F5F Stay Refreshed Hardware Desktop Bottleneck in RAM speed reaching 30,000 MB/s for AMD FX processors

Bottleneck in RAM speed reaching 30,000 MB/s for AMD FX processors

Bottleneck in RAM speed reaching 30,000 MB/s for AMD FX processors

Pages (3): Previous 1 2 3 Next
D
138
06-28-2017, 08:30 PM
#11
it looks like they might be talking about DDR3-1866 specifications or something similar. checking other processors shows different numbers—i7 2600 works with DDR3-1333 at 21.3GB/s, i7 3770 with DDR3-1600 at 25.6GB/s. amd fx supports DDR3-1866 and claims 29.9GB/s, which doesn’t match the real data. possibly the CPU isn’t powerful enough, so I increased the clock to 4700mhz for testing and got around 30,000MB/s. maybe the motherboard affects timing and reduces bandwidth.
D
DiamondKing126
06-28-2017, 08:30 PM #11

it looks like they might be talking about DDR3-1866 specifications or something similar. checking other processors shows different numbers—i7 2600 works with DDR3-1333 at 21.3GB/s, i7 3770 with DDR3-1600 at 25.6GB/s. amd fx supports DDR3-1866 and claims 29.9GB/s, which doesn’t match the real data. possibly the CPU isn’t powerful enough, so I increased the clock to 4700mhz for testing and got around 30,000MB/s. maybe the motherboard affects timing and reduces bandwidth.

J
Juan2610
Posting Freak
875
06-29-2017, 01:08 AM
#12
It all becomes clear when you understand how FX operates. 2400 CPU/NB is the typical optimal setting, offering little extra gain beyond that level. This is suitable for a regular desktop setup. Higher speeds are possible, but they require more complex adjustments. For peak efficiency, especially in overclocking scenarios, you can achieve better results. The example I tested earlier is a good reference: https://hwbot.org/submission/4664632_bon...0sec_938ms. For other projects, boosting RAM speed while increasing CPU/NB and core speeds can also help. Bones' SuperPi achieved a 9-second run with 141ms on an FX-8320. The main issue with FX performance is its lengthy cache pipelines, which limit efficiency. However, there are methods to enhance its speed in most cases: https://community.hwbot.org/topic/233924...ware-2024/. Look for the "BD patch and Conditioner (STILT Tools)" highlighted in black, as it can improve results. You might want to try BDC R1.03B or similar updates.
J
Juan2610
06-29-2017, 01:08 AM #12

It all becomes clear when you understand how FX operates. 2400 CPU/NB is the typical optimal setting, offering little extra gain beyond that level. This is suitable for a regular desktop setup. Higher speeds are possible, but they require more complex adjustments. For peak efficiency, especially in overclocking scenarios, you can achieve better results. The example I tested earlier is a good reference: https://hwbot.org/submission/4664632_bon...0sec_938ms. For other projects, boosting RAM speed while increasing CPU/NB and core speeds can also help. Bones' SuperPi achieved a 9-second run with 141ms on an FX-8320. The main issue with FX performance is its lengthy cache pipelines, which limit efficiency. However, there are methods to enhance its speed in most cases: https://community.hwbot.org/topic/233924...ware-2024/. Look for the "BD patch and Conditioner (STILT Tools)" highlighted in black, as it can improve results. You might want to try BDC R1.03B or similar updates.

I
iFreestyleCab
Junior Member
37
07-05-2017, 06:42 AM
#13
it doesn't appear to function well when switching from 2400 to 2600 MHz—it actually made things slower in games and didn't improve L3 cache performance. The AIDA64 reached 30000mb/s only with 2600 MHz, which is unusual. Adding +200mhz didn't boost speeds much, and at 2700 MHz it hit 32000mb/s, which seems odd. The extra +100mhz only added about 2000mb/s in PyPrime, so I’m giving up on that. My setup runs at 2600 MHz without issues, but I’m experimenting with secondary timings now. You're correct about the bandwidth gap after 2400 MHz—it's mostly about latency fixes. Running NB/HT at 2600 MHz still seems worthwhile for daily use.
I
iFreestyleCab
07-05-2017, 06:42 AM #13

it doesn't appear to function well when switching from 2400 to 2600 MHz—it actually made things slower in games and didn't improve L3 cache performance. The AIDA64 reached 30000mb/s only with 2600 MHz, which is unusual. Adding +200mhz didn't boost speeds much, and at 2700 MHz it hit 32000mb/s, which seems odd. The extra +100mhz only added about 2000mb/s in PyPrime, so I’m giving up on that. My setup runs at 2600 MHz without issues, but I’m experimenting with secondary timings now. You're correct about the bandwidth gap after 2400 MHz—it's mostly about latency fixes. Running NB/HT at 2600 MHz still seems worthwhile for daily use.

S
Skywonder216
Member
171
07-05-2017, 03:26 PM
#14
my board is designed for 2200 yet with a solid chip I’m confident I can reach 2900+ stable using tri-channel mode. Trying 6 dimms doesn’t show the full potential, but I’ll need to recheck with CFR. With 5 dims it boots and recognizes 2400, though I expect 2400 won’t hit 2400 again. The 2600 level requires a Hyundai CFR, but it only works for channels when there are 5 dims on. It seems comparable to Samsung D die designs, just less stable above 2V and best in cold temperatures. So I’d suggest going with 2DPC for a higher chance of reaching 2600+. These make sense if you don’t see similar specs elsewhere. On the other hand, you might achieve tighter primaries at 1.8-1.9V, which is where Samsung’s 4Gbit models usually cap. Any comments on DRAM voltage—most recommendations seem based on outdated info, and people didn’t really test real voltages back then. It probably faded with DDR5. This might be a bit outdated, though. My experience with the 970 Extreme3 was limited to FX 4130; I maxed out the CPU and called it quits for CPU LLC. I’m not sure about voltage under load, but I can’t say much. For FX, I’ll probably just take a screenshot of a 2700 Windows board if I get one working again—it might be stuck at RAM sticks. After replacing the socket cover (it won’t go faster than 2400), I’m still too curious about X58. My top pick for AMD would be LLano FM1, and it’ll take some time before I dive back into FX testing.**
S
Skywonder216
07-05-2017, 03:26 PM #14

my board is designed for 2200 yet with a solid chip I’m confident I can reach 2900+ stable using tri-channel mode. Trying 6 dimms doesn’t show the full potential, but I’ll need to recheck with CFR. With 5 dims it boots and recognizes 2400, though I expect 2400 won’t hit 2400 again. The 2600 level requires a Hyundai CFR, but it only works for channels when there are 5 dims on. It seems comparable to Samsung D die designs, just less stable above 2V and best in cold temperatures. So I’d suggest going with 2DPC for a higher chance of reaching 2600+. These make sense if you don’t see similar specs elsewhere. On the other hand, you might achieve tighter primaries at 1.8-1.9V, which is where Samsung’s 4Gbit models usually cap. Any comments on DRAM voltage—most recommendations seem based on outdated info, and people didn’t really test real voltages back then. It probably faded with DDR5. This might be a bit outdated, though. My experience with the 970 Extreme3 was limited to FX 4130; I maxed out the CPU and called it quits for CPU LLC. I’m not sure about voltage under load, but I can’t say much. For FX, I’ll probably just take a screenshot of a 2700 Windows board if I get one working again—it might be stuck at RAM sticks. After replacing the socket cover (it won’t go faster than 2400), I’m still too curious about X58. My top pick for AMD would be LLano FM1, and it’ll take some time before I dive back into FX testing.**

C
Cupcake_Rose
Posting Freak
844
07-06-2017, 12:00 AM
#15
Regarding BDC, it functions effectively in most situations—though not every case. Proper usage is essential. There are certain factors that don’t influence performance significantly, but overall it still delivers results. You can push up to 2600 CPU or NB if desired for optimal performance; however, stock or mild configurations typically limit this due to Vishera chip limitations at higher speeds. Subzero cooling helps avoid issues. The ideal range is around 2500 CPU/NB, and a 1.20v boost is sufficient to reach 2400 or even 2500. Avoid pushing LCC beyond 130% or 140%—my setup used 120% and it performed well without stress.
C
Cupcake_Rose
07-06-2017, 12:00 AM #15

Regarding BDC, it functions effectively in most situations—though not every case. Proper usage is essential. There are certain factors that don’t influence performance significantly, but overall it still delivers results. You can push up to 2600 CPU or NB if desired for optimal performance; however, stock or mild configurations typically limit this due to Vishera chip limitations at higher speeds. Subzero cooling helps avoid issues. The ideal range is around 2500 CPU/NB, and a 1.20v boost is sufficient to reach 2400 or even 2500. Avoid pushing LCC beyond 130% or 140%—my setup used 120% and it performed well without stress.

L
liflem
Member
163
07-06-2017, 07:19 AM
#16
they mention 1.725v is acceptable but 1.65v is suitable for daily use with cpu llc. the problem here is sudden voltage increases, and keeping it high could harm your CPU over time. most gb motherboards have one core per module, even basic ones without VRM heatsinks. a friend of mine had issues with his ASRock H61M board and lost trust in ASRock, so he voided it. FX is no longer appealing mainly due to the absence of AVX2 support. the top DDR3 options are Intel 4gen boards, featuring AVX2, PCIe 3.0, and high XMP RAM designed for those platforms. you can also add a TPM 2.0 module, though I’m not sure if it’s necessary.
L
liflem
07-06-2017, 07:19 AM #16

they mention 1.725v is acceptable but 1.65v is suitable for daily use with cpu llc. the problem here is sudden voltage increases, and keeping it high could harm your CPU over time. most gb motherboards have one core per module, even basic ones without VRM heatsinks. a friend of mine had issues with his ASRock H61M board and lost trust in ASRock, so he voided it. FX is no longer appealing mainly due to the absence of AVX2 support. the top DDR3 options are Intel 4gen boards, featuring AVX2, PCIe 3.0, and high XMP RAM designed for those platforms. you can also add a TPM 2.0 module, though I’m not sure if it’s necessary.

Z
zrb2004
Member
72
07-06-2017, 08:11 AM
#17
What BDC is involved? I'm checking NB/HT 2700mhz with a 1.35V supply. My CPU, NBM, and RAM are at 130% capacity. I plan to lower it to 120%. So far it seems unaffected since I didn't apply that much voltage. I suspect the issue isn't with the CPU or RAM but with the NB itself. When I set it to 2400mhz it runs at 1.4-1.45V, and pushing it higher to 2600mhz gives 1.45-1.5V. I'm keeping voltage between 1.3-1.35V for 2400mhz and 1.25V for 2600mhz. Noticing something similar to CPU LLC, it drops when I increase load beyond a point—then medium LLC kicks in to stabilize. With NB(IMC), however, the voltage keeps rising; at 1.3V it stays around 1.298-1.300 and at 1.35V it reaches 1.350-1.360V. It looks like the sensor might be reading incorrectly under certain conditions.
Z
zrb2004
07-06-2017, 08:11 AM #17

What BDC is involved? I'm checking NB/HT 2700mhz with a 1.35V supply. My CPU, NBM, and RAM are at 130% capacity. I plan to lower it to 120%. So far it seems unaffected since I didn't apply that much voltage. I suspect the issue isn't with the CPU or RAM but with the NB itself. When I set it to 2400mhz it runs at 1.4-1.45V, and pushing it higher to 2600mhz gives 1.45-1.5V. I'm keeping voltage between 1.3-1.35V for 2400mhz and 1.25V for 2600mhz. Noticing something similar to CPU LLC, it drops when I increase load beyond a point—then medium LLC kicks in to stabilize. With NB(IMC), however, the voltage keeps rising; at 1.3V it stays around 1.298-1.300 and at 1.35V it reaches 1.350-1.360V. It looks like the sensor might be reading incorrectly under certain conditions.

P
perrinoid
Member
137
07-14-2017, 01:41 AM
#18
...NB 2700mhz was not stable with 1.35v, atleast i did feel extra stutters in cs2 so i don't think its stable. btw on PYPrime 2.0 i get those: NB 2400 = 30sec NB 2600 = 29sec and here is another strange thing: DDR3-2133mt/s with NB(IMC) 2400mhz gets 30000mb/s read DDR3-2400mt/s with NB(IMC) 2400mhz gets 28000mb/s read so if i keep ram the same lets say DDR3-2400mt/s and i kinda compare NB(IMC) frequency: NB(IMC) 2400mhz Memory : Read 28000mb/s | Write 19000mb/s | Copy 27620mb/s | Letancy 57ns L3 Cache: Read 133.79GB/s | Write 39410mb/s | Copy 55480mb/s | Letancy 31ns Vs NB(IMC) 2600mhz Memory : Read 30648mb/s | Write 19549mb/s | Copy 28086mb/s | Letancy 56.8ns L3 Cache: Read 141.48GB/s | Write 41313mb/s | Copy 59996mb/s | Letancy 29.4ns
P
perrinoid
07-14-2017, 01:41 AM #18

...NB 2700mhz was not stable with 1.35v, atleast i did feel extra stutters in cs2 so i don't think its stable. btw on PYPrime 2.0 i get those: NB 2400 = 30sec NB 2600 = 29sec and here is another strange thing: DDR3-2133mt/s with NB(IMC) 2400mhz gets 30000mb/s read DDR3-2400mt/s with NB(IMC) 2400mhz gets 28000mb/s read so if i keep ram the same lets say DDR3-2400mt/s and i kinda compare NB(IMC) frequency: NB(IMC) 2400mhz Memory : Read 28000mb/s | Write 19000mb/s | Copy 27620mb/s | Letancy 57ns L3 Cache: Read 133.79GB/s | Write 39410mb/s | Copy 55480mb/s | Letancy 31ns Vs NB(IMC) 2600mhz Memory : Read 30648mb/s | Write 19549mb/s | Copy 28086mb/s | Letancy 56.8ns L3 Cache: Read 141.48GB/s | Write 41313mb/s | Copy 59996mb/s | Letancy 29.4ns

K
kleinne_meid
Member
228
07-15-2017, 06:45 AM
#19
DDR3-2133mt/s NB 2200 = 26000mb/s NB 2400 = 30000mb/s NB 2600 = 30000mb/s NB 2700 = 30000mb/s just guessing on this one i didn't test it Vs DDR3-2400mt/s NB 2200 = 26000mb/s NB 2400 = 28000mb/s why is this 28000 tho NB 2600 = 30000mb/s NB 2700 = 32000mb/s anyway i will try to get NB to 2800 to see if it scales to 34000mb/s ram 2388mt/s with NB at 2800 1.4v i get 33000mb/s read so... it does scale but for some reason at NB2400 i get only 28000mb/s with Ram at 2400mt/s maybe the motherboard does something to the hidden timings that drops -2000mb/s
K
kleinne_meid
07-15-2017, 06:45 AM #19

DDR3-2133mt/s NB 2200 = 26000mb/s NB 2400 = 30000mb/s NB 2600 = 30000mb/s NB 2700 = 30000mb/s just guessing on this one i didn't test it Vs DDR3-2400mt/s NB 2200 = 26000mb/s NB 2400 = 28000mb/s why is this 28000 tho NB 2600 = 30000mb/s NB 2700 = 32000mb/s anyway i will try to get NB to 2800 to see if it scales to 34000mb/s ram 2388mt/s with NB at 2800 1.4v i get 33000mb/s read so... it does scale but for some reason at NB2400 i get only 28000mb/s with Ram at 2400mt/s maybe the motherboard does something to the hidden timings that drops -2000mb/s

X
XaVaTaR
Member
77
07-15-2017, 08:41 PM
#20
You meant to clarify that "LCC" refers to the board's DIGI configuration options. For DIGI settings, understand that adjusting the voltage determines how much extra power is supplied to components like RAM. For instance, if you manually set your RAM voltage to 1.65V and then increase it to 110%, you're applying 110% of that value. The same applies if you select 120%. These adjustments are mainly used for overclocking or X-overclocking, allowing you to exceed the default BIOS limits. Be mindful of these settings because they’re designed for advanced users who want extra voltage control.

Using a calculator helps ensure accuracy. Calculate what actual voltage you're delivering based on your manual voltage choice and the percentage you select. Remember, the board is built for overclocking and can handle significant voltage increases if configured properly.

If your CPU or other components run too hot, especially under load, setting them too high can cause issues—typically around 65°C in most cases. Disabling the x87 instruction set (BCD) can improve performance, depending on your workload. You’ll notice a difference when comparing runs with and without BDC enabled.

Some graphics chips perform better at higher CPU speeds, so if yours supports it, you’re in good hands. Just avoid excessive voltage for that setting to prevent overheating or instability.
X
XaVaTaR
07-15-2017, 08:41 PM #20

You meant to clarify that "LCC" refers to the board's DIGI configuration options. For DIGI settings, understand that adjusting the voltage determines how much extra power is supplied to components like RAM. For instance, if you manually set your RAM voltage to 1.65V and then increase it to 110%, you're applying 110% of that value. The same applies if you select 120%. These adjustments are mainly used for overclocking or X-overclocking, allowing you to exceed the default BIOS limits. Be mindful of these settings because they’re designed for advanced users who want extra voltage control.

Using a calculator helps ensure accuracy. Calculate what actual voltage you're delivering based on your manual voltage choice and the percentage you select. Remember, the board is built for overclocking and can handle significant voltage increases if configured properly.

If your CPU or other components run too hot, especially under load, setting them too high can cause issues—typically around 65°C in most cases. Disabling the x87 instruction set (BCD) can improve performance, depending on your workload. You’ll notice a difference when comparing runs with and without BDC enabled.

Some graphics chips perform better at higher CPU speeds, so if yours supports it, you’re in good hands. Just avoid excessive voltage for that setting to prevent overheating or instability.

Pages (3): Previous 1 2 3 Next