Today's hardware announcements comparison

[David G. Hough on validgh]

Having trouble tracking all the new products?   I do.
Fortunately Sun marketing supplied the following from Dataquest.
I added the SPECfp92/$K ratios under comments :

Comparable System: 16 MB, 15- or 16-inch color, a 400 MB disk

System		Price	SPECint92	SPECfp92	Comments
----------	-----	---------	--------	--------------
HP 715/33	$ 7,395	  24.0		 45.0	   	6.1 FP/$K No EISA, 525 MB Disk	
HP 715/50	$14,095	  36.5		 72.1	   	5.1 FP/$K 19-inch color
SPARCclassic	$ 4,795	  26.4		 21.0		4.4 FP/$K
SPARCstation LX	$ 7,995	  26.4		 21.0		2.6 FP/$K
IBM 220		$10,885	  16.6		 26.1		2.4 FP/$K

Comparable System: 32 MB, 19-inch grayscale, a 400 MB disk

System		Price	SPECint92	SPECfp92	Comments
----------	-----	---------	--------	--------------
DEC 400/400S	$16,395	  62.6		107.0		6.5 FP/$K
HP 725		$17,895	  36.5		 72.1		4.0 FP/$K 525 MB disk
IBM Model 350	$22,895	  35.4		 74.2		3.2 FP/$K
IBM Model 340	$19,095	  27.9		 58.7		3.1 FP/$K	
SPARC 10/30	$18,495	  44.2		 52.9		2.9 FP/$K
SPARC 10/41	$24,995	  52.6		 64.7		2.6 FP/$K

Comparable System: 64 MB, 19-inch color, 1 GB disk

System		Price	SPECint92	SPECfp92	Comments
----------	-----	---------	--------	--------------
HP 735		$42,940	  80.0		150.6		3.5 FP/$K
DEC 500/500S	$42,995	  71.3		121.8		2.8 FP/$K
HP 755		$58,995	  80.0		150.6		2.6 FP/$K 2 GB min. disk
IBM 580		$72,582	  59.1		124.7		1.7 FP/$K 2 GB min. disk
IBM 560		$63,262	  43.2		 97.6		1.5 FP/$K
IBM 530H	$47,660	  28.5		 64.6		1.4 FP/$K
IBM 520H	$39,560	  21.5		 45.3		1.1 FP/$K


addendum: Dataquest omitted DEC high end; Sun didn't announce single-
processor SPECfp92 for SPARC 10/5x or 2000 since each has a minimum of 2 CPU's.

		base price
DEC 4000/600S	$70K	  84		143		1.8 FP/$K 150 MHz
DEC 7000/600	$168K	  97		182		1.1 FP/$K 182 MHz
DEC 10000/600	$316K	 107		200		0.6 FP/$K 200 MHz

The very high absolute floating-point performance of DEC's alpha systems
comes at a cost; code that wants to trap on floating-point exceptions 
and continue must be compiled with trap barrier instructions that slow things
down somewhat.   Programs that generate exceptions aren't allowed in SPEC,
since there is no standard way to specify what is to happen, and
hence SPECfp ratios don't reflect that cost.