F5F Stay Refreshed Hardware Desktop X58 uncore power draw?

X58 uncore power draw?

X58 uncore power draw?

Pages (2): 1 2 Next
K
kettususi
Junior Member
4
10-23-2016, 03:38 AM
#1
Adjusting my rig for folding means focusing on power savings. The x5650’s best setting is around 3.7ghz at 1.19v, but I’m using 1.256v with the BIOS off. I’m curious about the power consumption of the core and chipset since even at lower speeds the cores don’t heat up much. Running at 1.6v gives me a 6x4 d9qbj setup, which makes the dual-core clocks look weaker compared to voltage. I’m okay with going up to about 1.6v VTT but prefer staying around 1.5v to be safe and maintain good efficiency. Since I don’t have a GPU yet, this will be important for performance and power use.
K
kettususi
10-23-2016, 03:38 AM #1

Adjusting my rig for folding means focusing on power savings. The x5650’s best setting is around 3.7ghz at 1.19v, but I’m using 1.256v with the BIOS off. I’m curious about the power consumption of the core and chipset since even at lower speeds the cores don’t heat up much. Running at 1.6v gives me a 6x4 d9qbj setup, which makes the dual-core clocks look weaker compared to voltage. I’m okay with going up to about 1.6v VTT but prefer staying around 1.5v to be safe and maintain good efficiency. Since I don’t have a GPU yet, this will be important for performance and power use.

C
CaveMiner1215
Member
91
10-23-2016, 11:21 AM
#2
Typically, the main processing units consume the most power, which explains why the rest of the chip isn’t drawing much. I don’t remember temperatures rising significantly with any of my X-chipset setups.
C
CaveMiner1215
10-23-2016, 11:21 AM #2

Typically, the main processing units consume the most power, which explains why the rest of the chip isn’t drawing much. I don’t remember temperatures rising significantly with any of my X-chipset setups.

I
iDoNotEvenLift
Posting Freak
936
10-30-2016, 05:27 PM
#3
Welp seems the power usage is acceptable but performance won’t exceed 3500 with this RAM setup and those D9QBJS are a bad choice (should have picked quad rank D die or CFR). I’ve got three unused double-sided CFRs that I can run at around 125TRC at 2800, so I think I could switch them in and hit 100TRC at 2000 since RAM usage stays under 3GB. Not sure if dual-rank RAM will help or if running the uncore at 4000–4200 (around 1.6V VTT) is better.
I
iDoNotEvenLift
10-30-2016, 05:27 PM #3

Welp seems the power usage is acceptable but performance won’t exceed 3500 with this RAM setup and those D9QBJS are a bad choice (should have picked quad rank D die or CFR). I’ve got three unused double-sided CFRs that I can run at around 125TRC at 2800, so I think I could switch them in and hit 100TRC at 2000 since RAM usage stays under 3GB. Not sure if dual-rank RAM will help or if running the uncore at 4000–4200 (around 1.6V VTT) is better.

E
ElitexWizard
Member
210
11-03-2016, 02:27 AM
#4
My 970 became quite irritable around low voltages near 3260. My X5690ES held up well up to 3900, but performance was heavily influenced by bclk/qpi adjustments. Keeping it at about 3600 seems reasonable. It looks like you're applying a lot of voltage—if it isn't running smoothly between 1.4 and 1.45, chances are she won't perform much. Since you're running the mems at full capacity, things are looking okay.
E
ElitexWizard
11-03-2016, 02:27 AM #4

My 970 became quite irritable around low voltages near 3260. My X5690ES held up well up to 3900, but performance was heavily influenced by bclk/qpi adjustments. Keeping it at about 3600 seems reasonable. It looks like you're applying a lot of voltage—if it isn't running smoothly between 1.4 and 1.45, chances are she won't perform much. Since you're running the mems at full capacity, things are looking okay.

I
IMS209
Member
63
11-04-2016, 12:59 PM
#5
They appear to reach about 1.5v before performance drops sharply. My old x5660 handled around 3900 uncore at 1.45-1.48v and 4100 at 1.55v, then fell to 1.35v which limited the overclock only to about 3500 uncore. This pattern seems similar for Nehalem with my i7 930—capable of 2800 DDR3 at 1.5v, 3000 at 1.6v, and 3100 at 1.7v (thanks to the 1:1 uncore memclk on my ASUS board, which still works for 3100 DDR3). It looks like some processors struggle with high uncores. I suspect this affects how much RAM they can support, though my i7 930 seems to handle 3800 uncore at 1.6v well. The IMX doesn’t want to go beyond 3100 DDR3 on Samsung GD-ies, which I’ve reached before. I’m wondering if this limits the maximum clock speed for RAM, since I imagine my i7 930 could push 3000 DDR3 at 1.5v but struggles above that. Next time I’ll check online stores to pick better chips that can reach 4.5 at 1.4v or higher, or ones with high uncore support for Xeon W or Gullfield. Also, do you have experience with Foxconn, EVGA, or DFi boards? They’re all reputable, except for the Foxconn line which has improved recently. The memclk ratio matters a lot, especially for Foxconn boards (I can buy a high-end blade for $35 if it meets specs). If I hit a bad ratio, I could flash a BIOS and tweak voltages.
I
IMS209
11-04-2016, 12:59 PM #5

They appear to reach about 1.5v before performance drops sharply. My old x5660 handled around 3900 uncore at 1.45-1.48v and 4100 at 1.55v, then fell to 1.35v which limited the overclock only to about 3500 uncore. This pattern seems similar for Nehalem with my i7 930—capable of 2800 DDR3 at 1.5v, 3000 at 1.6v, and 3100 at 1.7v (thanks to the 1:1 uncore memclk on my ASUS board, which still works for 3100 DDR3). It looks like some processors struggle with high uncores. I suspect this affects how much RAM they can support, though my i7 930 seems to handle 3800 uncore at 1.6v well. The IMX doesn’t want to go beyond 3100 DDR3 on Samsung GD-ies, which I’ve reached before. I’m wondering if this limits the maximum clock speed for RAM, since I imagine my i7 930 could push 3000 DDR3 at 1.5v but struggles above that. Next time I’ll check online stores to pick better chips that can reach 4.5 at 1.4v or higher, or ones with high uncore support for Xeon W or Gullfield. Also, do you have experience with Foxconn, EVGA, or DFi boards? They’re all reputable, except for the Foxconn line which has improved recently. The memclk ratio matters a lot, especially for Foxconn boards (I can buy a high-end blade for $35 if it meets specs). If I hit a bad ratio, I could flash a BIOS and tweak voltages.

P
pa55w0rd
Member
181
11-04-2016, 01:52 PM
#6
I began with a Giga UD4, upgraded to a 965EX ES, then a 970, X5690ES, and eventually settled on an R3F that I used for about eight years. I miss it, but I don’t. We enjoyed pushing our limits and had some intense debates.
P
pa55w0rd
11-04-2016, 01:52 PM #6

I began with a Giga UD4, upgraded to a 965EX ES, then a 970, X5690ES, and eventually settled on an R3F that I used for about eight years. I miss it, but I don’t. We enjoyed pushing our limits and had some intense debates.

P
PedroO_
Senior Member
522
11-04-2016, 03:14 PM
#7
Search for X5675s—they tend to be the top bin/$. I can't recall their exact spot in the thread, but someone here helped sort through many chips to find the best options. EDIT: it was @Ground
P
PedroO_
11-04-2016, 03:14 PM #7

Search for X5675s—they tend to be the top bin/$. I can't recall their exact spot in the thread, but someone here helped sort through many chips to find the best options. EDIT: it was @Ground

G
Gabokazu
Posting Freak
814
11-04-2016, 06:38 PM
#8
The ud4 seems to offer better performance compared to the ud3r when it comes to RAM clock speeds. The outer channel struggles around 2600 MHz, but sometimes it can hit 2800 MHz with a bit of tweaking. Running at 2800 requires a stable voltage source since the inner and middle channels only need about 1.9V. In contrast, the xxones aren’t as forgiving. For boosting speeds, you’ll likely need a more reliable power supply. The innermost part needs just 2.06V to reach 2800, while the outer one demands 2.06V too, but it’s tougher to maintain stability. I’m leaning toward a gulftown i7 like the 970, 980X, or 990X for better results. There are some listings for a W3670 at a reasonable price, which could be worth considering if you’re flexible. As for 45nm CPUs, they’re definitely appealing. I’m also searching for a Nehalem with a golden IMC to push higher clock speeds on ISCs that don’t use PSC. My current 930 maxes out around 3100 with a jump to 3200+, while my 950 tops at 3000–2900 PSC. I’m hoping for a Nehalem that can hit 3400+ with a 1:1 uncore memory clock, which would make all the previous high-PSC records irrelevant. That would be great since it’s easy to find such chips and they’re usually plentiful.

I’m curious about these Xeons though—do they typically come with high-binned IMCs? It seems like this particular sample might be a special find, as most high-speed RAM records are harder to get with 1:1 uncore memory clocks. The fact that the memclk ratio is mostly stable makes it worth exploring further.

For the NEHALEM, I’m hoping for something that can run at 3400+ with a 1:1 uncore memory clock. It would really shake up the existing benchmarks I’ve seen online.
G
Gabokazu
11-04-2016, 06:38 PM #8

The ud4 seems to offer better performance compared to the ud3r when it comes to RAM clock speeds. The outer channel struggles around 2600 MHz, but sometimes it can hit 2800 MHz with a bit of tweaking. Running at 2800 requires a stable voltage source since the inner and middle channels only need about 1.9V. In contrast, the xxones aren’t as forgiving. For boosting speeds, you’ll likely need a more reliable power supply. The innermost part needs just 2.06V to reach 2800, while the outer one demands 2.06V too, but it’s tougher to maintain stability. I’m leaning toward a gulftown i7 like the 970, 980X, or 990X for better results. There are some listings for a W3670 at a reasonable price, which could be worth considering if you’re flexible. As for 45nm CPUs, they’re definitely appealing. I’m also searching for a Nehalem with a golden IMC to push higher clock speeds on ISCs that don’t use PSC. My current 930 maxes out around 3100 with a jump to 3200+, while my 950 tops at 3000–2900 PSC. I’m hoping for a Nehalem that can hit 3400+ with a 1:1 uncore memory clock, which would make all the previous high-PSC records irrelevant. That would be great since it’s easy to find such chips and they’re usually plentiful.

I’m curious about these Xeons though—do they typically come with high-binned IMCs? It seems like this particular sample might be a special find, as most high-speed RAM records are harder to get with 1:1 uncore memory clocks. The fact that the memclk ratio is mostly stable makes it worth exploring further.

For the NEHALEM, I’m hoping for something that can run at 3400+ with a 1:1 uncore memory clock. It would really shake up the existing benchmarks I’ve seen online.

S
samsampp
Member
114
11-05-2016, 05:38 PM
#9
Uncertain about it... the only positive aspect was the clock speed. Outside of that, the build quality felt average. The R3F really stood out compared to others.
S
samsampp
11-05-2016, 05:38 PM #9

Uncertain about it... the only positive aspect was the clock speed. Outside of that, the build quality felt average. The R3F really stood out compared to others.

D
Devies
Member
185
11-07-2016, 09:48 AM
#10
I'm wondering about the difference between a rampage board and a lower-end X58A UD3/4/5 board in terms of OC. Bclk isn't really my main concern except for running x/e xeons; I still need to slow down for them to hit 230+ anyway. Also, I just noticed I've already broken the 2500 and 2400 MHz frequency records for the i7 930 and 950, respectively. The first one uses phase change, and I'm beating it by 600 MHz on some air coolers.
D
Devies
11-07-2016, 09:48 AM #10

I'm wondering about the difference between a rampage board and a lower-end X58A UD3/4/5 board in terms of OC. Bclk isn't really my main concern except for running x/e xeons; I still need to slow down for them to hit 230+ anyway. Also, I just noticed I've already broken the 2500 and 2400 MHz frequency records for the i7 930 and 950, respectively. The first one uses phase change, and I'm beating it by 600 MHz on some air coolers.

Pages (2): 1 2 Next