The word on the street is that these might be even faster than the CUDA clients: http://n0cgi.distributed.net/cgi/dne...?user=mikereed
The word on the street is that these might be even faster than the CUDA clients: http://n0cgi.distributed.net/cgi/dne...?user=mikereed
[Aug 15 19:49:39 UTC] RC5-72: Benchmark for core #1 (IL 4-pipe c alt)
0.00:00:16.09 [132,721,054 keys/sec]
Windows 7 RTM x64, AMD Phenom 9850, Radeon 3850
[Aug 16 00:11:25 UTC] RC5-72: using core #0 (IL 4-pipe c).
[Aug 16 00:11:35 UTC] RC5-72: Benchmark for core #0 (IL 4-pipe c)
0.00:00:07.42 [589,471,423 keys/sec]
With a Core2 Q9550 and an ATI 4870 (GPU 750 MHz, Memory 950 MHz) running Vista 64bit.
Very nice, I'm using up all 4 CPUs for WCG with BOINC, and the d.net client is using up maybe 1-2% of 1 core to keep the GPU fully loaded.
Oh I think that will happen. Afterall, the stream thing is just getting going. I was reading on another project page (boinc one) can't remember which one, but one of the prog/admins was commenting that it was taking them much longer than they had originally thought it would to get both the cuda and the stream clients going. He intimated that there were some differences between them that created difficulties. So if it is hitting one place, it's likely hitting others as well.
I went and bought a HD 4650 to experiment with. I wanted one of the 48xx but the local store was out of them. And I'm running it on XP Pro, and there's pluses and minuses. Overall though I'm impressed with it. For one, it doesn't need the PCI Express electrical plug, just takes it's power from the m/b, so it can't be sucking too heavy a draw. It also isn't putting out much heat. And it's very quiet. It's crunching a single work unit every 24 seconds, a little bit slower than a cuda 9800GTX+ I have. The cuda one puts out more heat, makes noticeably more noise and requires the PCI Express electrical plug from the P/S. Of course the heat equates to more juice being sucked down. The stream worked the first try too, which is always a plus
I haven't tried playing with it under linux yet. Don't know if I will really. While I'm not a windows fan at all, it is definitely much easier to get video drivers and all changed around under windows than linux. And that becomes a bit of an issue when there are relatively frequent beta changes that require changes due to new libraries and all.
I was reading that the HD 48xx put out a lot of heat, but then so do the nvida 260's and 280's and up. So I intend to pick up a 48xx stream card and give that a run and see how it goes. But so far I think that the ATI/AMD stream is going to pan out just fine on rc5 anyway.
They've just announced that they should have the ATI integration into the BOINC 6.10.x stream within a month, it's basically the major thing they are working on now that the gridrepublic/facebook 6.8.x version is out the door..
Of course you can always try MilkyWay or Collatz Conjecture right now
Indeed! A 4870 can be had for £70-80 here at the moment. To get a similar keyrate from nVidia hardware would set you back a fair bit more as far as I can see.
Also, a 4650 costs about half as much as a 4870. Brucifer, you said yours is doing one block per 24 seconds. I'm not sure what size blocks those are, but if it's 2^32 keys, that works out to around 180 Mkeys/sec. Does that sound about right? If so, that's quite a big leap between the two cards..
The 4650 gives roughly 174,900,000 keys/sec on the - bench for core 0 which is the correct default core. Here the 4650 is currently running at $119 US. So yes, there is a HUGE leap between the 4870 and the 4650. So for less than double the price you get 3.3 times the keyrate which is one heck of a deal.
edit: My BFG GTX 260 is getting 234,310,064 keys/sec on core #6, cuda 2.2 client, so the 4870 is better than 2 times the keyrate of the BFG GTX 260. So unless there is a similar error in the crunched results from the stream client like there was with the cuda 2.1 and below clients, the 4870 is definitely a hot number on rc5!!!
Last edited by Brucifer; 08-19-2009 at 12:54 AM.
On RC5 work, with nothing else running on the computer.
GTX 285 294.25 Mkeys/s
GTX 280 269.13 Mkeys/s
GTX 275 282.94 Mkeys/s
using core #0 (CUDA 1-pipe 64-thd) and priority 2.
82 GHz and still climbing as all 3 of them have only been running for a couple days
http://teamstats.macnn.com/rc572/sta...sort=StatsWeek
That ATI card kicks some serious butt in RC5
ZipZoomFly has it for $169.99
http://www.zipzoomfly.com/jsp/Produc...tCode=10009805
Edit: Out of stock
Yeah.... I'm afraid that there's gonna be a lot of places "out of stock" as far as that ATI card goes. Just absolutely can't beat it for the price.
By my calculations, if the RC5-72 project was started from scratch using 10,000 HD 4870 cards, the keyspace would be exhausted in ~25 years. That smashes a rough estimate of 1,040 years at the current average rate.
Bear in mind that 10,000 is only 1/8th of the total number of participants.
I want one of these bad boys.
Wow, I actually have something that beats everyone else's gear?
It looks like MilkyWay and Collatz Conjecture for ATI you need to do some hacking to get stuff working properly. I'm assuming that this should sort itself out when the BOINC support ATI release is complete? I might just wait for that, not much time to mess around right now.
Last edited by Digital Parasite; 08-19-2009 at 12:29 PM.
4870 $139 - http://www.newegg.com/Product/Produc...82E16814129113
4890 $199 - http://www.newegg.com/Product/Produc...82E16814150359
4890 is 2x faster
Core Clock 850MHz vs 750MHz
Memory Clock 3900MHz vs 1800MHz
Memory Size 1GB vs 512MB
$120 https://www.ewiz.com/detail.php?name=XFX-477_51
- Mfr Part Number: HD477AYDLC
- Chipset: Radeon HD 4770
- Engine Clock: 750 MHz
- Video Memory: 512MB DDR5
- Memory Clock: 3.2 GHz
- Memory Interface: 128-bit
- Bus: PCI-Express x16 (Support PCI-Express 2.0)
- RAMDAC: 400 MHz
- Stream Processing Units: 640
I picked up a Radeon HD4850 last night, benchmarks are;
[Aug 21 15:35:18 UTC] RC5-72: using core #0 (IL 4-pipe c).
[Aug 21 15:35:29 UTC] RC5-72: Benchmark for core #0 (IL 4-pipe c)
0.00:00:08.84 [489,096,269 keys/sec]
[Aug 21 15:35:29 UTC] RC5-72: using core #1 (IL 4-pipe c alt).
[Aug 21 15:35:42 UTC] RC5-72: Benchmark for core #1 (IL 4-pipe c alt)
0.00:00:10.59 [411,941,905 keys/sec]
[Aug 21 15:35:42 UTC] RC5-72 benchmark summary :
Default core : #0 (IL 4-pipe c)
Fastest core : #0 (IL 4-pipe c)
and running just the single unit work units, it crunches one ever 8 secs. Puts out better than double what my BFG GTX 260 does, and puts out about the same amount of heat as the 260 but still runs quieter.
Just installed a new Radeon HD 4890, dual 6pin power connectors.
Benchmark at stock speeds is: 670,104,782 keys/sec
Smokes the pants off any nVidia card periodPHP Code:
[Aug 21 22:32:37 UTC] RC5-72: Benchmark for core #0 (IL 4-pipe c)
0.00:00:06.58 [670,104,782 keys/sec]
[Aug 21 22:32:37 UTC] RC5-72: using core #1 (IL 4-pipe c alt).
[Aug 21 22:32:47 UTC] RC5-72: Benchmark for core #1 (IL 4-pipe c alt)
0.00:00:07.59 [578,426,446 keys/sec]
[Aug 21 22:32:47 UTC] RC5-72 benchmark summary :
Default core : #0 (IL 4-pipe c)
Fastest core : #0 (IL 4-pipe c)
Wow,
now why did I think you would pick one of those bad boys up
Bok
yeah... I was reading a bunch of the "reviews on Best Buy, etc., and people were commenting on the streams running hot. Evidently they haven't run the GTX stuff, cause to me they seem like they run warmer than the stream stuff.
What I am noticing though is that since the stream is the first public beta release that the client is not as polished as the cuda client IRT using the cruncher and web surfing at the same time. This one I'm using isn't a high end one but it is comparable to the 9800GTX+ but it's more choppy than the 9800GTX. The 4800 series are better, but still have slower screen response than the GTX 2xx geforce stuff. I imagine that will improve as time goes by. Didn't buy the stuff for surfing though rather it was primarily for crunching.
For PSUs without two six-pin plugs, can a molex->6-pin adaptor be used for connecting to one of these cards?
I'm having trouble with a decision of which is better, perhaps you can help me?
2 4770s vs 1 4890
2 4770s vs 2 4850s
2 4850s vs 4890
2 4890s and be done with it already
Is core clock more important than memory clock? - for crunching only
I assume we don't enable cross-fire, so the clients would see two GPUs.
Power consumption and performance as a dedicated RC5 or F@H cruncher only - no gaming considerations
Good review that covers some of the angles...
http://www.hardocp.com/article/2009/...e_evaluation/9
Well, from this thread alone we can summarise:
HD 3650 - ..57.3 Mkeys/sec: 194? watts: 120 shaders
HD 4650 - 174.9 Mkeys/sec : ..55 watts : 320 shaders
HD 4850 - 489.1 Mkeys/sec : 110 watts : 800 shaders
HD 4870 - 589.5 Mkeys/sec : 150 watts : 800 shaders
HD 4890 - 680.0 Mkeys/sec : 190 watts : 800 shaders
HD 4890 - 728.0 Mkeys/sec : 200 watts : 800 shaders : Core Clock 950MHz : Memory Clock 1050MHz
Also, check this Wikipedia article. It's a neat comparison table.
It looks to me like core clock is more important.
: added by IronBits - references
http://www.guru3d.com/article/radeon...preview-test/2
http://www.techpowerup.com/reviews/Sapphire/HD_4650_OC
http://www.neoseeker.com/Articles/Ha...hd4890/14.html
.
With all that said...
http://www.newegg.com/Product/Produc...82E16814131156
How about a blow by blow write up on how you did it, like what all hardware is involved, etc. Curious about it as I'm noticing that performance for all cards sorta sucks on screen updates when crunching hard too. I thought about just doing a system for my surfing, etc., but unfortunately I seem to have a problem with a system sitting around the house that isn't crunching.
So the next best thing is to get one set up for both. However I first need to look at motherboards with multiple PCIE slots so I can run more than one gpu. Of course I guess and option would be to get one of the healthy nvida cuda motherboards and then go from there.
I have several computers, not in cases, sitting on shelves, configured the way I want them, for dedicated crunching only.
All hooked up to a 8 port KVM switch and 23" widescreen monitor for setups and tweakage of BIOS, accessed remotely using Real VNC.
The're referred to as headless boxen.
I have ONE computer, with all the latest and greatest bells and whistles, parts and pieces, case and all,
with a 30" Apple Cinema Display, for my personal pleasure, that I may or may not use for crunching.
Somehow my crunchers end up being just like my personal computer, sans case and monitor.
Man.................. what a RIP OFF!!!!!!!!! Did you see that little note up there that they are limiting that offer to ONLY 99 PER CUSTOMER???????
LOL Just thought I'd throw that cheap little shot out there.
That's a pretty smokin deal!! Getting back to the real world for a minute though, there is a lot of bucks going into crunchers for RC5 here. Luckily the stuff works on other projects too.
Naw, he did it because of the female shoe syndrom... they can have a closet full of shoes, but as that saying went by some gal somewhere, "If the shoe fits, buy it" so goes IB with "If a PCI-E slot is empty, fill it!" Obviously he had an empty slot.
LOL, actually, after you posted your numbers, I was so impressed, I became curious as to what the top of the line card could do ...
Then, found out there was an OC version, so I'll let you know how well that one works to, in a few days.
nVidia cards will be turned off and retired for now, except my gaming card of course.
However, now that you mention it...
Oh, and I have to stay ahead of Mustard don't you know!