2000MHz Hyper transport???

2000HTT is for dual core CPU's, I believe. It should be 1000MHz hyper transport on that single core chip.
 
HighwayAssassins said:
2000HTT is for dual core CPU's, I believe. It should be 1000MHz hyper transport on that single core chip.
Click to expand...

nope

it's 2000MHz for Single AND Dual cores
 
so where is this double pumper is it in with the dual channel RAM? i'm a little confused on the wording too, but do i actually get 2000MHz for my Hypertransport speed or is it actually 1000MHz and they can some how word it to make 2000MHz?
 
soorma07 said:
so where is this double pumper is it in with the dual channel RAM? i'm a little confused on the wording too, but do i actually get 2000MHz for my Hypertransport speed or is it actually 1000MHz and they can some how word it to make 2000MHz?
Click to expand...
HyperTransport speed isn't relevant to Memory bandwidth. Its determines I/O Bandwidth.
 
coldpower27 said:
HyperTransport speed isn't relevant to Memory bandwidth. Its determines I/O Bandwidth.
Click to expand...
ok so dual channel must be incorrect
man where is this 2000MHz because it's not in cpuz not in my bios just seems to be in advertisments and if it is actually 2000MHz then why do the opterons only have 1000MHz for their Hypertransport speed wouldn't that make them worse?
i would really just like to see some proof that there is a 2x multiplier somewhere giving me 2000MHz rather than the 1000MHz i see now

thanks for the replies
 
soorma07 said:
ok so dual channel must be incorrect
man where is this 2000MHz because it's not in cpuz not in my bios just seems to be in advertisments and if it is actually 2000MHz then why do the opterons only have 1000MHz for their Hypertransport speed wouldn't that make them worse?
i would really just like to see some proof that there is a 2x multiplier somewhere giving me 2000MHz rather than the 1000MHz i see now

thanks for the replies
Click to expand...
cpu-z doesn't report HT link speed.

it's just like ddr.. 200mhz ram is 400mhz effective because it sends data on the rising and falling edge of the clock signal. the HT link runs at 1000mhz for 939/AM2 and 800mhz for 754. due to the "double data rate" technique, it's effective bandwidth is the same as a 200mhz single data rate or 1600mhz single data rate bus would be. just stupid marketing trying to confuse
 
Its 1000MHz HyperTransport.. the 2000 you guys are seeing is the 2000 MT/s. ;) The HyperTransport is only 1000MHz. Its 1000MHz upstream, and 1000MHz downstream. (chipset to CPU) Its not doubled. www.hypertransport.org
MT/s - Megatransfers per second. A method of measuring the throughput of
the Direct Connect Architecture HyperTransport ports. 2000 MT/s
corresponds to a 1000 MHz HT clock speed, 1600MT/s is a 800Mhz HT clock speed.
Click to expand...
 
central brain identifier reports the HTL speed. It shows mine as 900 (running at 300Mhz HTT). Under the Cache/Memory tab, it shows the hypertransport I/O link to be Max of 1000Mhz.

So if the CPU said it was capable of 2000Mhz HTL...wouldn't it be reflected in the CBI program as well?
 
nst6563 said:
central brain identifier reports the HTL speed. It shows mine as 900 (running at 300Mhz HTT). Under the Cache/Memory tab, it shows the hypertransport I/O link to be Max of 1000Mhz.

So if the CPU said it was capable of 2000Mhz HTL...wouldn't it be reflected in the CBI program as well?
Click to expand...
Read my post 11.
 
w1retap said:
Its 1000MHz HyperTransport.. the 2000 you guys are seeing is the 2000 MT/s. ;) The HyperTransport is only 1000MHz. Its 1000MHz upstream, and 1000MHz downstream. (chipset to CPU) Its not doubled. www.hypertransport.org
Click to expand...
i don't kno man cuz that newegg link i had in the first post shows this
AMD Athlon 64 3700+ San Diego 2000MHz HT 1MB L2 Cache Socket 939 Processor - Retail
and by that i don't see 2000MT/s i see 2000MHz soo would that mean 4000MT/s??
wow this is really confusing

so far i like the sound of the ddr similarites with HTL, but is that actually true because that would mean it is doubled like ddr...
(cf)Eclipse said:
it's just like ddr.. 200mhz ram is 400mhz effective because it sends data on the rising and falling edge of the clock signal.
Click to expand...
 
soorma07 said:
i don't kno man cuz that newegg link i had in the first post shows this
AMD Athlon 64 3700+ San Diego 2000MHz HT 1MB L2 Cache Socket 939 Processor - Retail
and by that i don't see 2000MT/s i see 2000MHz soo would that mean 4000MT/s??
wow this is really confusing

so far i like the sound of the ddr similarites with HTL, but is that actually true because that would mean it is doubled like ddr...
Click to expand...

By the same token, they list Core 2 Extreme as having 2 x 2MB L2 Cache when it a single 4MB cache, and DDR / DDR-2 RAM is listed by it's effective transfer rates as wel. e.g. DDR-400 is listed as 400MHz instead of 200MHz or 400MT/s.

But I think that one of the above posters is correct in that it's not 1000MHz with DDR, but 1000MHz up and 1000MHz down. Up and Down is independant from each other and can transfer at the same time giving an effective 2000MHz when both points are sending data.
 
Everybody is confused. HTT Hyper transport is "FSB like" but superized between the cpu and chipsets. But the hyper transport links are in the cpu and chipsets. No bottle necks. Not normal FSB, different name 2 times the speed. There are 2 Hyper transport links. 1000mhz each yes dual "duplex" dual pumped. Makes 2000mhz. AMD always have that advantage over intels Legasy FSB. HTT is 2 times the speed and has insane bandwidth.

Don't try to compare the 2.
 
Serge84 said:
Everybody is confused. HTT Hyper transport is "FSB like" but superized between the cpu and chipsets. But the hyper transport links are in the cpu and chipsets. No bottle necks. Not normal FSB, different name 2 times the speed. There are 2 Hyper transport links. 1000mhz each yes dual "duplex" dual pumped. Makes 2000mhz. AMD always have that advantage over intels Legasy FSB. HTT is 2 times the speed and has insane bandwidth.

Don't try to compare the 2.
Click to expand...

Let's see.
AMD's HTT
16bits x 1000MHz = 16000Mbps or 2,000MBytes/second x 2 links = 4,000MBytes/second
Intel's FSB
64bits x 1066MHz = 68224Mbps or 8,530MBytes/second.

where did you learn math Serge?

HTT's superiority come from the fact that it's scalable (add more links for more bandwidth) and takes fewer traces to make a link than a "Legasy" FSB.
 
ryan_975 said:
But I think that one of the above posters is correct in that it's not 1000MHz with DDR, but 1000MHz up and 1000MHz down. Up and Down is independant from each other and can transfer at the same time giving an effective 2000MHz when both points are sending data.
Click to expand...
wow that actually makes sense

and also if that is true than how can
ryan_975 said:
Let's see.
AMD's HTT
16bits x 2000MHz = 32000Mbps or 4,000MBytes/second x 2 links = 8,000MBytes/second
Click to expand...
because it would be 16bits x 1000MHz x 2 links am i correct? because according to the first quote you can't have 2000MHz x 2 links only 1000MHz per link
 
ryan_975 said:
Let's see.
AMD's HTT
16bits x 2000MHz = 32000Mbps or 4,000MBytes/second x 2 links = 8,000MBytes/second
Intel's FSB
128bits x 1066MHz = 136448Mbps or 17,056MBytes/second.

where did you learn math Serge?

HTT's superiority come from the fact that it's scalable (add more links for more bandwidth) and takes fewer traces to make a link than a "Legasy" FSB.
Click to expand...

Intel's frontside bus is 64bit, not 128bit. So it's really 8.5GB/s when you run at 1066MHz FSB. Also, hypertransport can run at 500MHz and you will feel no performance degredation. Try running the Intel frontside bus at 533 versus 1066 and let me know which one handles better ;)

Also, Hypertransport 3 has a peak bandwidth of 20.8 GB/S PER LINK. So a full duplex HTT3 connection would be 41.6GB/s. Let's see, it would take an intel front side bus of... 5GHz effective to reach that? Is my math right there? Yep.

64 x 5000 = 320000Mb/s or 40000 MB/s.

Damn, that Intel frontside bus kind of sucks in comparison. Let's hope CSI tries to at least compete with HTT as a system bus.
 
ryan_975 said:
nah, just a mix of retailer ignorance and marketing deception.
Click to expand...
QFT!

soorma07 said:
Which HT is currently in use is it only HT1?

also i didn't realise how different Intel and AMD actually were in their FSB vs HTT until now :)
Click to expand...

The AMD processors currently come with HT, not HT2 or HT3. Whether there are other devices that use HT2 or even HT3 currently, I do not know.
 
Owned.

robberbaron said:
Intel's frontside bus is 64bit, not 128bit. So it's really 8.5GB/s when you run at 1066MHz FSB. Also, hypertransport can run at 500MHz and you will feel no performance degredation. Try running the Intel frontside bus at 533 versus 1066 and let me know which one handles better ;)

Also, Hypertransport 3 has a peak bandwidth of 20.8 GB/S PER LINK. So a full duplex HTT3 connection would be 41.6GB/s. Let's see, it would take an intel front side bus of... 5GHz effective to reach that? Is my math right there? Yep.

64 x 5000 = 320000Mb/s or 40000 MB/s.

Damn, that Intel frontside bus kind of sucks in comparison. Let's hope CSI tries to at least compete with HTT as a system bus.
Click to expand...
 
To the guys who pointed out I used incorrect numbers, i've corrected them. But my math was still sound.
robberbaron said:
Intel's frontside bus is 64bit, not 128bit. So it's really 8.5GB/s when you run at 1066MHz FSB. Also, hypertransport can run at 500MHz and you will feel no performance degredation. Try running the Intel frontside bus at 533 versus 1066 and let me know which one handles better ;)

Also, Hypertransport 3 has a peak bandwidth of 20.8 GB/S PER LINK. So a full duplex HTT3 connection would be 41.6GB/s. Let's see, it would take an intel front side bus of... 5GHz effective to reach that? Is my math right there? Yep.

64 x 5000 = 320000Mb/s or 40000 MB/s.

Damn, that Intel frontside bus kind of sucks in comparison. Let's hope CSI tries to at least compete with HTT as a system bus.
Click to expand...

For one thing, I never said Intel's FSB was superior to AMD's. It's old tech and falling behind. The point of my post was that the guy was saying that HT1 was 2x as fast as Intel's FSB. With my math (even with the corrected numbers) HT1 trails FSB by a bit per link pair. You can talk about HT3 all you want but that's way down the road and pertinent to this conversation. Another thing why are you talking about slowing down the connections? Of course there's going to be a difference, AMD has an integrated memory controller, Intel doesn't.

Does Intel need something other than a bloared 14 year old technology? yes

Is AMD's HT1 superior to FSB? yes, because it can scale quite nicely.
 
so if you don't notice a difference in running the HT at 500...is there any benefit of running it faster than 1000?

For example, I have my HTT set to 300. HTL is 1200. Is there any difference in running it at 1200 vs 900?
 
nst6563 said:
so if you don't notice a difference in running the HT at 500...is there any benefit of running it faster than 1000?

For example, I have my HTT set to 300. HTL is 1200. Is there any difference in running it at 1200 vs 900?
Click to expand...

For most apps, you won't notice much diffrerence if any at all because you're not affecting memory bandwidth. just the link to the rest of the system. Now for games you will notice since the HT link is the only connection between the CPU and GPU. That is as long as your not GPU bound.
 
it's not a matter of HTT being faster than FSB.. it's just that with amd's integrated memory controller, the memory bandwidth doesn't have to go through the HTT, and on intel, it does. as soon as intel goes to integrated memory controllers, the FSB will be MORE than enough bandwidth.

nst6563 -> http://eclipseoc.com/index.php?id=6,47,0,0,1,0
 
nice ;)

looks like it really doesn't mean squat to run it at higher than 1000 anyway.

Does it mean more if you run the HTT higher? difference between say..300 and 200? or 400 and 200? My system certainly felt snappier when I changed the HTT to 300 whilst keeping everything else the same speed (2.4ghz on cpu, 200mhz ram).
 
Like I siad HyperTransport determines I/O Bandwidth which doesn't effect performance really on the whole. Memory Bandwidth is basically independent of Hypertransport.

Unless you run Hypertransport at only 8Bits. The HyperTransport link should be kept at spec and everything is fine.
 
§kynet said:
Illustrates it perfectly. On board DDR(2) controller AND HT=no FSB bottlenecks, which only gets amplified the more processors per system.
Click to expand...

Not necessarily, those processors still have to share the same link to memory.
 
ryan_975 said:
Not necessarily, those processors still have to share the same link to memory.
Click to expand...
elaborate? if you mean the link inside the chip connecting the cpu to the memory controller.. i think it can handle PLENTY of bandwidth, probably >50gb/s ;)
 
(cf)Eclipse said:
elaborate? if you mean the link inside the chip connecting the cpu to the memory controller.. i think it can handle PLENTY of bandwidth, probably >50gb/s ;)
Click to expand...

I was talking about getting amplified as more processors get added bit. The memory link is still limited to the total bandwidth provided by the RAM itself. Also I was assuming the comment I quoted was referring to more sockets and not cores.

Now if each socket has it's own set of RAM to work with that's another story, but then you have to deal with coherency issues.
 
ryan_975 said:
I was talking about getting amplified as more processors get added bit. The memory link is still limited to the total bandwidth provided by the RAM itself. Also I was assuming the comment I quoted was referring to more sockets and not cores.

Now if each socket has it's own set of RAM to work with that's another story, but then you have to deal with coherency issues.
Click to expand...
gotcha, i misinterpreted then :( :p

and yes, the bandwidth is sorta split up between the cores. however.. imo, the bandwidth available is simply out of hand. it's the latency that really matters, when two cores go after something in the memory around the same time, it's the turnaround time, how long it takes for the second core in line to be able to access the memory, that makes more of an impact. :p

with multi-socket K8 systems, each socket does get it's own set of ram. NUMA ftw. this is the only time where HT link speeds are critical


whew, this is getting off-topic, but i like the direction it's going. <3 ram :D
 
lol... Ramophile?

**Edit** I can't talk... I'm a Corephile. Your (para)phrase, "When 2 cores go after something in the RAM" I got the shivers...
 
Arcygenical said:
lol... Ramophile?

**Edit** I can't talk... I'm a Corephile. Your (para)phrase, "When 2 cores go after something in the RAM" I got the shivers...
Click to expand...

woohoo, when something goes off topic around here it REALLY goes off topic don't it. :)
 
You must log in or register to reply here.
Back
Top