PDA

View Full Version : Thunderbird Benchmarks



paulcs
14th April 2000, 06:24
I don't know where he got them:

http://www.hardocp.com/articles/amd/tbirdbenches.html

Paul
paulcs@flashcom.net

Jammrock
14th April 2000, 09:08
I'll take 4! All on the same motherboard with 4 GB DDR RAM, a G800 64 MB DDR RAM version with a 15,000 RPM fiber channel hard drive, a SB Live! platnium, and a bag of chips.

Jammrock

------------------
Athlon 650, Biostar board, 128 MB PC133 (Crucial), G400 32 MB DH, SB Live! w/ Digital I/O, 10/100 NIC, lots of case fans, etc...

paulcs
14th April 2000, 13:46
I'll settle for one, although the aluminum/slot format may be short-lived, and, I've read, the socket/copper format will be introduced sometime in the future.

I would like a Thunderbird-based system, but the upgrade path appears short for the Slot A version.

Paul
paulcs@flashcom.net

Jammrock
14th April 2000, 15:48
All Thunderbirds and Spitfires will be 0.18 micron copper based chips. I guess the Dresden fab is getting enough yields that AMD can produce enough chips out of the one factory.

I would assume that when the final switch to copper is made that the Austin fab will be retooled to match the dresden fab.

As for slot vs socket. The Thunderbird/Spitfire chips are suppose to ship in both flavors sometime in the late Spring, early summer.

Jammrock

Dyre
14th April 2000, 15:52
I just hope the current GFD's will work with these new bad boys, as we all know how horriable FSB overclocking has been with athlons so far.... http://forums.murc.ws/ubb/smile.gif

paulcs
14th April 2000, 18:31
I swear, I just read yesterday that the first Thunderbirds would be aluminum and Slot A, then Dresden would kick in a short time later with copper.

Anyway, it appears Intel has got another chip shortage on their hands, with parts above 700 MHz to be hit the hardest. All the more reason for AMD to get the Thunderbird out a little early and make Satan look silly again.

Paul
paulcs@flashcom.net

paulcs
14th April 2000, 21:24
Pictures:

http://www.hardocp.com/news_images/2000/april_2000/thunderbird1a.jpg

http://www.hardocp.com/news_images/2000/april_2000/thunderbird2a.jpg

Paul
paulcs@flashcom.net

paulcs
15th April 2000, 15:38
I'm seeing a lot of conflicting information about the Thunderbird's die. I've read in several places that Fab 25 (Austin) will initially produce aluminum dies followed shortly by Fab 30 (Dresden) with copper dies.

I still think we all have to get ready for the transition to socketed CPU's and motherboards. It's going to happen. There is no sound engineering reason to keep the chip on a PCB if L2 cache is on the chip. It then becomes a marketing issue, and the lower manufacturing costs of socket CPU's will eventually win out.

If you upgrade motherboards and CPU's every four to six months, of course, it won't be a problem. http://forums.murc.ws/ubb/smile.gif

Paul
paulcs@flashcom.net

Greebe
15th April 2000, 20:45
I have got to say I'm not impressed! My Athlon runs proportionally = to it cept that it was run on a KX133 MB. Therefore the mem bandwidth benchmarks beat mine... and that's it!

Sandra 2000 registered
CPU benchmark 2383 mips
1141 mflops

CPU bandwidth 378 MB/s
433 MB/s

MM benchmark 2368 it/s
3959 it/s

Of coure I don't really feel like down grading my CPU to 750mHz (it's currently running at 850 1:3)

paulcs
15th April 2000, 21:11
I can only imagine the damage you'll do when you get your hands on one of them. http://forums.murc.ws/ubb/smile.gif

Paul
paulcs@flashcom.net

DentyCracker
15th April 2000, 22:07
It may very well be a "crippled" CPU. I was reading what Chris Tom ( www.amdzone.com (http://www.amdzone.com) ) was saying

Taz
16th April 2000, 11:21
I've been reading over at JC's, (www.jcnews.com/pc), that theres been a bit of a row between Kyle (HardOCP) and Chris (Amdzone) over the validity of those benchmarks. Probably best to wait for further benchmarks before making any kind of judgement.

Greebe
16th April 2000, 12:31
I've spoken with JC personally and we both agree that something is definitely wrong with the results on HOCP.

paulcs
16th April 2000, 13:03
I think part of the problem was the accusatory nature of Chris's first post:

"-X-Static let me know that Hard OCP has some supposed 750MHz Thunderbird benchmarks up. My instincts say these are fake."

That's pretty rough. It denotes an intentional fraud, and there isn't a lot of room for interpretation.

Paul
paulcs@flashcom.net

Himself
16th April 2000, 16:06
I can't wait for socket A to get rid of this POS SS7 motherboard. (Hi Gurm! http://forums.murc.ws/ubb/smile.gif) Actually, no, my motherboard does everything I ask of it, I just want a faster cpu. http://forums.murc.ws/ubb/smile.gif Wish there were competition in Athlon mb chips though, instead of just VIA.

[I think slots are the only way to go for SMP, other than that I agree, if you are only going to use one cpu, a socket is just as good, if you can cool it enough.]

JCholewa
17th April 2000, 14:33
First off, I want to point to
" http://www.jc-news.com/pc/bbs/index.cgi?read=20228 ", which is a thread discussing the validity of the images and benchmarks and wcpuid info. The root of the thread seems to throw every possible thing he can find against HOCP's case.

Personally, this is my opinion:

The wcpuid results are valid. CPUID codes are like table keys -- they'll spout out a string of simple letters and it's up to the program to determine, usually by way of a look-up table, what the characters mean. So wcpuid, for instance, would have no idea what the codes for Tbird's L2 are unless AMD gave H.Oda classified information and gave him permission to make that information public. Which is unlikely. http://forums.murc.ws/ubb/wink.gif

The image ... for the most part, it looks legit. There have been some assertions that the die looks the wrong size. According to fact, Athlon die is 102sqmm, I believe. According to rumour, Spitfire die is 99sqmm. Given AMD's general cache density, this should mean that Tbird will take around 115-120sqmm of die area -- let's round it and say it's about fifteen to twenty percent larger than the Athlon die. This should be a noticable amount, but I've been too lazy to compare and contrast.

I have two problems with the report. Firstly, Kyle hyped himself as the carrier of the first Tbird benchmarks (or first alleged Tbird benchmarks -- I do not mean it in a bad way when I say "alleged", btw) out there. There was a possibly reliable double usenet post showing benchmarks of the Spitfire and Thunderbird alongside Coppermine and Coppermine-128. Whether these are real or not, Kyle should probably have shown some discretion here.

Though I did think it was funny on a level of genius the way Kyle presented the benchmarks -- a plea for secrecy and quiet, then a huge "AMD THUNDERBIRD BENCHMARKS" 3D image link. LOL!!! http://forums.murc.ws/ubb/biggrin.gif

The benchmarks. I've seen regular 750MHz Athlon scores beating these reported Thunderbird scores. I cannot see the Thunderbird holding only same clock for clock parity with Athlon, given what I know of the microarchitecture. So, in my opinion, one of a few things are true:

(A) The benchmarks are hoaxes - this is less likely, because the photograph does look pretty real, and Kyle is not in the habit of lying to people, and Kyle said that his source is trustworthy

(B) Kyle is suffering from Thresh Syndrome - That's right! The part benchmarked is an early beta of the Tbird (maybe). Remember, Thresh's beta K7 benchmark preview (a preview which I enjoyed, I should note)? Very early silicon, and I was dismayed when the Thresh team pushed it as being a nearly production level revision in a followup article. So, maybe this is a Tbird silicon A1 with cache only operating at 64-bit width and stepped down to half clock rate. Who knows?

(C) Board BIOS Bad, Bro! - A fellow at my website's BBS was recently having a problem where his friend's Athlon-550 was performing like a Pentium 233. Yikes! What's up with that? Basically, he cleared the CMOS, and it started running like a mad dog (er, I mean, it started performing to spec). It's possible that the board on which the Tbird was tested does not have some necessary BIOS patch or driver enhancement, and this erodes the performance of the processor.

Er ... I can't think of any more. But you get the idea. I have no problem with Chris Tom's outright accusation of Kyle -- I mean, hey, that's often Kyle's own style! I am proud that Kyle tried to back up his report in a manner more fruitful than Thresh's's.

I'll try to do a comparative look at the die with a regular Athlon. But I think that the best way to find out is to wait for real Tbirds to come out, after which point we'll know what it should have looked like and how it should have performed. http://forums.murc.ws/ubb/smile.gif

-JC
PC News'n'Links
http://www.jc-news.com/pc

JCholewa
17th April 2000, 14:50
Jammrock: Some base info to confirm and enhance your post:

The Dresden HiP6L process drops effective gate lengths (Leffs) by a little more than 15%. The copper interconnects probably help a little, too. Atiq Raza, formerly Vice Chief Poobah of AMD, once suggested that AMD would get a speed grade or two of extra frequency out of this process, though this was a total WAG. But, in general, I agree with his assessment, and it's possible that it was even somewhat conservative.

The high clocked Tbirds will come out of Dresden. Low clocked Tbirds and Spitz will appear from Austin, though AMD will be slowly moving Austin capacity over to Flash memory (a very lucrative business), more and more as time proceeds on its course.

Dresden has already started processing wafers, and recently stepped up from 500 wafers per week to 600 wafers per week. Total capacity of Dresden will eventually reach between 5000 and 6000 wafers per week, and we'll see this level some time in year 2001.

I'm not doing the calculations for area now, so I'm just wildly guessing at 200 Tbird die per wafer in Dresden. Let's pretend yields after packaging and everything are 75%. That's 150dpw. At 600wpw, that's 90000 Thunderbirds per week.

Don't take that number as law. I'm probably exaggerating something, like yields or dpw or something. I'll figure it all out some other time. http://forums.murc.ws/ubb/smile.gif

FWIW, AMD is officially estimating the following ramp (numbers are, I believe, parts sold, not parts produced) of Athlon and its derivatives (Tbird/Spitz):

Q1: 1.2 million
Q2: 1.8 million
Q3: 3.6 million
Q4: 7.2 million

-JC
PC News'n'Links
http://www.jc-news.com/pc

[This message has been edited by JCholewa (edited 17 April 2000).]

paulcs
17th April 2000, 16:13
Kyle show discretion? But then he wouldn't be Kyle.

Now Kyle has his hands on a Voodoo 5. I wonder if he has drivers. http://forums.murc.ws/ubb/smile.gif

Paul
paulcs@flashcom.net

[This message has been edited by paulcs (edited 18 April 2000).]

Jammrock
17th April 2000, 16:57
Hey! If the real Thunderbirds are running within 10% of the posted benchmarks on HOCP, I'll be extatic.

Copper vs. Aluminium for the first Thunderbirds...from what I have been reading, Dresden is already pumping out Alpha/Beta chips using 0.18 micron dies and copper traces. Also, the rumor mill says that the yields per wafer have been exceptionally high, which means that Dresden could go full force in Thunderbird production any day. Assuming the rumors are true.

I have not read anything about Fab 25 (Austin) even making the Thunderbirds or spitfires. If Dresden can pump out enough chips to meet initial demand, I don't see why they need to produce anything in Austin until its retooled. If I find out anything new, I'll post.

Jammrock