R4000/100MHz or R4400/150MHz
Compared to R4400/200MHz
First, a description of the original query. Here is a comparison of the data from the two spec sheets:
Aug 1993 Sept 1994 3D Lines 640K 1M 3D Lines GouraudZ Depth Cued 275K 452K Tmesh, Flat No Z 250K 408K Tmesh, Gouraud Z, Lit 113K 181K Quads, FlatZ 84K 129K Quads, GouraudZ, Lit 43K 67K
Note that there were probably OS and graphics library improvements during the year which spans these two data sheets.
Comparison for XZ:
About 50% better. The difference between old XL and new XL is even more striking, more than 100% better (eg. 3D Lines jumps from 490K for R4000 to 1M for R4400/200. The other figures have similar increases).
Actually, I suspect that the older XZ figures were probably measured with an R4000/100. A rather odd thing to do when they could have used an R4400/150... but I notice that all the chip performance figures on the older sheet are 'estimated' (MFLOPS, MIPS, SPEC89, SPEC92); perhaps an R4400/150 wasn't actually available to run the XZ tests with.
Also, the figures will be underestimates now since I have no data for R4400/250 or R10000 in use with XZ. I know R10000 can be used with Extreme.
The Extreme figures don't change that much between data sheets (10% max, 3% in most cases).
Note that the Quads figures are from the original IMPACT home page (I keep copies :), so actually they ought to be higher (they were measured with an R4400/250, not an R10K).
So that's the mystery. Now for Dan's explanation; he gave me a copy of a post made by an SGI employee which should shed some light on what's going on. Here is that post, written by Brad Reger of SGI's ISD (Interactive Systems Division):
From: Brad Reger (email@example.com) 22 Dec 1994 16:29:19 GMT > > I have some some new Indigo2 boxes which, I am told, have > > the XZ graphics package. However, the hinv tells me that > > the graphics card is "GR3-Elan". How do I determine whether > > I have Elan or XZ graphics? > > GR3 is the XZ board. GR2 is Elan. GU1 is Extreme. While this may be true, I suspect it isn't very interesting to most people. What our marketing folks want to call things doesn't usually matter that much either except that it's the customer- visible name. To confuse matters, we in engineering (along with marketing) sometimes change things in hardware or software, but leave the marketing name the same. We'll never (consciously) remove a feature, but there have been times where we've added things and left the name the same. XZ is an example. Purple Indigo: Entry graphics has no hardware GE (it's done in s/w) XS graphics has 1 GE XZ graphics has 2 GE's Elan graphics has 4 GE's Original Green Indigo2: XL has no GE's (done in s/w again) XZ had 2 GE's EX has 8 GE's Later Green Indigo2: XZ has 4 GE's Some history (my interpretation): Purple Indigo came out with Entry graphics. Great machine. Great price. Graphics just okay. Most people wanted more bitplanes and the option of putting in a hardware Z-buffer. Elan graphics comes out in purple Indigo. Excellent graphics. Completely blew away anyone else doing desktop 3D graphics. Along with Elan, we came out with XS (1 GE) with optional hardware z-buffer and optional 24-bit. Now have high, medium and low graphics configs. Turns out the market really wanted 2-GE performance. One wasn't quite enough and 4 was too expensive. So we came out with the XZ. Now have the performance of 2 GE's, 24-bit, and standard hardware z-buffer. Great product. Sold lots of 'em. [this may have happened after the Indigo2 intro] Green Indigo2 comes out. Intro with high-end EX graphics (8 GE's and double the raster performance). Best graphics on the desktop. A quarter later we intro the XZ (2 GE) config and one quarter after that we intro the XL (0 GE) config. All three sell well. As competition heats up (and some things become possible in hardware) we realize the XZ needs a boost. We bump the Indigo2/XZ from a 2 GE config to a 4 GE config. Nothing else changes, including the name. We figure people will be happy to get more for the same price. Problem is, people get confused and ask lots of questions. Now on to 'hinv'. As far as identifying graphics configs, it uses a pretty simple algorithm. It first looks at the base graphics board (LG*, GR*) and then just counts the number of GE's it sees. Then it goes through a fixed mapping of #GE's to marketing name. GR2 is purple (or brown). GR3 is green. So you get GR2-XZ if it sees a GR2 board and 2 GE's. You get GR3-XZ if it sees a (green) GR3 board and 2 GE's. When it sees 4 GE's it wants to call that an Elan. So you'll get either GR2-Elan (purple) or GR3-Elan (green). Since we never had an Elan product in green, you now have to interpret this as the 4 GE version of XZ. Long-winded. Did I explain enough? -- Brad Reger firstname.lastname@example.org Hardware Systems Engineering Silicon Graphics - Interactive Systems Division (green & purple boxes)
The above explanation certainly clears up the mystery performance jumps as far as XZ is concerned, but that still leaves XL. My guess is that the simple use of a faster CPU and perhaps better graphics libraries is responsible for those improvements, though I'd be happy to hear from anyone who has more accurate information.
Thanks Dan and Brad!