Well they didn't ship my 920 out today so I canceled my order. I'm going to wait and see what monday and tuesday brings when the rest of you get your chips. If batboy has good luck with his 940 then I might get one of those. I would even get a 950 if it will give me my 5Ghz dualcore.
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Presler 930 Results...
- Thread starter Ross
- Start date
- Joined
- May 1, 2003
- Location
- Seattle
Good try Ross, too bad it doesn't seem to be a 5Ghz chip. I had the same thing with my 4000 when they first hit. I think batboy is right on the money.
NOOOO!!
Cancel the shipping, cancel the credit card, cancel everything, although that was a pretty entertaining thread
I'm debating between a 955XE or FX-60, leaning 955XE now due to cold bug. Only thing I don't like is the 4X 2nd slot on P5WD2. Anyone know how much that hurts CF setups?
Sentential said:
NOOOO!!
Cancel the shipping, cancel the credit card, cancel everything, although that was a pretty entertaining thread
I'm debating between a 955XE or FX-60, leaning 955XE now due to cold bug. Only thing I don't like is the 4X 2nd slot on P5WD2. Anyone know how much that hurts CF setups?
- Joined
- Sep 16, 2003
- Location
- Knoxville, TN
My 920 is on the way hyper, you can gauge your results from what I have. Btw for those interested this one was purchased from mwave and Ross's was from costcentralhyperasus said:
- Joined
- Sep 16, 2003
- Location
- Knoxville, TN
Ya its fairly significant. Anywhere from 3-4k on 3d03El<(')>Maxi said:
From what I'm seeing if this is truly a bin issue my 920 will clock better than Ross's 930. Why? Because 920s are blanketed chips.
You gotta remember that they do not test every single chip. They take random samples from random wafers then bin the whole wafer to that level. From a wafer you are talking almost 100 chips. Most importantally some chips are always garunteed to be "failed" chips from another batch. I hate to say it Ross but your 930 and the majority of 930s appear to be failed 940s.
By that basic theory and the recency of the package date I can say its safe to assume that 99.9% of all 9xx chips have been cut from maybe a dozen different wafers because I do not expect more than maybe 2,000 chips to be in retail right now.
That being said the odds of *all* 930s being garbage are infinatly high leaving the remaining chips to pick up the slack. This will be further proved if my 920 does significantally worse or better than Ross's 930.
So the only hope you have of hitting in the level of the 955XEs are to get a blanket wafer (ie a non binned 920 group) or a binned 940 wafer (since there is no 950 yet)
Last edited:
- Joined
- Oct 9, 2004
- Location
- Asheville NC
El<(')>Maxi said:NOOOO!!
Cancel the shipping, cancel the credit card, cancel everything, although that was a pretty entertaining thread
I'm debating between a 955XE or FX-60, leaning 955XE now due to cold bug. Only thing I don't like is the 4X 2nd slot on P5WD2. Anyone know how much that hurts CF setups?
Granted I'm not an engineer, I believe that even the latest pci-E graphic cards do not take up the BW of pci-E 1x.
I ordered a 920 and it's been shipped, will be here monday.
- Joined
- May 1, 2003
- Location
- Seattle
Well if it's that bad I guess I'm going with FX-60 and hope for the best. Are their any 8X/8X or 16X/16X boards close to release that run the XE?
- Joined
- Sep 16, 2003
- Location
- Knoxville, TN
There are a couple but that is generally going to be with only i975X based chipsets which will be out soonEl<(')>Maxi said:
- Joined
- Oct 28, 2001
- Location
- Columbia, SC 29063
Sentential said:Ya its fairly significant. Anywhere from 3-4k on 3d03
From what I'm seeing if this is truly a bin issue my 920 will clock better than Ross's 930. Why? Because 920s are blanketed chips.
You gotta remember that they do not test every single chip. They take random samples from random wafers then bin the whole wafer to that level. From a wafer you are talking almost 100 chips. Most importantally some chips are always garunteed to be "failed" chips from another batch. I hate to say it Ross but your 930 and the majority of 930s appear to be failed 940s.
By that basic theory and the recency of the package date I can say its safe to assume that 99.9% of all 9xx chips have been cut from maybe a dozen different wafers because I do not expect more than maybe 2,000 chips to be in retail right now.
That being said the odds of *all* 930s being garbage are infinatly high leaving the remaining chips to pick up the slack. This will be further proved if my 920 does significantally worse or better than Ross's 930.
So the only hope you have of hitting in the level of the 955XEs are to get a blanket wafer (ie a non binned 920 group) or a binned 940 wafer (since there is no 950 yet)
Where else are you seeing reports on the 930's?
- Joined
- Oct 9, 2004
- Location
- Asheville NC
El<(')>Maxi said:
Maybe it's not bad at all, check this out.
"Obviously, the PCIe x4 is not able to drive the same amount of bandwidth as a full PCIe x16 slot, but running a graphics card in a PCIe x4 slot surprisingly did not hurt performance in any meaningful way. When we thought about it, we could see why. A PCI Express x4 slot is capable of delivering 2.0 GB/s of bandwidth, compared to 8.0 GB/s of a PCI Express x16 slot. In comparison, AGP 8x can deliver 4.0 GB/s, where AGP 4x can deliver 2.0 GB/s of bandwidth, which is roughly the same as a PCI Express x4 slot.
Basically, running a PCI Express graphics card in a PCIe x4 slot will basically be giving it the same bandwidth as an AGP 4x slot. Which, frankly, is still a lot of bandwidth for a GPU to utilize. We threw our Radeon X800 XT PCI Express card in the PCIe x4 slot to see how the lack of bandwidth would hurt performance.
. Radeon X800 XT PCIe x16 PCIe x4
Doom 3 - Site3 Timedemo (1024 x 768) 100.5 99.4
UT2004 - Rankin Botmatch (1024 x 768) 187.57 187.53
UT2004 - Colossus Botmatch (1024 x 768) 72.73 72.66
Obviously, this does not constitute a full set of benchmarks, rather a quickie set we ran when we were finishing up this article. What it basically shows is that modern cards still are not pushing the needs of a high-bandwidth graphics slot as of yet, as our Radeon X800 XT PCIe card ran at nearly full speed when run in a much slower PCIe x4 slot.
What does this mean in the end? Well, in theory, we had two GeForce 6600/6800 series PCI Express cards, it looks like that SLI would be possible on the X6DAE-G2 motherboard. While nVidia recommends a PCIe x16 and a PCIe x8 slot for an SLI configuration, only having a PCIe x4 slot for the secondary card should not be a bottleneck. If a Radeon X800 XT PCIe card can barely max out a PCIe x4 slot, using mid-range 6600/6800 cards in this slot should not hurt SLI performance. In theory, of course.
After writing this section, we got confirmation that it is indeed possible to run an nVidia SLI setup on the Supermicro X6DAE-G2 motherboard using PCI Express x16 and PCI Express x4 slots. Obviously, this is a major feature advantage over Tyan’s competing Thunder I7525 platform, despite the fact that we really can’t utilize this today. We are seeing nVidia’s first SLI-enabled PCI Express graphics cards slowly hitting the market though, so we should actually be seeing working SLI setups within the coming weeks!"
-credit to gamepc.com
- Joined
- Sep 16, 2003
- Location
- Knoxville, TN
There isnt any yet, but statistically speaking the odds of seeing other 930s with the same results is *very* high since most wafers are consistant these days.S_Wilson said:
Just from a buisness perspective the way Intel bins their chips has to be highly effective with a decent margin of error otherwise they would have a high return rate.
- Joined
- Oct 28, 2001
- Location
- Columbia, SC 29063
Sentential said:
Sorry, I thought I read something different than what your original post says. Now I am confused why I asked that, lol.
- Joined
- Sep 16, 2003
- Location
- Knoxville, TN
No problem. I am wholly convinced that there is an OEM / Dell factor going on. MFGRs are more likely to let higher end wafers slip if the demand for a certian chip is high enough to warrent it. They'd rather spend slightly more in terms of cost if it means they will get their money right away.S_Wilson said:Sorry, I thought I read something different than what your original post says. Now I am confused why I asked that, lol.
If you look at Dell's catalouge you will see almost nothing in the way of 930s being sold. The majority are either 920 or 940s. If there was a way to see how many chips of each type they are making it would be easier to pick out which ones are the best to buy.
Generally the ones with the highest population will have the highest varience and therefor should see the biggest spread in terms of clocks.
- Joined
- May 1, 2003
- Location
- Seattle
Thats an interesting theory Sen
I hope your right about it.
I hope your right about it.
- Joined
- May 11, 2004
- Location
- Montana tech Butte mt.
Hmm interesting becuase the 830s seemed to be the 8x0 series sweet spot.
Also I am thinking that binning may have came into play becuase if you look at Intel CPUs in general the 5x0 series the early ones up to the 540 didn't clock so well. the 630 wasn't really a clocker either. Once it gets all sorted out this will probably be fine. Consider also that the packaging might not perfected. Early china chips can Kiss my tail but hey the newer ones are even starting to kill off the mighty costa ricas and others.
So who knows one problem of being the first is you are a guinee pig though.
Also I am thinking that binning may have came into play becuase if you look at Intel CPUs in general the 5x0 series the early ones up to the 540 didn't clock so well. the 630 wasn't really a clocker either. Once it gets all sorted out this will probably be fine. Consider also that the packaging might not perfected. Early china chips can Kiss my tail but hey the newer ones are even starting to kill off the mighty costa ricas and others.
So who knows one problem of being the first is you are a guinee pig though.
- Joined
- May 11, 2004
- Location
- Montana tech Butte mt.
Oh yes maxi Mikeguava said that the 16x 4x hurts for 05 more than anything else for 01 and 03 it seemed to matter less.
The i975 should change that though athough until you make them really cold Intels just don't smoke AMDs in anything except 03 and superpi.
The i975 should change that though athough until you make them really cold Intels just don't smoke AMDs in anything except 03 and superpi.
- Joined
- Sep 16, 2003
- Location
- Knoxville, TN
speed bump said:
See I beg to differ on that. 630s when they first were out were doing signficantally better than the 640s and still to this day they clock better overall. The same thing can be said for the 660s. 660s pre 670 bins were fantastic but now the majority of 660s are failed 670s. Right now the hot chip to buy is either a 630 or a 650.
Why? Because Dell buys so many 650s that I bet they use higher end wafers to fill the gap in demand which then rolls over to the retail market. However demand for 660s and 640s is relitivly small since its not cost effective and therefor the demand is lower.
The 830 was the same way. 830s were terrible at lauch, I remember that as clear as day. Partially because no one wanted them, the majority of people could only afford 820s and those looking for high end went with 840s.
Since the demand was to high for 820s at the onset the same thing happened. They de-binned previously higher grade chips to satisfy demand. AMD does this all the time, especially with the S939 Opterons and the 3000+.
Mainly it has to do with population count. The smaller the group is the more controlled it tends to be. The more that are in a group the higher your odds of getting one that shouldnt be in there.
Right now if you want to take a gamble Id suggest either a 940 or a 920 until my submission proves otherwise. I will do a clock for clock comparison and if mine is signifcantally better or worse it will tell us alot about how confident Intel is about its yields.
In addition do not count on Cedar Mill being a fix-all CPU. Since intel went with a spit die there is almost nothing stopping them from taking failed 9xx and disabling the dud core leaving you with a mediocre chip.
There is going to be a way to figure out which is which eventually but its somethign to keep in mind when you purchase these things.
In order to buy a successful chip you have to do 1 of two things. Either buy one thats under their radar (ie a 920) or buy a chip thats above their binning average (such as a 650) but below there cap for the failed EEs.
My guess is that the chips to watch for will be: Pentium D 920, Pentium D 940? (depends if they are failed 950s or not), Pentium D 950, Pentium 4 651 (if there is a 671) or Pentium 661 (is there is no 671), Pentium 4 631.
The odds of the 661 and 641 being failed copies of their higher end brothers is significantally higher than a 631 being such. The same can be said of the 920.
Overall we will see in due time, all I can offer is logical explinations. For all I know this could just be how Presler is. I may not even be able to hit 4ghz (shame the thought) but its entirely possible that this series has a scaling issue.
- Thread Starter
- #60
Sorry I didn't check in sooner...lotsa good stuff going on now I personally am never buying any Intel x3x proc again because I have no luck with them for OCing (the 630 I had for all of 2 days was a dud as well).
Interesting Sen. My x700 Pro vid card is like that...the core says x700XT and it clocks like crazy for a Pro Yes, there will be a 671, but not until April or something.
I really, really hope batboy's 940 does well. If so, I might even consider getting one instead of a 661 after I benched all night. For the severe let down in OC-ability, I am very pleased to see the bench results at the low level I did (4350). At just 290FSB/4.35GHz it pretty much beat the crapola out of my 670 @ 350FSB/4.9GHz (which is no slouch!) across the board....er, except SPi If someone can get one of these 900s ramped up to 325+ FSB, it will be insane...especially 3Ds. I benched everything I had installed and will post screen shots tomorrow.
Quick rundown...rounded numbers:
670 @ 4.9GHz/350FSB -> 930 @ 4.35GHz/290FSB
PCMark04 7950 -> 9200
PCMark05 4500 -> 6050
PCMark01 28400 -> 28500 (single card, this was the worst gain)
PCMark03 11450 -> 11750 (CF x1600XT)
PCMark05 5600-> 5800 (single x1600XT)
PCMark05 8600 -> 8800 (CF x1600XT)
Those 3Ds aren't huge gains, but the 930 is beating the 670 using 60 less FSB, which is a TON of performance difference when it comes to benching. I also benched at the same card clocks that I used on the 670, but for whatever reason, the cards will clock even higher now. Also, I am sure my fragged HDDs (1 SATA/1 IDE) and 1.5GB swap files (x2) weren't helping the PCMark scores
On another good note, as I just mentioned, this proc really cleaned up CF. Not only in terms of benchmarks, but in how well it ran. On my 670 with CF enabled, I can't OC the cores 1MHz without artifacting in Nature in 01 or 03. I was able to OC the cores 15MHz on the 930. Yeah, it's not a lot by comparison, but that's a ton on these 1600XTs cores and it makes a big difference in scores.
I really wish I could understand how a huge 55C drop in temps from air to phase struggles to give a 500MHz higher clock It just seems impossible that an air cooler will run it that close to its ultimate limit. If I keep this thing for any reason (the benches got me thinking about it), I am going back to air and saving myself the electric bill from phase
As for the 4x PCI-e bandwidth limiting CF...I really don't know if that's as much an issue on 1800s that use a cable to send the slave data to the master (ie. only one way on the bus to the 2nd card), but for the 1300/1600s that only talk to each other over the PCI-e bus (both ways), I'll bet it makes a difference.
I personally am never buying any Intel x3x proc again because I have no luck with them for OCing (the 630 I had for all of 2 days was a dud as well). Interesting Sen. My x700 Pro vid card is like that...the core says x700XT and it clocks like crazy for a Pro
Yes, there will be a 671, but not until April or something.I really, really hope batboy's 940 does well. If so, I might even consider getting one instead of a 661 after I benched all night. For the severe let down in OC-ability, I am very pleased to see the bench results at the low level I did (4350). At just 290FSB/4.35GHz it pretty much beat the crapola out of my 670 @ 350FSB/4.9GHz (which is no slouch!) across the board....er, except SPi
If someone can get one of these 900s ramped up to 325+ FSB, it will be insane...especially 3Ds. I benched everything I had installed and will post screen shots tomorrow.Quick rundown...rounded numbers:
670 @ 4.9GHz/350FSB -> 930 @ 4.35GHz/290FSB
PCMark04 7950 -> 9200
PCMark05 4500 -> 6050
PCMark01 28400 -> 28500 (single card, this was the worst gain)
PCMark03 11450 -> 11750 (CF x1600XT)
PCMark05 5600-> 5800 (single x1600XT)
PCMark05 8600 -> 8800 (CF x1600XT)
Those 3Ds aren't huge gains, but the 930 is beating the 670 using 60 less FSB, which is a TON of performance difference when it comes to benching. I also benched at the same card clocks that I used on the 670, but for whatever reason, the cards will clock even higher now. Also, I am sure my fragged HDDs (1 SATA/1 IDE) and 1.5GB swap files (x2) weren't helping the PCMark scores
On another good note, as I just mentioned, this proc really cleaned up CF. Not only in terms of benchmarks, but in how well it ran. On my 670 with CF enabled, I can't OC the cores 1MHz without artifacting in Nature in 01 or 03. I was able to OC the cores 15MHz on the 930. Yeah, it's not a lot by comparison, but that's a ton on these 1600XTs cores and it makes a big difference in scores.
I really wish I could understand how a huge 55C drop in temps from air to phase struggles to give a 500MHz higher clock
It just seems impossible that an air cooler will run it that close to its ultimate limit. If I keep this thing for any reason (the benches got me thinking about it), I am going back to air and saving myself the electric bill from phase As for the 4x PCI-e bandwidth limiting CF...I really don't know if that's as much an issue on 1800s that use a cable to send the slave data to the master (ie. only one way on the bus to the 2nd card), but for the 1300/1600s that only talk to each other over the PCI-e bus (both ways), I'll bet it makes a difference.
