Overclock.net banner
41 - 60 of 62 Posts
While i agree the 33% increase in CU with proper memory speeds should net more than we’re seeing. A measly 10% uplift over the 7800xt in the best case scenario just doesn’t add up. I think it’s intentional market segmentation rather than poor quality silicon.
Yeah the 7900GRE is very reminiscent of the olD HD7870xt, Which was a cut down tahiti die, that you would normally find in the much quicker 7950's and 70s except it was masquerading a tier down as a 7870. The difference here is 7870XT was a decent chip with decent OC capabilities. Where it seems 7900GRE is far more limited by artificial limits put on it.
 
Well this GPU is not new and they already have a lot of time to fix it but sure saying that is a bug instead of artificial market segmentation is much better from a PR standpoint.

This is not the first time they do something like this:


That bug got fixed or not?
The 6800/6900 XT's limits were never fixed, but could be bypassed with MPT.
Early samples could only reach 3000 MHz however, more required a respin of the silicon
 
Discussion starter · #43 ·
AMD, Intel and Nvidia have always done this., non X, X and XT. Lower quality silicon gets lower limits. Just because DRAM chips can achieve certain speeds there is nothing saying the IMC/Fclock can handle them. Intel has locked down overclocking for decades unless you buy the more expensive chip. Nvidia just eliminated overclocking completely.
My 12400f at 5320mhz all core says otherwise but sure is not "officially supported" without paying for pretty much the same silicon but with some bits on the firmware changed and to me this is just outright extortion as corporation don't have ANY right of imposing restrictions AFTER selling the product to the customer.
Font Screenshot Software Technology Electronic device


And keep in mind that even if the IMC/Fclock can't handle 20gps memory there is chance that increasing the SoC voltage will work to make it stable but AMD locked the vbios against any mods so clearly they don't want anybody to experiment with this because most likely they have something nefarious to hide.
 
Discussion starter · #47 ·
Running 2550 myself rn might try 2600 but 2650 crashed my pc when in a highly populated lobby in VRChat on my index. 2600 might work though since it took a long ass time before it crashed at 2650
Ok good to see they unlocked the vram clock limit but AMD is doing this for PR reasons and not out of the kindness of their heart so take a look at hwinfo to see what the SoC voltage and memory voltage to make sure they are not ****ing around with that for some nefarious placebo effect strategy:

Rectangle Font Screenshot Parallel Number

This is from the 7900xtx sensors so make sure that the SoC voltage and memory voltage are similar and not lower.:unsure:
 
Would like to see core clocks hit over 2800+mhz but I'm guessing that they won't allow that. The most I have seen so far is people hitting around 2700mhz core and 2600mhz memory. Which will give it a nice boost but won't cannibalize the XTs. IPC wise there's not much of an increase over the 6000 series. My OC'ed 6950XT pulls solid numbers close to stock 7900XT. Based on the 7900GRE specs if you could get the core clock up closer to 2900mhz along with that memory overclock you should be getting 7900XT performance if not a little bit more for $500-550. Not bad.
 
Sapphire pulse getting 2587 MT/s, but failing at whatver was closest to 2650 (2637?)

However, anyone know about limits on linux. I just updated my arch install and the limits in /sys/class/drm/card1/device/pp_od_clk_voltage are still 2316 MT/s (1158x2)


Code:
cat /sys/class/drm/card1/device/pp_od_clk_voltage

OD_SCLK:
0: 500Mhz
1: 2430Mhz
OD_MCLK:
0: 97Mhz
1: 1125MHz
OD_VDDGFX_OFFSET:
0mV
OD_RANGE:
SCLK: 500Mhz 2803Mhz
MCLK: 97Mhz 1158Mhz
VDDGFX_OFFSET: -350mv 0mv
 
I'm still running on an overclocked vega 56. I saw the 7900 gre and started drooling because I felt like it had the OC potential vega had. Bummer this software lock hasn't been fixed yet. I suspect it's only a matter of time.

One way to overclock the vega was a registry mod. Has anyone looked at that yet?
 
My sapphire nitro+ GRE green screens at 2475mhz. Can run fast timings at 2400mhz. odd that the memory chips are rated for 2500 but they can't actually run at that. 980mv on the core is stable.

Mem SOC 0.898v
SOC Current 36A

Mem volt VDDIO 1.35v
Mem volt VDDCI 0.834v

Mem current VDDIO 22.3A
Mem Current VDDCI 12.6A

Seen a few people on reddit not able to get much more out of their memory while others can hit quite high.
 
My sapphire nitro+ GRE green screens at 2475mhz. Can run fast timings at 2400mhz. odd that the memory chips are rated for 2500 but they can't actually run at that. 980mv on the core is stable.

Mem SOC 0.898v
SOC Current 36A

Mem volt VDDIO 1.35v
Mem volt VDDCI 0.834v

Mem current VDDIO 22.3A
Mem Current VDDCI 12.6A

Seen a few people on reddit not able to get much more out of their memory while others can hit quite high.
Have you tried 2500 MHz? This might be a case of 2475 MHz running with a timing strap intended for a slower speed, and the strap switching at 2500 MHz specifically.
 
  • Rep+
Reactions: Spladam
Necro bump because I'm really curious about how the GRE would perform with this this procedure:
It's a full die and has weirdly low current limits, it could benefit a lot from extra current. Could be a sleeper card with this method.
No matter what's done with power/current limits, the low clock limit is going to continue to be a barrier unfortunately.
 
GPU or VRAM clock? Is the limit at driver or bios level?
Have a look at 7900XT current limits

Blue Photograph White Text Screenshot


For Core/SoC/USR, at least they're in hwinfo list. Now look at 7900GRE:

Blue Photograph White Text Font


That's quite the difference. There's also two other rails that don't have limits mentioned in hwinfo, Vmem and Vddci but last one doesn't matter that much. You can now offset the current for all these rails.
I was able to get my (7800XT) VRAM to 2850MHz, benchmarked at that speed, and seemed to pass memtest_vulkan 5 minutes test at 2800MHz, though I'd back off a bit, 2750MHz or so. LLMs were fine in a short test I made.
If software allows setting 2800MHz then it will probably work at that speed provided you offset the current to help with that.
 
GPU or VRAM clock? Is the limit at driver or bios level?
Have a look at 7900XT current limits

View attachment 2710136

For Core/SoC/USR, at least they're in hwinfo list. Now look at 7900GRE:

View attachment 2710137

That's quite the difference. There's also two other rails that don't have limits mentioned in hwinfo, Vmem and Vddci but last one doesn't matter that much. You can now offset the current for all these rails.
I was able to get my (7800XT) VRAM to 2850MHz, benchmarked at that speed, and seemed to pass memtest_vulkan 5 minutes test at 2800MHz, though I'd back off a bit, 2750MHz or so. LLMs were fine in a short test I made.
If software allows setting 2800MHz then it will probably work at that speed provided you offset the current to help with that.
Looking at TPU OC results in their reviews, they were able to get GRE right up to the 2800Mhz limit in-game which is why I'm thinking power is not the primary limiting factor.

Computer hardware


Raising limits with EVC could eke out a little more performance in some high-power applications but it won't be anything like 7900XT where EVC can be the difference between 2900Mhz (power-limited) and 3200Mhz or even 3400-3500 for benching. The GRE is just too clock-limited and it's too bad because you're right that otherwise it could be a strong contender for OC.
 
But Vcore has exactly 50% less current limit vs 7900XT. That has to play into it barely reaching 2800MHz. I doubt it has some kind of weird BIOS clock limit.
Curious how LACT interface looks on Linux, for 7900GRE, does it allow setting clock higher than 2800MHz? Like 3300MHz for example?
The low clocks could simply be enforced with a low current limit, or VI curve. I'd certainly try the current gain mod if I had one.
 
But Vcore has exactly 50% less current limit vs 7900XT. That has to play into it barely reaching 2800MHz. I doubt it has some kind of weird BIOS clock limit.
Curious how LACT interface looks on Linux, for 7900GRE, does it allow setting clock higher than 2800MHz? Like 3300MHz for example?
The low clocks could simply be enforced with a low current limit, or VI curve. I'd certainly try the current gain mod if I had one.
It does appear to have a weird BIOS limit unfortunately. Radeon Overdrive limits have been stored in VBIOS for a long long time; one used to be able to bypass them with registry PowerPlayTables but AMD has locked down / removed that capability on newer cards.

Here are 7900GRE HWBot results including submissions from some of the best OC'ers in the world and not one of them got past the 2.8Ghz limit 😕
AMD Radeon RX 7900 GRE @ HWBOT
Fwiw it's the same situation with 7600 (3Ghz limit) and 7600XT (3.2Ghz limit) and afaik those have not been bypassed either.

I'll do some digging on Discord but I do believe EVC has been tried on GRE.
I'm not saying the current limit is irrelevant, but I am saying there is a lot of evidence that even if one were to raise PL the card wouldn't go over 2.8Ghz.
 
41 - 60 of 62 Posts