Let's sort this out :-)
- DS10 - 466Mhz
- XP1000 - 667 Mhz proc. (now)
 - Specfp95base for the UP2000 is almost the same as for the DS20,
but this was measured under NT. Judging by the latest results in c't,
linux still is a bit better than NT, but it's just a tiny bit.
- No idea about the power supply - you'll want
the power supply together with the board anyway.
- we ordered a UP2000 (netherlands),
for 25.000 hfl (12.500$) with 2 procs/2mb cache,
512MB ram, reasonable disk (9GB SCSI) etc

Compaq c/fortran for linux certainly is a must for this box....
http://www.**-**.com/

Actually:
Has anyone got some numbers for the difference between
2/4 Mb cache ?

Christian

typo, that should read:
the same as for the XP1000

Christian

That depends strongly on your application.

-- g

Another question, if I buy Up2000 machine, should I use 600 W power
supply? The price of this power supply is resonable? Finally I am
wondering about the price of Up2000 machine in US or other country
outside Korea. Samsung doesn't open the price of Up2000 in Korea.

I will install Linux in 21264 and will use Compaq Fortran & C. Is it
possible in Up2000?

--
Lee, Bo-sung
Ph.D. Candidate in Aerospace Eng.
Seoul National Univ.
mailto:bs...@pvmcube4.snu.ac.kr

I was told to use 600W power supply. A 400W one should work but there
are no reserves, so the 600W is recommended (which is quite a bit more
expensive).
I got quotes for cases w/600W power supply: (1 Euro ca. = 1.045 USD)
Towers at 480 Euro
19" 3U rackmount: 860 Euro

Prices for complete systems (2x4MB Cache Alpha @667 MHz, 256 MB RAM)
about 10700 Euro

should be possible. No first hand experience. Is there anybody out there
running Linux on UP2000?

- Alex

Mainly, we should compare UP2000 to DS20. Both are dual processor
capable with 256-bit wide memory. The DS20 has slightly higher memory
bandwidth due mostly in part to the 200-pin SODIMMs (UP200 uses 168-pin
DIMM). DS20 max. memory is 4GB compared to 2GB for UP2000. UP2000 is
available in 667MHz with 2 or 4MB cache or in 750MHz with 8MB cache. By
the way, the UP2000 with 750MHz/8M provides the highest SPECcpu95
ratings ever, both int95 and fp95. The UP2000 will be available with
1GHz 21264, when this cpu is released. We believe that this will be the
first 1GHz Alpha platform.

The chipset used in both DS20 and UP2000 is the same, Tsunami. The only
difference is the memory controller chip.

Where the DS20 has the slight edge in memory bandwidth, UP2000 makes up
for this, and more, with the faster CPU. Testing to date shows that the
UP2000/667MHz is much faster than DS20 in most areas.

UP2000 was designed specifically for Linux, it does not run Tru64 UNIX
or OpenVMS. And, well we all know where NT is going. DS20 was designed
specifically for Tru64 UNIX and can run Linux. Our testing shows that
UP2000 is overall a better choice for Linux, than DS20 - in terms of
performance, support and most importantly, pricing.

Yes, although you can squeek by (barely) with a 400W supply, if you buy
a dual CPU with a couple of storage devices.. you should really consider
600W as a minimum. With 1 CPU and 1 or 2 devices, 400W is sufficient,
provided it is of high quality.

You can receive a quote on UP2000 from
http://www.harddata.com/alpha/dpx264.html
or, send me a note at chr...@harddata.com

You mean the Compac Fortran and C for linux? If so, yes.

--
Chris @ Hard Data Ltd.

Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.

What do you mean by 667 Mhz proc. (now). Originally I thought
that the XP1000 is always equipped with 500Mhz. All the offers
I can get are for that proc. Is the faster proc. just an upgrade
option? By the way what is the highest frequency possible
with the board in the XP1000.

Christian

--
Christian Hasse
Institut fuer Technische Mechanik
RWTH Aachen
http://www.itm.rwth-aachen.de/staff/hasse/Christian.Hasse.html

http://www1.compaq.com/pressrelease/0,1494,wp~85_2|ob~6804_1_1,00.html
was the first benchmark results reported with the 667 Mhz box.

Pit

You are correct. However, now you can get XP1000s at 667 mhz. They
just started shipping in volume, apparently.

-- g

XP1000 now (also) ships with 667 Mhz 21264a.

Not quite sure, IIRC the old XP wasn't based on tsunami chipset.
The new one OTOH probably is.

The DS,UP boards are
based on the tsunami chipset.
And Samsung already ran a 1Ghz sample in the DS20 board.

Current is 750 Mhz - at least that's what API says,
Dunno wether they actually ship that.
Rumours are that IBM is doing 833Mhz in copper -
but they defintely don't ship that yet.

Dito

    Yes, lets sort a couple of things out.

    The DS20 and the UP2000 are based on the Tsunami chipset.
    The UP1000 is not. It is based on the AMD IronGate chipset.
    This information is a http://www.alpha-processor.com

    Incorrect. The 1Ghz sample was run on a UP2000 at PC Expo
    in June of 1999.

                                                                                    mike

hehe , ok  ;-)

Let's put it this way:
Difference as compared to 21164 2/4 Mb ?
How would memory bandwidth and
the brainy 21264 factor into this equation ?

Christian

Oh well ....
On checking:
Actually it was a UP2000.

Christian

: > Bo-sung Lee wrote:
: >
: > > Hello Everyone.
: > > I heard Samsung released 667 Mhz 21264 CPU/Mainboard named Up2000
: > > recently. And Compaq now sells DS10/20, XP1000 with 500 Mhz 21264. In my
: > > opinion, 667 Mhz CPU is faster than 500 Mhz CPU about 25%.
: > > But my collegue said that if chipset or cashe doesn't support CPU Up2000
: > > may be slower than Compaq machines. Is it true? If someone tested two
: > > machines, let me know the results.
: > >

                               XP1000       UP2000        OS & compiler
                               21264A       21264A        performance
                               667MHz       667MHz        delta between
                               4MB L2       4MB L2        Tru64 Unix
                               Tru64 Unix   NT4 SP3       and NT4
                               ----------   -------       -------------
    SPECint95                  37.5         32.1          1.17x
    SPECint_base95             33.2         31.8          1.04x

    SPECfp95                   65.5         53.7          1.22x
    SPECfp_base95              53.9         49.0          1.10x

    SPECint_rate95             337          288           1.17x
    SPECint_rate_base95        298          288           1.04x

    SPECfp_rate95              590          478           1.23x
    SPECfp_rate_base95         485          436           1.11x

[snip]

--
 Hai  Vo-Ba

: > Christian Simmendinger <chr...@nat.vu.nl> writes:
: >
: > > Has anyone got some numbers for the difference between
: > > 2/4 Mb cache ?
: >
: > That depends strongly on your application.

: hehe , ok  ;-)

: Let's put it this way:
: Difference as compared to 21164 2/4 Mb ?

I would substitute for Greg here: That depends strongly on your application.

Since you would normally use different optimization flags to get
the optimal performance from 21164 and 21264[*], many important
characteristics of your code could change substantially between
21164 and 21264. This in turn makes blanket statement wrt the
performance meaningless.

Cheers,

/Serge.P

[*] If you are really serious about performance, you might also have
to use different optimization flags for 2Mb and 4Mb cache configurations,
particularly for computational kernels which can benefit from cache
blocking and loop unrolling. As an example, you can tune DGEMM-based
BLAS3 from netlib to use block size of 256. This will give you a nearly
optimal performance on a system with 4Mb cache - but will run an order
of magnitude slower on a system with 2Mb. You can also tune the same
code to use block size of 128 - in which case you'll loose may be 5%
on 4Mb system, but will run at exactly the same speed with 2Mb of cache.

--
home page: http://www.cobalt.chem.ucalgary.ca/ps/