Here's a little teaser....

[dave_graham]

Has ANYONE run 3d06 yet? This is pretty lame

can't. 3992 doesn't support pcie graphics cards. the only board that is capable are the 3600-based boards (2915-E, 2932-E, etc.) right now and....good luck getting one.

I don't think anyone has anything capable of running it here. There are workstation boards, with onboard video. I think they're at best running XGI Volaris...

3DM06 has nothing to do with new CPU architectures, and a Barcelona system will most likely not be indicative of either Phenom or QuadFX. Between the errata issue, ECC ram, and workstation chipsets there's little point in pursuing that one...

ATI ES1000 chip here with a 32mb buffer chip. lemme tell ya...it ROCKS!

for amd's sake, i hope the chips and boards it sold to their actual customers function. cause, if their multi billion dollar clients are unhappy with how their new servers are running, amd is going to get eaten alive...

i think the manufacturers that actually care about this are rolling their own (dell, sun, hp, ibm).

Not sure but I think that a lot of sales are for the entire server. I'm sure AMD will make sure they run before they get to the customer. They will probably also have a lot of drop in sales, just exchanging cpus. For those I'm sure the AMD field reps will be johnny on the spot and will undoubtably have made sure the configuration will run before they exchange cpus

sure.

Get not integrated graphics?

not possible at this time, but patience.

Mate stop being so close minded.

K10 is the main arch, most that AMD can do on phenom to improove performance (IPC) is to increase the mem controler speed and that is it.

clock speed, memory controller speed, silicon fixes (which are minor percentage points, but, still....it does help), etc. platform DOES have an impact esp. as traffic is routed through the SB.

dave

[metro.cl]

can't. 3992 doesn't support pcie graphics cards. the only board that is capable are the 3600-based boards (2915-E, 2932-E, etc.) right now and....good luck getting one.



ATI ES1000 chip here with a 32mb buffer chip. lemme tell ya...it ROCKS!



i think the manufacturers that actually care about this are rolling their own (dell, sun, hp, ibm).



sure.



not possible at this time, but patience.



clock speed, memory controller speed, silicon fixes (which are minor percentage points, but, still....it does help), etc. platform DOES have an impact esp. as traffic is routed through the SB.

dave

Dave i said IPC that doenst count clock speed

[dave_graham]

Dave i said IPC that doenst count clock speed

IPC = Instructions PER CLOCK. so, given that clock speed is a necessary function of the Memory Controller speed, yeah, that'd be an important increase.

dave

[s7e9h3n]

Not sure but I think that a lot of sales are for the entire server. I'm sure AMD will make sure they run before they get to the customer. They will probably also have a lot of drop in sales, just exchanging cpus. For those I'm sure the AMD field reps will be johnny on the spot and will undoubtably have made sure the configuration will run before they exchange cpus

The components I used with my chips were configured by AMD and assembled by Colfax (the company whom AMD has contracted out to assemble all of their custom rigs built to spec). If my setup isn't considered an "AMD built" product, then nothing is

[s7e9h3n]

Whatever you do, try to monitor all core frequencies if you can (RM CLock might give more accurate real time frequencies). I personally know if PLLs and BIOS are erring the frequencies shown in CPUZ or most software can be wrong, Franck and Samuel will also tell you this as many C2Ds also showed this problem. It seems these cores on these early platforms love running at different frequencies though.

Use AMD's "Power Monitor" utility found here:

http://www.amd.com/us-en/assets/cont...r115_setup.zip

[s7e9h3n]

can't. 3992 doesn't support pcie graphics cards. the only board that is capable are the 3600-based boards (2915-E, 2932-E, etc.) right now and....good luck getting one.

Wait a sec, doesn't your mobo have 2 pci-e slots? And they both don't support video cards?

[tsuehpsyde]

The box I have to work with is a test box for work, which they've been so nice as to let me play around with it in Windows and do some testing. However, taking it off the shelf and tossing hardware in there (which would require bringing my own PCI-E videocard in, assuming it would jive with the 8x PCI-E on the board I have access to) isn't really a viable option unless I get the okay, and even then it would have to wait until Wednesday at the lastest (my first day off in 6 days). And quite frankly, I'd like to have some time off after almost working a week straight.

That, and I question the stability of the machine itself due to the problems I had, and I have a feeling 3D would give me the same problems. However, I'll see what I can do about 3DMark01 today by installing DirectX...

Ninja Edit: The machine is showing up in the properties tab as (Default Monitor) on *blank*, so I can't get into the advanced options (button is grayed out), and I can't find the actual video drivers needed to get it working. I'll keep at it...

[tsuehpsyde]

Use AMD's "Power Monitor" utility found here:

http://www.amd.com/us-en/assets/cont...r115_setup.zip

This shows all 8 cores at 1900MHz and the NB at 1400MHz.

[KTE]

Wait a sec, doesn't your mobo have 2 pci-e slots? And they both don't support video cards?

Thats what the specs suggest: http://www.tyan.com/product_board_detail.aspx?pid=235

[s7e9h3n]

Thats what the specs suggest: http://www.tyan.com/product_board_detail.aspx?pid=235

lol...maybe I should learn to do some research before asking my next question

[JohannesRS]

btw, could someone please tell me why NOT all tyan socket F MBs are showns as "quad core compatible", despite all showing the new opteron logo?

[tsuehpsyde]

CPU-Z Latency:



In other news, I moved the 2347s over to another identical rig, but instead of 8 x 2GB PC2-5300, it had 4 x 1GB PC2-5300. The other rig had a set of 2218s in it and was working fine with the same BIOS. Tossed in the Barcelonas, and it wouldn't boot unless I lowered it to 266MHz. So the IMC or the BIOS is to blame, but I won't be getting any 333MHz benches sadly.

[JohannesRS]

for god sake...

or the IMC will kill the K10 so bug it is... or the bios will be cleared up to show a best performance ever, cause being the AMDs so dependent on memory speeds and bringing those results with so slow memories, it must be great at proper timings. :P

[s7e9h3n]

CPU-Z Latency:



In other news, I moved the 2347s over to another identical rig, but instead of 8 x 2GB PC2-5300, it had 4 x 1GB PC2-5300. The other rig had a set of 2218s in it and was working fine with the same BIOS. Tossed in the Barcelonas, and it wouldn't boot unless I lowered it to 266MHz. So the IMC or the BIOS is to blame, but I won't be getting any 333MHz benches sadly.

Wow, what's up with that vcore on the cpu? On a better note......CONGRATS on the world's first Barcelona OC ! Almost a whole 10mhz

EDIT: BTW, you have memory divider options in bios?!?!? What other [interesting] settings are there?

[tsuehpsyde]

The voltage is being reported incorrectly from what I can tell. The BIOS and that AMD power program both show something completely different (and much lower). Don't pay it any attention. I honestly think CPU-Z is telling lies about my HT Link as well.

And fear the 10Mhz OC! Rawr!

Getting Sandra stuff now for Ozzimark's satisfaction, will post those when I finish up, as well as Everest.

[tsuehpsyde]

And some more....









The latency from everest was the biggest wtf. My first run was actually 27.7ns. I said no way, and re-did the test and got 29. Then I tested and got like 28.1, 33, 36.3, 32, 29, etc. I had to run it a few times and managed 27.9ns, but I did get 27.7ns on my initial test, which is a bit curious...

[perkam]

Mate stop being so close minded.

K10 is the main arch, most that AMD can do on phenom to improove performance (IPC) is to increase the mem controler speed and that is it.

I thought Phenom (desktop) is clocked at 2.0ghz-2.6ghz+ ??

Perkam

[Lightman]

tsuehpsyde your Everest latency test is amazing! My personal best on AM2 platform with DDR2 1000MHZ 4-4-4-12 was around 37.9ns!
On the other hand SiSoft Sandra test are a bit disappointing and from memory tests I'm thinking your Windows doesn't know what NUMA is .

Of course many thanks for sharing! Every input is important, even the one witch bad results!

[KTE]

Yeah that's AFAIK too, perkam.

TechReport got a latency of 91ns at 8192KB block, 512-byte stride (for a 2GHz part), whereas tsuehpsyde, you my friend got 208ns.

That L1 and L2 latency is nearly exactly the same as a 1GHz K7. Everest latency is diametrically very low though, although it isn't the highest I've seen.

That pretty much sums up all of the rest of the scores too. Lowest bandwidth scores I've seen yet.

Can you run the Sandra multi-core efficiency test please. I can imagine if only one socket is doing the work here.

[tsuehpsyde]

Sure, I'll go run the multi-core test now. And yes, even if these scores suck due to issues, at least they're something to look at and compare against other stuff for reference. Anything I can do to help out.

[tsuehpsyde]

And here she is:



Pretty flat footed, don't ya think? :p Maybe CPU-Z is actually showing the HT Link speed correctly. Not sure what else would be causing such low cores across the board.

[Motiv]

Very strange results. Any chance of the SPI 1M so we can have a comparison against Daves 1.9 chips. Might give us some indication on how off the motherboards are as he had 42 seconds.

[tsuehpsyde]

Very strange results. Any chance of the SPI 1M so we can have a comparison against Daves 1.9 chips. Might give us some indication on how off the motherboards are as he had 42 seconds.

Go back and look, already did one. :p

[Motiv]

Go back and look, already did one. :p

I swear to god I've looked and looked, I just can't find it. I am very tired tho

[informal]

Now we are 100% sure something funky is going on with all 3 rigs in question here .