CompuServe Thread

#Prod'ness of Multitask'

14 messages in this thread
#43278From: Scott S.Dec 11, 1986 1:27 AM
'386 eh? You should check out the Benchmark example in the DL libs. If you upgraded to a 68020 you'd have better than a 80386…one can only imagine what the 68030 is like! Try the '150WINDOWS.EXE' program, it might change your view of the envelope on the maximum number of tasks that can be opened. (Only a simple example though) Thought it (your messages) was well said also…and said…and said… <Grin> Scott
#43283From: Don Curtis/SYSOPDec 11, 1986 1:48 AM
Scott, I think you replied to the wrong person….I think you wanted to talk to Patmore not me. Don
#43288From: Patmore DouglasDec 11, 1986 2:37 AM
]Scott, Thanks for the response. I haven't consulted the DLs yet – maybe you can narrow down my investigative options by specifying which library the comparitive bechmarks can be found (and maybe also the name of the file). I guess my assertion of the power and superiority (apparent or real) of the Intel '386, comes from ariticles written in various computer magazines about the possibilities offered by this chip. I believe the Computer Shopper had an ariticle in it: one, two or three months back, scoping the possibilities of the application of powerful artificial intelligent programs on a machine utilising this chip. AI machines have a reputation of requiring a lot of system power navigating through tremendous amounts of memory, and requiring large amounts of processing power. Byte's (magazine) special IBM issue described the many facets of the chip: on board MMU; pipelining (I believe it entertains six pipelines); the allowance of an 8086 environment running concurrently with the more advanced environment – a product of the chip's own innovative architecture; runs at 3.64 Million of Instructions Per second etc. I must admit that despite this, I haven't made any real technical comparisons between this chip and the 68020 or 68030. But I get the impression that, and my instincts tell me, this new Intel chip is overall superior to both. Of course I could be wrong, maybe you or someone else can give me specific reference to the `benchmarks you were telling me about. Regards. P.D.
#43294From: Todd WeissDec 11, 1986 5:30 AM
From what I've heard, the '386 is running at about 3 to 4 MIPS at 16 Mhz with an average of about 4 clicks per instruction. This "seems" to be significantly faster than the 68020 (don't know about the 68030 yet) as far as the speed/instruction. You can however get 68020's at 25 Mhz while "I believe" that the fastest '386 is 16 Mhz. Anybody with any spec's or docs are welcome to U/L them. I've read that document in the D/L also and I'm not positive it makes sense as the average instruction time seems to be considerably faster for an 80386. –> Todd
#43312From: Bela Lubkin/SysopDec 11, 1986 11:59 AM
The benchmarks in DL11:BENCH.DOC are not all that valid because they are with dynamic memory; you must compare the 80386 results with the uncached 68020 results. That makes them just about equal, at the same clock speed. I'd like to see the same benchmarks with cache on both processors, because real-world machines are either going to have cache or static memory. Anyway, the '386 >may< have faster instruction times (I don't know), but it suffers from less registers and a much less symmetric instruction set. It may execute instructions faster, but it's going to take more of them (and more memory references, which >cost<) to get anything done. In any case, if the '386 and the '020 are close, the 68030 will blow the doors off the '386. The 68881 already clobbers the 8087/80287, and the 68882 will be that much faster (2-4x, actually). The 80387? Haven't seen any references to it at all, except that '386 machine and add-in-board manufacturers are leaving a socket open for it… – Bela
#43423From: Tom CattrallDec 11, 1986 11:37 PM
Bela– Gee, I thought I had you all educated about the 386 vs. 020 and then you go say something like that! 🙂 If you remember from a while back, the IEEE Micro article has both dynamic and static ram test results. When using fast external memory there are no wait states and the effect is the same as an external cache memory. With the fast static memory the 386 is still somewhat slower for equal clocks. Substituting full speed 020's gives much more speed yet. Finally, the 68030 is predicted to be 2x a 68020. To summarize using the quicksort benchmark (the only real program in the list) gives the following table showing speed normalized to the 80386 being 1.00. The 680×0 times are with cache enabled. cpu speed static(45ns) dynamic(150ns) 80286 (16) 0.85 0.98 80386 (16) 1.00 1.00 68020 (16) 1.66 2.49 68020 (25) 2.59 3.90 68030 (25) 5.18 7.80 (predicted) Besides being a year or 2 later than the 68020, the 80386 is also definitely slower. Perhaps the 486 will catch up with the 68020. I lost your address. If you let me know I'll definitely copy this stuff and send it to you. Can't have you going around spreading these vicious rumors. 🙂 Tom
#43430From: Larry Phillips/ICUGDec 12, 1986 12:05 AM
Thanks again Tom. I had forgotten that these had been posted before. I am really getting tired of the sort of people that the IBM and clones seem to gather. They don't seem to care about anything nut Locust 123 and when they read an article on a '386, they just make the assumption that if IBM is going with it that it must be best and fastest. Regards, Larry.
#43464From: Bela Lubkin/SysopDec 12, 1986 1:20 AM
If you want to fight off these claims, your only choice is to type in the tables from that IEEE article and upload the works to DL11. <grin, but do it anyway…> I remembered being convinced by data that was for static memory, but when I looked in the DL, it wasn't there! I still have a nagging feeling that the '386 can be cached better than just hanging no-wait-state static RAM off the bus… but I am pretty well convinced that the 68020 beats the 80386. However, it would be very wrong to draw any such conclusions from DL11:BENCH.DOC alone. – Bela
#43316From: Don Curtis/SYSOPDec 11, 1986 2:31 PM
Patmore, They are in DL 11 as BENCH.DOC. Don
#43346From: Patmore DouglasDec 11, 1986 6:34 PM
Much obliged for your assistance again Don. Regards. P.D.
#43424From: Dave Haynie/C128 LandDec 11, 1986 11:43 PM
The 80386 is probably an OK chip, as far as Intel is concerned. They've finally bagged the 64K limitations (probably the last serious MPU designer to do so), and the on-chip MMU is nice. Its kind of missed on several things, though, when compared to the 68020 and the 68030. The 68020 has a reasonably large instruction cache on-chip. An on-chip cache will always out-perform an external cache, both from a cost point of view and from a "free external bus time" point of view. The 68030 extends this design with a (presumably) better instruction cache organization, a DATA cache as well, burst mode accesses to nybble-mode DRAM, and an on-chip MMU. The on-chip caches are a big win, since things happen internally, a cache miss doesn't slow anything down. This is not true of many external cache designs. Now, while you might argue that the on-chip MMU can offer the same advantages, that's not always that much of a concern. The only thing that really slows you down with MMU translations is in a virtual memory situation, in which logically addressed memory is not part of physically addressed memory. In either case, the physical memory must be restored from slow storage. External MMUs in a 68020 system are capable of completing translation in time for no wait state memory accesses. The main advantages of the on-chip MMU are (1) cost effectiveness, and (2) that creates an MMU standard for the processor you're using. Of course, a poorly designed on-chip MMU is a big loss. -Dave
#43592From: Patmore DouglasDec 13, 1986 8:03 AM
All, I will concede to the fact that I had read whatever articles I did, in the context of a 'closed IBM frame of thinking'; i.e. the discussions of articles concerning applications with the '386, and the technical specifications of the chip, were done without comparison to a competitive microprocessor, and with the implications of using that mpu in the same situation as the same subject chip. My major influence to the idea of the '386's superiority over the 68020 and even 68030, was determined by an article in Byte magazine's IBM special issue, page 89, which simply describes the subject chip (the '386), as being one of fascinating achievement; its technical superiority over previous Intel mpu's; a comparison between the power of this chip and a mainframe frame system (an IBM incidentally); comparative benchmarks between a system using this chip (running at 16Mhz) and the Vax-11/780 – Interestingly, the '386 shows a 72% increase in perforamance in executing the Dhrystones over its competitor, and in the October issue of Computer Shopper, a 14.28 Mhz 68020 turbo Amiga does a 50% outperformance over the same Vax-11/780 – Taking for granted that 14.28 is approx 11% slower than 16, this and other factors, should have indicated some suspicion of the fact that the 68020 and the '386 might at least be somewhat competitive if not only in this area of testing (the Dhrystones), but overall. NB. These comparisons of times are quite very vague and do not premise to provide a real case of one chip against the other: they just indicate a suspicion. (contiued……)
#43593From: Patmore DouglasDec 13, 1986 8:04 AM
The benchmarks in file Bench.doc does indeed provide some indication that the 68020, cached or uncached, is competitive with the '386 to the extent that the tables reveal. I will however be quick to point out there was uncertainty in my assertions of the Intel chip over the Motorola's, but I am glad to say that I believe the 68020 (and definetly the 68030) is probably the better of the two. You have all won a soul.
#43376From: Scott S.Dec 11, 1986 8:31 PM
Guess I was caught by my own laziness…Hopefully Patmore will read my response by following the thread. Scott