Hypertransport-100 vs 2000

Hansfragger · 04-21-2006, 05:57 PM

Hello. I am deciding between an Opteron 150 and a A64 4000+. The specs list the HyperTransport frequency as:
Opty 150 - 1MHZ.
A64 4000+ - 2MHZ.

This difference is the only thing I haven't seen anyone discuss. What diff if any would this make in gaming?
I will be OC'ing this proc with a ASUS A8N32 SLI and OCZ PC4200 2x512 with liquid cooling. Thanks in advance!

(I missed the typo in the title-sorry)

tsuehpsyde · 04-21-2006, 06:05 PM

First, I'll assume you mean 1GHz versus 2GHz.

And just so you're aware, they both use the same speed. Confused? Well, the Opteron was listed as using the actual speed of 1GHz, while the 4000+ was listed as using the effective speed of 2GHz (it's double pumped).

So they're using the same thing.

Hansfragger · 04-21-2006, 06:17 PM

Thanks. I try to not let any typos slip by, but I'm a little beat after a long day. Since they are the same performance, I can now make my decision. I am going for the Opty. I appreciate your quick response.

bender222 · 04-21-2006, 07:07 PM

Quote:

Originally Posted by tsuehpsyde

First, I'll assume you mean 1GHz versus 2GHz.

And just so you're aware, they both use the same speed. Confused? Well, the Opteron was listed as using the actual speed of 1GHz, while the 4000+ was listed as using the effective speed of 2GHz (it's double pumped).

So they're using the same thing.

I was not aware that it was double pumped, i thought it was full duplex. the old Ht was 800 mhz both ways hence 1600 mhz total and the new one is much the same but 1000 mhz both ways.

tsuehpsyde · 04-21-2006, 07:14 PM

Yeah, I think it's full duplex. It's still 2GHz effective.

Furen · 04-21-2006, 07:25 PM

It's full duplex AND double pumped, so it does have an "Effective" (how I hate these) 2GHz rate in each direction. Of course it's only 16-bits wide so still quite "inferior" to Intel's 64-bit, quad-pumped FSB (In bandwidth, though Intel needs to move memory traffic through its FSB while it it remains on-chip in AMD chips).

IvanAndreevich · 04-21-2006, 07:32 PM

Quote:

Originally Posted by Furen

It's full duplex AND double pumped, so it does have an "Effective" (how I hate these) 2GHz rate in each direction. Of course it's only 16-bits wide so still quite "inferior" to Intel's 64-bit, quad-pumped FSB (In bandwidth, though Intel needs to move memory traffic through its FSB while it it remains on-chip in AMD chips).

Ok man. It's a bidirectional 1 GHz link with a base frequency of 200 MHz. It's not "double-pumped".

You have to know absolutely nothing about the subject to say that HTT is inferior to Intel's FSB

In reality, Intel's FSB is way out of date. Their next big thing will be similar to Hyper Transport in a couple of years or so.

ozzimark · 04-22-2006, 02:14 AM

Quote:

Originally Posted by Furen

Of course it's only 16-bits wide so still quite "inferior" to Intel's 64-bit, quad-pumped FSB

then tell me why i can drop the HT link speed all the way down to 400mhz from 1300mhz with NO impact in performance

the HT link is way more than we really need for most cases... only an exceptionally demanding set of video cards even come close to utilizing it..

matt samoohi · 04-22-2006, 12:33 PM

its 2ghz projected as the other guys said.

Furen · 04-22-2006, 04:07 PM

Quote:

Originally Posted by IvanAndreevich

Ok man. It's a bidirectional 1 GHz link with a base frequency of 200 MHz. It's not "double-pumped".

You have to know absolutely nothing about the subject to say that HTT is inferior to Intel's FSB

In reality, Intel's FSB is way out of date. Their next big thing will be similar to Hyper Transport in a couple of years or so.

Wow.

First off, the "base clock" means absolutely nothing. The HT link operates at 1000MHz REAL clock. It also transfers on the rising and falling edge of each clock, hence the "double-pumped" thing. So you get 2GT/sec in each direction, since it IS full duplex, can we agree on this? Now, multiply by the width (16bits in each direction) and you get a aggregate bandwidth of 8GB/sec, which is the number AMD usually quotes.

Now let's look at Intel's FSB. 266MHz, quad-pumped with a 64bit width. Because of the quad-pumped thing you get an effective rate of 1066MTs/sec, each transfer being 64bits, so you get a total bandwidth of 8.512GB/sec. This is, of course, not counting the 333MHz FSB that Woodcrest will be using. I know HT has been shown to work at higher clock speeds and in wider widths, but there's no CPU out there that actually uses it above 1GHz at 16bits.

About the "inferior" thing (NOTICE I put it in quotes and said that this only applied to bandwidth). Intel can throw more data through its FSB than AMD can throw through its HT links, that's all I meant. Of course, since AMD's memory controller is on-die AMD does not need to move memory transfers off-chip, while Intel has to pass its memory transfers through the FSB. Most of the traffic that passes through Intel's FSB, in fact, is memory traffic, which is why Intel gains so much from just increasing the FSB clock. So Intel's FSB can transfer MORE DATA but AMD doesnt need to, since only IO traffic has to touch the HT link.

HT is better because it transfers more data per pin compared to Intel's parallel FSB and is a royalty-free technology (which makes it cheaper to implement, this is especially true in multi-socket environments), but the HT link would be a bottleneck if the memory controller was not on-die.