Has anyone started crunching with AMD's new 8 core CPU?

I'm planning to build an AMD rig next and have been torn between trying the FX-8150 or just going with a Phenom II X6 1100T.

On paper the 8150 seems to make since. 8 wu's for the CPU and 4 wu's for the GPUs if I go with 2 GTX 470's. 12 total would be pretty sweet!

---

My Computers:
â–ˆ Blue Offline
â–ˆ Green Offline
â–ˆ Red Offline

One of my team mates has an 8120 and had it OCed to 4540 Mhz on air. Its a darn stable CPU and with 8 true cores you'll not have the nasty lag that the intel Hyperthreaded CPU's would give

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

The FX supports AVX instructions so could be useful for Seti V7.
I´m not sure i can resist.

---

With each crime and every kindness we birth our future.

One of my team mates has an 8120 and had it OCed to 4540 Mhz on air. Its a darn stable CPU and with 8 true cores you'll not have the nasty lag that the intel Hyperthreaded CPU's would give

What nasty lag is this you are talking about?

---

One of my team mates has an 8120 and had it OCed to 4540 Mhz on air. Its a darn stable CPU and with 8 true cores you'll not have the nasty lag that the intel Hyperthreaded CPU's would give

What nasty lag is this you are talking about?

Presumably the fact that because the HT cores are not completely separate (they share some hardware), they can interfere with one another at times. That's why a HT core does less than the same amount of work as 2 non-HT cores.

---

One of my team mates has an 8120 and had it OCed to 4540 Mhz on air. Its a darn stable CPU and with 8 true cores you'll not have the nasty lag that the intel Hyperthreaded CPU's would give

What kind of numbers are they getting out of it?

---

My Computers:
â–ˆ Blue Offline
â–ˆ Green Offline
â–ˆ Red Offline

If you want efficient SETI cruncher for reasonable price, it is 2500K. Bulldozer has only 4 FPU blocks(although twice wider, but it has no good use ATM), which matters for SETI.
Most efficient is 2600K, but it is up to you to decide does it worth or not. HT is not an issue, time per task is longer, but summary output is increased 30-40 and even more percent. Bulldozer cannot beat even 2500K, not to speak about 2600K, its even worse than 6-core AMDs(where 6 totally independent FPUs are available.
You can look at my WUs completion times, those crunched by CPU. 4660 MHz, bellow 100 watts power usage. Bulldozer beats here, its hotter when overclocked, and yet more power hungry(obviously).
HT may affect GPU WU completion times, only if u dedicate cores to serve GPUs. But if u run two or more tasks per GPU, it doesn't matter, GPU pipes are fed well.
Sorry for the offtopic, I had to point these.

---

Thanks for the input. The 2500K does look like a great chip.

I've been kind of leaning toward the AMD's since it seems like they do better with the number of x16 slots. AMD 990FX chipset can do 16x, 8x, 8x. VS. 16x, 8x, 4x with the P67 chipset. But I have seen Intel rigs running 3 & 4 GPUs, but not sure what Motherboards/chipsets there using.

---

My Computers:
â–ˆ Blue Offline
â–ˆ Green Offline
â–ˆ Red Offline

PCIe performance may affect crunching, in measurable numbers, but everything over x4 2.0 will give u hardly measurable difference(needs multiple runs to clarify that the difference does exist).
But slower CPU may affect it far more. Not only in pure CPU crunched tasks, but in CPU time used for feeding GPU. I've seen same GPUs(460 in my research) taking over a minute CPU time for an average WU. This is on AMD x6, clocked at 3.8, IIRC. Where on my CPU they take around half minute. This, put over 10 minutes total time for a WU, makes huge difference.
I had numbers how PCIe speed affects performance, but I can't find them now. Its like 1% slower on x8 than x16, literally few seconds. CUDA 4 has improved host to device transfer speeds, so now it can be even less.
Intel rigs may have NF200 multiplexor chip, to multiply PCIe lines and switch between devices, having an active transfer. In SETI case, where transfers are small, it may sevre well, but still remains questionable, bcs added latency from the chip itself.
I have ran three GPUs on Intel X48, where lines are x16, x16, x4. I can say, yes, there was difference, but its far from being something, which would make me to consider MB with full blown x16 slots.
In fact, overclocking your GPU a bit(10 MHz on 800 MHz basis) may render this difference to nil.

---

One of my team mates has an 8120 and had it OCed to 4540 Mhz on air. Its a darn stable CPU and with 8 true cores you'll not have the nasty lag that the intel Hyperthreaded CPU's would give

What kind of numbers are they getting out of it?

It's hard to tell currently. He does run other projects but Heres a link to the computer
He currently has it at 4521mhz

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

Disaster. 5000 vs 3000 seconds for average WU, compared to my 2500K.

---

Thanks for the input. The 2500K does look like a great chip.

I've been kind of leaning toward the AMD's since it seems like they do better with the number of x16 slots. AMD 990FX chipset can do 16x, 8x, 8x. VS. 16x, 8x, 4x with the P67 chipset. But I have seen Intel rigs running 3 & 4 GPUs, but not sure what Motherboards/chipsets there using.

You might want to have a look at these articles on Toms Hardware.
GeForce And Radeon On Intel's P67: PCIe Scaling Explored
AMD FX: Energy Efficiency Compared To Eight Other CPUs

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Well, its not that bad maybe. seems 2500K and 8150 have similar performance, IF this Bulldozer runs 8 tasks simultaneously, which cannot be said at a glance.
Still 2500K is better choice, bcs of the price, power efficiency and single threaded performance, for all around usage. Also, bcs better singlethreaded performance it would feed GPUs better, tho.

---

Thanks for the input. The 2500K does look like a great chip.

I've been kind of leaning toward the AMD's since it seems like they do better with the number of x16 slots. AMD 990FX chipset can do 16x, 8x, 8x. VS. 16x, 8x, 4x with the P67 chipset. But I have seen Intel rigs running 3 & 4 GPUs, but not sure what Motherboards/chipsets there using.

GPUs are the obvious way to go if you're looking for high powered number crunching. PrimeGrid is an extreme example for how efficiently a GPU can be utilised!

So to my mind, the balance made in Bulldozer is a very good idea if you also have APU/GPU number crunching available. It's more a question of how well the software fits the CPU. Bulldozer is the first of the direction for the future.

It will be very interesting to see how performance for the various Boinc projects compare on Bulldozer.

Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)