Overclock.net banner
Cancel
Create thread New Forums

[Official] NVIDIA RTX 5090 Owner's Club

Tags
5090 blackwell nvidia rtx
44 reading
Jump to Latest
14,821 - 14,840 of 18,210 Posts
#14,823 ·
I dont know about Gigabyte-GPUs...
Is there a GPU with ´side connection´?
 
Save
Reply Quote Rep+ Rep+
#14,826 ·
#14,827 ·
Save
Reply Quote Rep+ Rep+
#14,828 · (Edited)
It’s hilarious that before the 5090 launch Gigabyte already planned to hide connector (for Stealth series) by leaving room for extra power socket on the side, so they wouldn’t need to revise the PCB. They simply moved the actual socket to the side and called it day. Making it possible 1200W with two 12vhpwr which is way safer for shunt mod
 
#14,829 · (Edited)
motivman said:
its not worth the switch, but I will keep the EKWB block. At least I now have access to the bios switch, and it seems higher quality. I prefer the port placement on the Bykski though... LOL
Click to expand...
Way more safer on Bykski. To get better temps you need more pumps. I've got a spare D5 lying around just bought a housing for it and will add my 4th pump. Should get way more head pressure hitting those cold plates.
 
Save
Reply Quote Rep+ Rep+
#14,830 ·
Snorkel said:
I don't care it performs well or not :p I just want to remove power limit without solder my card
Click to expand...
But if you have to pull 700-800w to get the same performance as 600w on GB bios, doesn't make sense to use it.
 
#14,831 · (Edited)
Obidkhan said:
View attachment 2719148
You are right. This is 5090 Aorus Master PCB which looks exact the same as Aorus Xtreme. 50A rated MOSFETs and 24 Phase VRM + 7 Phase Memory. Does Astral’s 80A rated MOSFETs make difference for OC stability? I still think old 80A would be better than new 50A
Click to expand...
No, it shouldn't make a notable diff imo. Well the thing with power stages is that the amp rating doesn't say much. The 50a part could be better at transients and such but slightly worse in efficiency than the older 80a part at the end of its efficiency curve, but efficiency is only a concern with VRM cooling and ig the total available power budget as less efficient VRMs will lose more power that would otherwise be sent to the GPU.
A 5090 will pull like 700-800a sustained(30-35a per power stage) when not power limited at the most in most gaming applications which the 50a part should be able to handle just fine.
 
  • Rep+
Reactions: Obidkhan
Save
Reply Quote Rep+ Rep+
#14,832 ·
Jwick said:
No, it shouldn't make a notable imo. Well the thing with power stages is that the amp rating doesn't say much. The 50a part could be better at transients and such but slightly worse in efficiency than the older 80a part at the end of its efficiency curve, but efficiency is only a concern with VRM cooling and ig the total available power budget as less efficient VRMs will lose more power that would otherwise be sent to the GPU.
Click to expand...
Thx for clarification. Best 3DMark scores mostly gained using Astrals with Gigabyte Xtreme BIOS so I wondered if that’s due to 80A MOSFETs
 
#14,833 ·
i did NOT use LM on card, and i will take your advise and not clean the card fully like i did prior. Ill do just like you said and clean off only what is needed to have clean surfaces and thats it.
 
  • Rep+
Reactions: mirkendargen
Save
Reply Quote Rep+ Rep+
#14,834 ·
Obidkhan said:
Thx for clarification. Best 3DMark scores mostly gained using Astrals with Gigabyte Xtreme BIOS so I wondered if that’s due to 80A MOSFETs
Click to expand...
This is a power stage spec used by 4090 FE ( 70a MP86957 ).
Plot


This is vishay sic654- found on Tuf 5090s.

White Text Plot Font Diagram


You can't compare the results directly because the voltage is different, but yea it is marginal although the MPS part loses about 25% less power than vishay at 30a.

Well, bin, cooling and bios are the most pertinent to scores than VRM capability tbh, although i suppose the FEs may be a bit subpar in its capability due to limited space.
 
  • Rep+
Reactions: Papusan and Obidkhan
Save
Reply Quote Rep+ Rep+
#14,835 ·
Jwick said:
This is a power stage spec used by 4090 FE ( 70a MP86957 ).
View attachment 2719176

This is vishay sic654- found on Tuf 5090s.

View attachment 2719177

You can't compare the results directly because the voltage is different, but yea it is marginal although the MPS part loses about 25% less power than vishay at 30a.

Well, bin, cooling and bios are the most pertinent to scores than VRM capability tbh, although i suppose the FEs may be a bit subpar in its capability due to limited space.
Click to expand...
Very well could just be the FE bios if something is holding them back. Obviously no way to prove it of course.

Just a lot of odd inconsistencies with 50 series. Bios and drivers.
 
#14,836 ·
Jwick said:
This is a power stage spec used by 4090 FE ( 70a MP86957 ).
View attachment 2719176

This is vishay sic654- found on Tuf 5090s.

View attachment 2719177

You can't compare the results directly because the voltage is different, but yea it is marginal although the MPS part loses about 25% less power than vishay at 30a.

Well, bin, cooling and bios are the most pertinent to scores than VRM capability tbh, although i suppose the FEs may be a bit subpar in its capability due to limited space.
Click to expand...
I think when shunting and you remove the power limit, mayor limiter is bin quality, then cooling, then VBIOS (assuming no XOC VBIOS). Now with XOC, then is bin quality and then cooling.

Take for example my MSI VG, when shunted it max was about just 3322-3332Mhz, at 1.07V. Even if wanted to give it more power it would just crash. I guess with the XOC VBIOS it would scale more until 1.125V.

But with that max voltage when shunted, even with a WC or LN2 or something it would never score higher than any similar shunted card that does 3350Mhz, or 3400Mhz, or even some that do 3500Mhz.

Bin quality is just the most important (and random) decider of performance.

To think that my TUF is even slower, I think some FEs can be better than both TUF and VG easily if you get a good bin.
 
  • Rep+
Reactions: domdtxdissar
Save
Reply Quote Rep+ Rep+
#14,837 ·
Panchovix said:
I think when shunting and you remove the power limit, mayor limiter is bin quality, then cooling, then VBIOS (assuming no XOC VBIOS). Now with XOC, then is bin quality and then cooling.

Take for example my MSI VG, when shunted it max was about just 3322-3332Mhz, at 1.07V. Even if wanted to give it more power it would just crash. I guess with the XOC VBIOS it would scale more until 1.125V.

But with that max voltage when shunted, even with a WC or LN2 or something it would never score higher than any similar shunted card that does 3350Mhz, or 3400Mhz, or even some that do 3500Mhz.

Bin quality is just the most important (and random) decider of performance.

To think that my TUF is even slower, I think some FEs can be better than both TUF and VG easily if you get a good bin.
Click to expand...
Panchovix said:
I think when shunting and you remove the power limit, mayor limiter is bin quality, then cooling, then VBIOS (assuming no XOC VBIOS). Now with XOC, then is bin quality and then cooling.

Take for example my MSI VG, when shunted it max was about just 3322-3332Mhz, at 1.07V. Even if wanted to give it more power it would just crash. I guess with the XOC VBIOS it would scale more until 1.125V.

But with that max voltage when shunted, even with a WC or LN2 or something it would never score higher than any similar shunted card that does 3350Mhz, or 3400Mhz, or even some that do 3500Mhz.

Bin quality is just the most important (and random) decider of performance.

To think that my TUF is even slower, I think some FEs can be better than both TUF and VG easily if you get a good bin.
Click to expand...
for 3300mhz, your vanguard only scores 17.5k right in SN? which is low considering you can do 16.5k+ with about 2900mhz. 2900 to 3300 is more than 10% increase in clocks, so it should at least get 18k right? unless there is some internal current limit or clock stretching. Remember, the shunts are only raising the input 12v current limit. There can still be chip current limit of some sort present that may differ between chips like with cold bug throttling.
 
Save
Reply Quote Rep+ Rep+
#14,838 ·
Jwick said:
for 3300mhz, your vanguard only scores 17.5k right in SN? which is low considering you can do 16.5k+ with about 2900mhz. 2900 to 3300 is more than 10% increase in clocks, so it should at least get 18k right? unless there is some internal current limit or clock stretching. Remember, the shunts are only raising the input 12v current limit. There can still be chip current limit of some sort present that may differ between chips like with cold bug throttling.
Click to expand...
Yeh, but more like from 3050Mhz to 3300Mhz, so almost 10%. The difference is I just used an offset since voltage keeps pegged at 1.07V, vs without shunts where voltage is variable and then you want to maximize the clocks per voltage point.

I didn't see a limiter at least, except Vrel.

But yeah 17500 is also quite disappointing. It is like 3.5% more performance for ~200W more. The mayor advantage tbh was the card wasn't getting early power limited at 500-600W range.

So imagine the TUF shunted, it would score like 17000 and that's a big IF :ROFLMAO:
 
Save
Reply Quote Rep+ Rep+
#14,839 ·
Panchovix said:
Yeh, but more like from 3050Mhz to 3300Mhz, so almost 10%. The difference is I just used an offset since voltage keeps pegged at 1.07V, vs without shunts where voltage is variable and then you want to maximize the clocks per voltage point.

I didn't see a limiter at least, except Vrel.

But yeah 17500 is also quite disappointing. It is like 3.5% more performance for ~200W more. The mayor advantage tbh was the card wasn't getting early power limited at 500-600W range.

So imagine the TUF shunted, it would score like 17000 and that's a big IF :ROFLMAO:
Click to expand...
Did you need 3050 for 16.5k cause i did it with <2900 and i was still power limited throughout:poop:. I am actually curious what this chip will do with shunts because it is literally at 600w with only .9v. at 2800. I can't shunt it on this cooler even i wanted to because it hits 70c+ at only 600w..
 
Save
Reply Quote Rep+ Rep+
#14,840 ·
newls1 said:
i did NOT use LM on card, and i will take your advise and not clean the card fully like i did prior. Ill do just like you said and clean off only what is needed to have clean surfaces and thats it.
Click to expand...
I personally scrubbed the hell out of my 5090, it had those white oem fuji style pads and the residue was on another level. It just wouldn’t come off. I had to clean the hell out of the card. It was all but submersed in alcohol, I had alcohol under the core and everything probably used about 50 Qtips and paper towels on it. It was finally spotless upon mounting the block. Crazy that your card died man. Hope your luck is better next time.
 
Save
Reply Quote Rep+ Rep+
14,821 - 14,840 of 18,210 Posts
About this Discussion
18K Replies
579 Participants
Last post:
synergon
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling