anyone ahve any thoughts on TERA there supposed to be kicking SGI butt.. here is a thread of converstaton between tera and "buyer" ..easy to tell who is TERA.....
Subj: Re: tera rebuttal info
Date: 97-06-11 17:48:38 EDT
On Tue, 10 Jun 1997
GG,
I'll butt heads with anyone over high-performance computing. It's a
subject near-and-dear to me. I'm not convinced that the MTA is
everything Tera claims. I'm also not convinced that Tera is a
viable company even if the MTA does work to spec. That's my opinion,
nothing more. Tera, of course, will have different opinions but
because I'm the "customer" and they're the "seller", they must
respond to each of my reservations or there's no sale, figuratively
speaking. The burden of proof is on Tera. That said, let the debate
begin. I would like to see their rebuttal.
Best regards,
Henry
> > OK, so one MTA CPU can do an integer sort
> > faster than a Cray CPU. Big deal.
>
> Ahem. It's faster than any other CPU in the world.
It's also more expensive than any other commercial CPU in the world.
I'll take a dozen Pentium-Pro or PowerPC CPUs running parallel over
a single Tera CPU. They may not be as fast but they're certainly
cheaper. And because they're so much cheaper, I can scrap the lot when
something more advanced comes along. In the end I'll do much better
than if I had put my entire equipment budget for the next five years
into one machine from a financially insecure, virtually unknown
company.
> > What's that got to do with multi-threading or multi-processing
>
> Everything.
Nothing. The sorting benchmark was performed in serial, not parallel.
Has Tera released any parallel benchmarks that I don't know about?
> > I'd bet that there are $10,000 multiprocessor
> > PowerPC's that can be programmed to parallel-sort integers faster
> > than a Cray.
>
> You'd lose. Look at the public results. The little multiprocessors
> are slow (in Cray terms) and they don't scale up. If such machines could
> be built so that they would scale to an interesting size (and performance
> scaled up with cost), then we'd be out of a job.
Considering how fast the PC industry responds (six-month obsolescence
cycles), how much time do you think Tera has left before there's a
cost-effective, scalable parallel desktop computer.
> > Cray CPU's aren't particularly good with integers but
> > they're second-to-none for floating point calculations.
>
> Where do you get these assertions? And why do you feel the need
> to share them with us? Don't you imagine we know what Cray
> machines can do?
Maybe I wasn't writing it to you. I don't know or care who you are.
Besides, you haven't refuted the assertion.
> Cray Research builds lots of machines. At the high end, they have
> 2 basic styles: the T3E and the T90. The T3E is a distributed-memory
> machine built out of a large number of Alphas. The T90 is a parallel-
> vector machine built out of a small number of vector processors.
> We see ourselves as competing with the T90. (The T3E and other distributed-
> memory machines aren't really in the same class.)
If you think the Tera MTA is competing with the Cray T90 you're deluding
yourself - it's possible to buy a working T90 right now. Tera is more
likely to find itself competing with the Origin2000. Will the MTA be faster
and cheaper than the competition's latest effort? It's a competitive market
and the competition is working hard too.
> > And speaking of multiprocessor PC's, have you noticed that they're
> > becoming more common? Have you noticed how cheap they are?
>
> Am I allowed to be sarcastic? No?
Sarcasm? From my perspective you're whistling past the graveyard. How
many MTA's did you say Tera has sold? None? Sorry, my mistake. How
many did you say were on backorder? Now how many T90's are already in
operation around the world? You have little room for sarcasm.
> >Unfortunately for Tera it seems that's the trend in the computer market
> >right now.
>
> It's _a_ trend. So is the internet, so is Java, etc.
The comparison is obtuse. The trend towards smaller, faster, cheaper
machines has eliminated a lot of computer companies over the years.
That's the point I'm making.
> The market we're aiming at is the one that is desperate for more power.
> So desperate, that they've been willing to buy a lot of junky machines in
> the past and then blow millions of bucks trying to get them to perform.
Once again, I agree. The high-end market is already buying T90's from Cray.
> > I know of quantum chemistry labs (that require massive CPU power)
> > that are doing their calculations in parallel on dozens of cheap, networked
> > Pentiums.
>
> If they require _massive_ CPU power, they won't be satisfied with dozens
> of Pentiums. That sort of machine just isn't in the same scale.
> Academic labs buy what they can afford. I'd love to sell 'em machines,
> but that market is very competitive, and it's better for us, as a company,
> to avoid competition. So we aim at the very high end where the only people
> who come close to being competition are Cray. And then we beat them
> by scaling to larger machines than they can build.
I won't argue that academic labs buy what they can afford. However, buyers
in the high-end market don't have unlimited equipment budgets either. Given
the size of the high-end market, it doesn't take much to saturate it.
Do you think the high-end market is large enough to sustain Tera? How many
computer companies have gone bankrupt targeting only the high-end market?
How long will Tera have to remain viable just to recoup its initial research
costs?
> And everybody uses parallelism. PEntiums use a limited form of instruction
> level parallelism. Crays use vector parallelism. Distributed memory
> machines use data parallelism. The cool thing about the Tera is that
> it can take advantage of all these forms of parallelism. Makes us
> general purpose _and_ fast.
Are you actually shipping those machines? How much do they cost and how
many have you sold so far?
> Tera's demonstrated that it can build a high-performance, shared-memory
> machine that can be scaled to very large configurations for a reasonable
> price. (I know it sounds like a lot, but compared with the competition,
> we're a better deal.) Multithreaded parallel processing is a solution,
> not a problem to be solved.
Tera hasn't demonstrated anything except the design of a CPU that sorts
integers really well. That's the only information that they've shared with
the investing public. I'll believe the problem of multithreaded parallel
processing has been solved when I see it in action.
> >If they don't solve this problem it would appear that MTA is a big waste
> >of money and not even the US government will buy it.
>
> The government will byy it, labs will buy it, and industry will by it.
> And all for the same reason: It offers great performance for reasonable cost.
Once again, I'll believe that when I see it. You remind me of the
engineers from Thinking Machines. They used to build computers too.
Best regards and good luck with the MTA. I hope it succeeds. |
