nsw@cord.UUCP (Neil Weinstock) (03/18/89)
Well, I thought I'd play the "run home from the user's group meeting and post about what was presented" game. It was the March JAUG meeting. MASSIVE DISCLAIMER: The info below is impressions from a 1 hour sales pitch and demo. I am not affiliated with GVP in any way. I apologize if this posting is very commercial in nature, but I feel that this is a very significant product, and people ought to know about it. Well, then, let's get right to it. GVP introduced a 68030 card. Asynchronous. Initially in 16 MHz or 25 MHz versions, and they're working on a 33 MHz version. Socket for 68882 with its own clock. Also to be introduced is a 32 bit memory card with either 4 or 8 MB (which will naturally cost a bundle, memory prices being what they are at the moment.) It takes advantage of the 030's burst mode to run at 0 wait states most if not all the time (not sure of the details here). The 68030 *can* DMA into the 32 bit memory. The CPU card *has* sockets for EPROMS to allow UNIX to run. The CPU card also has, built in, a hard disk interface for some special variety of Quantum drive (not SCSI or ST506, something weird). This I found a bit odd, but hey. Some numbers? Well, they ran a Ronin benchmark which showed the 68030, running at 20 MHz (not a version they will be selling), motoring along at over 4 times the speed of a stock 2000. That's *without* any 32 bit memory. With the 32 bit memory, the benchmark came in at over 8x the standard Amiga. Major caveat here, we had little idea what the Ronin benchmark was doing exactly. But in talking with the GVP folks afterwards, they said that they were typically finding that those numbers were reasonable in their tests. They said that they were finding that their 25 MHz 030 board without 32 bit memory or floating point processor consistently outperformed a 2620 with 2 MB of 32 bit RAM. Make of that what you will. They demoed running a Mandlebrot set generator with it. Filled the screen in around 5 seconds. Anyway, this all sounds too good to be true. Maybe it is. But under any circumstance, GVP appears to have a real powerhouse board on their hands, one which would certainly make Mac II owners jealous. Oh yeah, you wanna know the price? For the 16 MHz 030, $849, and for the 25 MHz 030, $999. These are not typos. These are not introductory prices. 4 MB of 32 bit RAM will be ~$2000. Sounds like it would be worthwhile, though, to own the CPU without the RAM, at which point it becomes *very* affordable. Actually, the pricing here is so low as to be implausible. But it appears to be real. Did I mention that Max Toy was there also? What a meeting. Things are getting fun in Amigaland! /.- -- .. --. .- .-. ..- .-.. . ... .- -- .. --. .- .-. ..- .-.. . ...\ / Neil Weinstock | att!cord!nsw | "One man's garbage is another \ \ AT&T Bell Labs | nsw@cord.att.com | man's prune danish." - Harv Laser / \.- -- .. --. .- .-. ..- .-.. . ... .- -- .. --. .- .-. ..- .-.. . .../
daveh@cbmvax.UUCP (Dave Haynie) (03/21/89)
in article <741@cord.UUCP>, nsw@cord.UUCP (Neil Weinstock) says: > Keywords: GVP 68030 > Summary: Hold on to your hats, boys, it's party time. > The 68030 *can* DMA into the 32 bit memory. If the 68030 can't DMA into 32 bit memory, you've got a problem, since you'd expect the CPU to be able to access this memory. Perhaps what they're claiming is that 16 bit devices from the expansion bus have DMA access to 32 bit memory, as on the A2620. > They said that they were finding that their 25 MHz 030 board without 32 bit > memory or floating point processor consistently outperformed a 2620 with 2 > MB of 32 bit RAM. Make of that what you will. Either fiction, or they're being really careful about which benchmarks they run (for instance, the FPU only helps when you're running floating point code, not integer code or FFP-library code). Or perhaps they just forgot to run FastMemFirst when they compared themselves to the A2620 :-)? Certainly a 25MHz '030 with the data cache enabled (currently a dangerous thing to do if you've got DMA happening) will run the inner loop of some code faster than a 14.3MHz '020 with 32 bit wide FAST memory, if such innner loops fit in the cache. Another interesting thing about the '030 is, like the '020, it always does longword prefetched, but in this case, for both instruction and data fetches. What this amounts to is that the CPU always has to run two memory cycles to read from 16 bit RAM, even if it's only interested in a byte of memory. Which means that, if you take away all 32 bit memory and turn off all the caches, a 14.3MHz '020 will go slower than a 7.16MHz 68000, and certainly a 14.3MHz '030 will go yet again slower. The extra clock speed in this case won't buy you much, since the CPU will be running memory bound almost exclusively. The caches make up some of the difference -- a 14.3MHz 68020 on 16 bit memory runs about 20% faster with it's instruction cache on than the 7.16MHz 68000. However, give it fast 32 bit RAM, and it's going to run 200%-400% faster, possibly better, in an Amiga system. No question that this would blow away an A2620 if you hook up a 25MHz FPU and some real 32 bit RAM. But it'll end up costing about twice as much. > Anyway, this all sounds too good to be true. Maybe it is. But under any > circumstance, GVP appears to have a real powerhouse board on their hands, > one which would certainly make Mac II owners jealous. And there's a good target to aim for. An A2500 performs about the same, CPU-wise, as a Mac II. Even though the clock speed is slightly less, the Mac II makes it's '020 do so much extra system work, it's effective speed in many cases comes out less. An A2000 with this GVP board and 32 bit memory should run about twice the speed, roughly, of a Mac II. Lots of Mac fans know this too; you can buy similar accelerator cards for Mac SEs, up to 33MHz, that tend to make the Mac II look silly. > Oh yeah, you wanna know the price? For the 16 MHz 030, $849, and for the > 25 MHz 030, $999. These are not typos. These are not introductory prices. Also, without math chips. Which does strike me as being strange, since most accelerator board without 32 bit memory really need to include the math chip to give much additional performance. The data cache will certainly help, when you can use it (pretty safe if you're not running any DMA), or if you're not too concerned about it's safety. > Actually, the pricing here is so low as to be implausible. Not really. Consider an A2620 with 2 meg of 32 bit RAM goes for somewhere around $1500, math chip included. > Did I mention that Max Toy was there also? What a meeting. Things are > getting fun in Amigaland! I'm really glad that GVP decided to at least make a real 68030 board, instead of just another me-too design. > /.- -- .. --. .- .-. ..- .-.. . ... .- -- .. --. .- .-. ..- .-.. . ...\ > / Neil Weinstock | att!cord!nsw | "One man's garbage is another \ > \ AT&T Bell Labs | nsw@cord.att.com | man's prune danish." - Harv Laser / > \.- -- .. --. .- .-. ..- .-.. . ... .- -- .. --. .- .-. ..- .-.. . .../ -- Dave Haynie "The 32 Bit Guy" Commodore-Amiga "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: D-DAVE H BIX: hazy Amiga -- It's not just a job, it's an obsession