriiing and already seen.

 

[Damn_Smooth]

http://3.bp.blogspot.com/-8e20pCOODxA/ThAbzbYfQkI/AAAAAAAAA1Y/CN1VPrwI6Bk/s1600/newska1.png

5.1GHz on Air....OBR says the proof video will come soon

He's trolling so trollin'

I would not complain at all if that were true, but I still don't trust that guy.

 

[repman244]

Me neither, remember when he posted those Llano screens showing it run at ~4.7GHz?
Well read this: http://www.xtremesystems.org/forums...lano-OC-amp-benchmark-collection-thread/page2
Especially post 47.
I think OBR is full of shit, but that's just me.

 

[repman244]

Here we go:

 

[seronx]

Here we go:

http://1.bp.blogspot.com/-aJTldEb1Kms/ThApx9sveSI/AAAAAAAAA14/dvIF3YKov_8/s1600/cine11.png-oc.png
http://3.bp.blogspot.com/-roteoiOV-N0/ThApvJpAcEI/AAAAAAAAA1w/7TDEK93UsQY/s1600/superpi_oc4.png

Those black things all over that are so enthused in hugging those numbers

I don't really care for the benchmarks...since I only really need SSE4.1/4.2/5 and AVX

I don't care about Integer Performance at all it could be equal to Phenom II for all I care as long as I get SSE4.1/4.2/5 and AVX

Ya, his ES Sample is multiplier locked....I did calculations on what I know about the architecture
and ya....his version isn't a black edition

3.520GHz