Heh, that seems rather crazy.
The V20 is the 8088-compatible version, so it is limited by an 8-bit bus and chipset. It's a few percent faster than a regular 8088 at the same clockspeed, but still way slower than a 286.
Even the V30 (8086-compatible, so 16-bit bus) will be slower than a 286. The 286 is a much more advanced and efficient CPU.

Those numbers seem pretty strange... the 8086 at 5 MHz is less than half the speed of an 8088 at 10 MHz. That doesn't make sense. That'd mean that even if we assume perfect performance scaling with clockspeed, the 8086 would be slower than the 8088.

i think you cannot compare mips numbers directly; their meaning for real performance depends on width of the machine word at least; i also think there will be machines which will do more with less instructions especially if were looking at a given set of problems to solve for that given machine;

mrau wrote:

i think you cannot compare mips numbers directly; their meaning for real performance depends on width of the machine word at least; i also think there will be machines which will do more with less instructions especially if were looking at a given set of problems to solve for that given machine;

I think the confusion is mainly because people take 'MIPS' literally.
In practice there was the standardized Dhrystone benchmark to measure 'MIPS' (integer) and Whetstone to measure 'FLOPS' (floating point).
These benchmarks were available in C source code. The idea was to compile it for the target system and run it, then compare results against other systems.
So you weren't actually measuring the number of instructions per second. You were measuring how long a known piece of code took to execute on a system. So the 'MIPS' were relative to a base system (I believe some PDP/11). The code was constructed to run ~1 MIPS on that base system.

LOOK: https://www.flickr.com/photos/94839221@N05/al … 157665135854734
LOOK https://www.flickr.com/photos/94839221@N05/al … th/22741784972/
LOOK: https://www.youtube.com/watch?v=-T_Ay3ZWYKQ

The NEC V20/V30 tend to confuse some benchmarks, as some instructions are blazingly fast compared to the Intel implantation, and those instructions are over represented in some bencmarks

I think a V20 8mhz should be even fast as a 286-6 mhz probably.

Robin4 wrote:

I think a V20 8mhz should be even fast as a 286-6 mhz probably.

Not a chance.
Why do people hype these V20/V30 CPUs so much? In a lot of software you can't even tell the difference between a regular 8088/8086 or a V20/V30.
A 286-6 is probably twice as fast as an 8088 at 8 MHz on average. A V20 would be a few percent faster than the 8088, depending on the software you run (think in the range of 10-20% on average).
Just look at the cycle counts per instruction: https://zsmith.co/intel.html
And for V20/V30 here: http://datasheets.chipdb.org/NEC/V20-V30/NEC_uPD70116.pdf
Take eg a multiply operation: https://zsmith.co/intel_m.html#mul
A 286 is 4-5 times as fast as an 8088 on those instructions, and still about twice as fast as a V20/V30.
It's in a completely different league.

It might have been a different story if Intel had released something like an "80288" CPU (with an 8-bit external data path).

Maybe this is a good opportunity for someone here on Vogons, who has a turbo XT, an NEC V20, V30 & a 6 or 8 MHz 286, to benchmark these CPU's, using some of the popular benchmark programs?

Scali wrote:

Why do people hype these V20/V30 CPUs so much? In a lot of software you can't even tell the difference between a regular 8088/8086 or a V20/V30.

I'm no psychologist, but I think it is because they love XTs,
which in turn is based on the idea that the 5150 (XT is 5160 actually) was the "very first PC" everrr.
So having a hot-rod PC is the holy-grail for some of them. And that's were the V20/V30 CPU thing is coming into play.
It's the cheapest, easiest to install and most available upgrade for this machine class.

Of course, that doesn't apply to all users. Some users do simply have a technical interest in older machines and are
thus fascinated by this oddity. Some also install these CPUs to have the enhanced instruction set found in the 80186/80286 CPUs.
For example, this can be handy for several zip programs or the Win 3.0 VGA driver. Compilers like Power C also have an 80186 switch.

In some way or another, this reminds me of the 8080 vs. z80 thing. The z80 was often used as a replacement for 8080 processors.
It wasn't pin-compatible, but slightly faster (hardwired functions), had a better instruction set and only required a single voltage.
Another cool feature was the integration of a dram circuitry, I think. Intel later realized this and launched the 8085,
which had similar enhancements but came a bit too late.

Anyway, it's funny that people do always mention the V-processors along with the IBM-PC, even though they likely weren't
even intended for it in the first place. Remember, NEC also had it's own computer series - the PC-88 and PC-98.
Whereas the former was an 8080/z80 (µPD780) based machine. So was it pure coincidence, that the enhanced PC98 got a processor
(V20/30) with an 8080-emulation mode ? I don't think they made that chip only for best CP/M compatibility..

"All PC98 computers will also run PC88 games. That's right! NEC worked very hard to be sure that the PC98 is backwards compatible
so you can run your PC88 floppies on it without any major problems. That's how I play my PC88 games, by the way."
Source: http://www.gamefaqs.com/PC98/934099-nec-pc98/reviews/155931

More about japanese computers can be read here.

(Note: We know that the IBM 5150 wasn't the first personal computer per se.
Altair 8800, IMSAI 8080 and several other machines before used that term, but sadly people don't care.)

Robin4 wrote:

I think a V20 8mhz should be even fast as a 286-6 mhz probably.

Not a chance.

I have a 10mhz v30, and the benchmarks I ran put it in the same league as a 10MHz 286. The Sierra games I threw at it denitely felt like AT speed. There is a pretty big speed jump from 8088 to 8086. I want to say about 50% faster.

Anonymous Coward wrote:

I have a 10mhz v30, and the benchmarks I ran put it in the same league as a 10MHz 286.

And how does a regular 8086 at 10 Mhz compare in those benchmarks?
Another issue with '286' is that it's rather meaningless. There is so much variation in chipset/motherboard performance from that class of computers, that you really can't generalize 286 performance from a single 286-based machine.
I have a late 286-20 machine, which blows early 286 machines completely out of the water with its performance. Benchmarks tend to see it as a ~30 MHz 286 equivalent.

wouldnt the biggest impact on all these cpu be external ram speed and wait states

BloodyCactus wrote:

wouldnt the biggest impact on all these cpu be external ram speed and wait states

That, and the type of code you run.
Eg, 8088 is slow at EA... V20 is fast at EA. If you optimize your code for 8088, you avoid slow EA. Therefore, code optimized for 8088 won't gain much on V20.

Anonymous Coward wrote:

I have a 10mhz v30, and the benchmarks I ran put it in the same league as a 10MHz 286. The Sierra games I threw at it denitely felt like AT speed. There is a pretty big speed jump from 8088 to 8086. I want to say about 50% faster.

Well, the original 6MHz PC/AT probably wasn't the fastest thing around. Slow set of chips, wait states, slow memory, etc.
It's quite possible that a more recent Turbo XT was less bottlenecked in this regard. But speaking of the CPU itself,
a 286 should be far more advanced (pun intended) than a 808x/V20/V30.

Scali wrote:

Another issue with '286' is that it's rather meaningless. There is so much variation in chipset/motherboard performance from that class of computers, that you really can't generalize 286 performance from a single 286-based machine.
I have a late 286-20 machine, which blows early 286 machines completely out of the water with its performance. Benchmarks tend to see it as a ~30 MHz 286 equivalent.

I support that. Some 286 machines are quite speedy and do even outperfom a 386SX at the same speed.
Can the V20 do something similar ? Possible, but unlikely. It has to perfom some kind of multi-plexing (16bit data on 8bit data bus),
like the 386SX has to do internally (32bit data on 16bit data bus).

BloodyCactus wrote:

wouldnt the biggest impact on all these cpu be external ram speed and wait states

I think so. But perhaps the benchmark doesn't care for this and just measures how fast certain instructions are processed
or how much cycles they need. NEC was quite good at optimizing these things, so it's possible that some insturctions are indeed
more efficient than on the 286. However, does this affect real world performance, aswell ?
I don't know. Better ask some expert like Trixter.

jesolo wrote:

It might have been a different story if Intel had released something like an "80288" CPU (with an 8-bit external data path).

They didn't in particular (except maybe 80188, which could be seen as a 286 minus protected-mode), but IBM created a mutant called PC/XT 286. 😄

jesolo wrote:

Maybe this is a good opportunity for someone here on Vogons, who has a turbo XT, an NEC V20, V30 & a 6 or 8 MHz 286, to benchmark these CPU's, using some of the popular benchmark programs?

Good idea. Should be amusing, could perhaps even end in some kind of pillow fight! 😉

Jo22 wrote:

but IBM created a mutant called PC/XT 286. 😄

The irony is that the PC/XT 286 uses zero-waitstate memory, where the PC/AT uses a waitstate on memory. This makes the PC/XT 286 considerably faster at the same clockspeed (which again goes back to my point of 'there's no such thing as 286 performance').

My experience with 8088-based systems is that there is far less variation in performance. Many clones use the original Intel 82xx-chips on their motherboard, making them cycle-exact to the real thing. Other clones may have integrated chipsets (such as the Faraday FE2010 in my Commodore PC20-III), but performance is near cycle-exact.
They probably deliberately designed XT clones that way. Most of them boot up in 4.77 MHz mode, even if they have a turbo. They were probably designed so that they are (almost) cycle-exact to a real IBM at 4.77 MHz, because this was important for compatibility back then.

Agreed, this would make sense for Turbo XTs. I guess best PC/XT compatibility was their main selling point (besides lower manufacturing costs and beeing affordable).
A lot of first generation programs used the IBM 5150 as a reference and weren't even aware of newer computer models.
However, some earlier clone models weren't advertised as iBM compatible, but just intended to be MS-DOS compatible (fine for Lotus 1-2-3 and WordStar).
I'm tinking of the Sirius 1 and other machines from the DOS 2.x era.. So perhaps some of them still hold one or more surprises for us. 😉
Anyway, these are just my thoughts and my knowledge about XTs is still lacking, so please anyone feel free to correct me..

jesolo wrote:

It might have been a different story if Intel had released something like an "80288" CPU (with an 8-bit external data path).

Maybe this is a good opportunity for someone here on Vogons, who has a turbo XT, an NEC V20, V30 & a 6 or 8 MHz 286, to benchmark these CPU's, using some of the popular benchmark programs?

Which benchmarks to run?

I recently replaced an 8088 10MHz for a V20, just for the heck of if. I never owned a V20 and I had the same questions about it.
Benchmarks and personal (subjective of course) observation show a difference of ~5%, which is nowhere near a 286. Software used was targeting XTs and that was my intention.
I believe that for many reasons, Scali pointed them out excellently, this is the correct comparison to be made when a V20 is involved. XTs vs 286 AT in general is another discussion for me, although ironically this is what this thread is all about 😀

konc wrote:

I recently replaced an 8088 10MHz for a V20, just for the heck of if. I never owned a V20 and I had the same questions about it.
Benchmarks and personal (subjective of course) observation show a difference of ~5%, which is nowhere near a 286. Software used was targeting XTs and that was my intention.
I believe that for many reasons, Scali pointed them out excellently, this is the correct comparison to be made when a V20 is involved. XTs vs 286 AT in general is another discussion for me, although ironically this is what this thread is all about 😀

True, but to be fair I can't blame them (or us). Only a few people have ever seen those CPUs in anything else than these machines.
The 286 was originally intended for somewhat more sophisticated tasks, I believe.
Think of data centres, telecommuncations, research and development and much more.
It is somewhat sad that this chip is nowerdays barely considered to be more than a footnote in history (as the "braindead chip").
And even more, only a few of them were ever allowed to run somehting else than DOS and got more than one lousy megabyte of memory.
Same applies for the PC/AT 5170 in some way or another. It introduced some many things we take for granted nowerdays,
yet only a few people even remember this model (and if they do, then often in a negative sense).
Or as some people say "Who does remember the second man on moon ?"