• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Radeon RX 9070 Can be Flashed with RX 9070 XT BIOS to Increase Power Limits and Clocks

aah, i remember flashing my 5850 with a 5870 bios, ah the memories :toast:
 
I remember flashing my HD2900 Pro to HD2900XT back in the day.
 
I would say that it doesn't even need GDDR7, just faster GDDR6.
This is old news, but it seems that everyone is content with using 20 Gbps GDDR6. Hey, whatever is cost-effective right?
If it had 24 Gbps GDDR6 we would surely see some improvements.

I'm not saying it isn't bottlenecked to some degree but that fast cache voodoo actually works, the proof is that the 9070 XT, a card with 645 GB/s (slighty higher than the 7800 XT, lol) bandwidth is able to get (quite) close to cards that have (sometimes significantly) higher bandwidths:
5070 Ti - 896 GB/s
4080 - 717 GB/s
4080 Super - 736 GB/s
7900 XTX - 960 GB/s
5080 - 960 GB/s (this one is clearly stronger but not due to bandwidth alone)
For starters the current gap is 11-13%. 24 Gbps at best would give it a 20% increase, out of 35%(assuming the 128 ROPs don't become the new bottleneck, which I don't think they will) that would still about 10% on the table, plus whatever latency increase turbo-clocked GDDR6 has, so I'd still go for 28G7 .
 
I will flash one of our leakiest cards, overclock it, and run it through the test suite.

Thanks! I totally am not abusing the quote system to falsely represent anyone and ask them to do us a solid!
(Please overclock the core first; I'm kind of curious how high it can scale...and also it should need less ram bw so it should be able to clock pretty high w/o even touching it and still be fine).

Remember how 7700xt clocked to 3113mhz and 7800xt ~2930 at best? Yeah, that was fun(ny). I'm expecting that type of result (unless there is a bios clock lock).
 
Last edited:
Ahh the memories did this on so many radeons going all the back to the 8500 LE
 
them losing their soul since becoming more like nVIDIA.
I'd say they're more nVidia than nVidia themselves.

What AMD need is snatching as much marketshare as possible at lowest prices possible. No matter how devastating the short-term and mid-term damage is. It's been more than five years since they had any luxury to do otherwise but still here we are with less than 10% of what's essentially flooding the market with enough cheap stuff. 9070, even in the best case scenario, bought at 550+VAT, fully unlocked to match the 9070 XT wattage and overclocked to roughly match it in performance, is still not a great value card. Just an okay value one. Okay isn't enough in this case.

To make matters worse, no way you can get this GPU for MSRP. Not a 100% chance you can flash the BIOS and get lucky. We, buyers, want AMD to actually compete, not to squat around like an absolute monkey. Feels like they'll be reintroducing 7900 XT(X) at ever so slowly lowering launch prices without doing anything better than that.
 
For starters the current gap is 11-13%. 24 Gbps at best would give it a 20% increase, out of 35%(assuming the 128 ROPs don't become the new bottleneck, which I don't think they will) that would still about 10% on the table, plus whatever latency increase turbo-clocked GDDR6 has, so I'd still go for 28G7 .
That's not how memory controllers work. You can't magically switch ram types (unless it's built into the controller, which in this case I don't think it is).

Also, the difference between 1080p/1440p is 35%, but again, it's about ~1440p60 or up-scaling 4k there-abouts in certain titles, or keeping in the VRR window for the later (at stock).
It wouldn't require as much as you may think. I'm not saying that they could absolutely accomplish it for Wukong at any decent settings, but many titles where that would become a '4k' card and replace 7900xtx.

For instance, think of something like this. Remember that the 9070 xt in this scenario (W1z uses the slowest card, in this case a Pulse) is running at ~2808/2505mhz.
If it scaled linearly, that would require ~3891mhz. If the RT bandwidth requirement for 2.97ghz is 20gbps (as is likely the case, as models at 2518mhz run proportionately faster), that would require 26.2gbps.
Which again, is the absolute speed of a typical 5080 core's capability (but would likely not be limited by RAM as seen here, as AMD [just like nVIDIA] will likely increase it [to 32GB, perhaps vs nVIDIA's 24GB]).

Make sense? Obviously they'd likely sell it as somewhere between 48-60, with the goal of absolute performance (overclocking/undervolting) being ~60. Just like it is for 1440p on 9070 XT in many games.

This is why this bios mod is kind of rad. Some people will get the "oh, now 9070 is not a ~40fps card but a ~48fps in-the-vrr window at high settings' card. Some won't. That's what makes these things neat.
The people that eventually buy a cheap 9070 and do that thing will feel special, like they cheated the system, and I think that's cool. It also helps bc it was super obvious what AMD was doing and this fixes it.
This theoretically takes 9070 from not-good-enough to actually good-enough in a lot of scenarios. Be that ~1440p60 in (quality up-scaled) AC RT, or maybe ~48 mins where at stock it might be ~40.
Go look at reviews for 1% lows and you'll see what I mean. It's A LOT of games; very similar to 5070. Putting that in the VRR window (even if you have to make a couple tweaks) is actually a really big deal.
Again, this is how I think the cards always should have been segmented; one as ~60 and the other ~48 mins in many games at high settings; at least through an OC (as in, IYKYK). That's the AMD secret sauce.
I get 9070 xt is not 60fps min stock in a lot of titles bc of the stock ram speed, but 9070 was relegated to lower settings/resolution imo for no reason other than to compete with 5070 and make XT look good.
Now it's tangibly better than a 5070 theoretically, which as it's AMD it needs to be IMO, while XT just gives that nice solid 60fps most of the time. Again, that's the way it should have been from the beginning imo.

1744037912503.png
 
Last edited:
So, the yields are indeed great for N48 and so, AMD cuts down healthy dies to offer a few 9070s in the market. And that's why the official price gap is so small. Many called that since prices were revealed. Great value GPU when bought close to MSRP, especially for the ones who love tinkering with hardware and software to get out of it more than they bought.
 
how do we unlock the extra shaders (CUs and Stream processors)?
 
Oh man this brings back some memories. I remember doing this with an AMD card a long time ago as well, but the actual model eludes me.
Probably ruins the per/W though.
Heck yea, the nostalgia is real! A few have mentioned going all the way back to the 9500. My first big boy GPU purchase was a 9700 Pro, and at the time I probably had no clue what a vbios even was. It wasn't until owning the 4850 that I became aware of bios flashing to the 70 series cards.
how do we unlock the extra shaders (CUs and Stream processors)?
Unfortunately, the probability is extremely high that you can't. Back in the day they would lock them in the vbios, but eventually, they started to physically sever them via lasers.
 
To make matters worse, no way you can get this GPU for MSRP. Not a 100% chance you can flash the BIOS and get lucky. We, buyers, want AMD to actually compete, not to squat around like an absolute monkey. Feels like they'll be reintroducing 7900 XT(X) at ever so slowly lowering launch prices without doing anything better than that.

I keep telling people; these prices won't last forever. It's been a month. I said give it until the end of June (Q2) or July (Prime Day), maybe slightly later.
You can't buy them (especially at MSRP) because they're sold out; they're making more. There's a $1500 5080 at my local BB, I can hook you up IYW. If you act fast, they also have a $1000 5070ti.

You're right, they are reintroducing 7900xt and 7900xtx...kind of...but shh. Those were never bad cards, nor were they generally bad prices (after they settled, which these haven't yet).

You know what the difference is, there, chief? RT. Up-scaling. The things people swore up and down were the most important things to ever exist (even though nVIDIA uses it to force a perpetual upgrade cycle).
By increasing the performance penalty of DLSS and it's featureset, upping RT ratio, and remixing specs so they fit their business goals, but generally not improving anything.

And worth 30% more money, on cards that now perform WORSE than their AMD counterparts in non-rt scenarios, sometimes RT; still 48fps 7900xt. In the VRR window. Almost like AMD's designs make sense.
I swear it was only yesterday people were telling me 4070ti was a better card than 7900xt.
Which NEVER made any sense, but apparently people believed that. Now they don't, because nVIDIA did the thing where they magically can't keep 60fps even with an OC (AGAIN), but still fell for it. 7900xt 60.
It's almost like you can predict 5070ti/5080 won't keep 1440p48 and 1440p60 (if perhaps up-scaled) when Rubin releases and AMD parts will, bc that's literally nVIDIA's business model some people can't comprehend.
And if these modded 9070s don't compete with 5070ti I will eat my hat, because they will. That's the point; just as XT has with 5080 for all practical intents and purposes. But, you know, don't tell anyone.

Guess what AMD is doing. Competing. Standardizing. What AMD does. Still. Again. Cheaper than nVIDIA, whom either stagnates their price/perf or sometimes lowers it, which boggles my mind.
While AMD always does the same thing; prices lower and allows an OC that will meet nVIDIA's next part up at the next realistic performance target, apparently this time on the down low IG.

You know what 5070ti/5080 are? Not quite 4080 or 4090, which is actually mind-numbing that people are buying at the same price. They are literally less silicon/ram, like AMD, running higher clocks, like AMD.
For more money...not like AMD. nVIDIA forgot the part where you pass the less-silicon higher-clocks (through maturity of the process) part onto the consumer. But muh 5% perf from GDDR7 now...JFC people.
nVIDIA's architecture is actually LESS efficient than RDNA4. Seriously. Compare normalized; as-in if they had a part with similar units/clocks; AMD would often be faster. This is how a 3200mhz 8192sp = 10752@2640mhz.
And that's with their unneeded GDDR7 vs very cheap GDDR6.

You sure I can't interest you in that $1500 almost-as-expensive-but-not-nearly-as-good-as-4090-from-over-two-years-ago? They say it's in high-demand and expected to sell out.
I have no clue why it's sitting there, people just keep walking by it and nobody is buying it...they must not see it. Also, BB needs to higher a better cleaning crew bc I can see the dust on it. Pardon nVIDIA's.

Pretty sure there isn't a 9070xt available within 1000 miles, even at it's 'outrageous' price of an OC edition being $730, HALF, nor the $650 9070 (which I actually do think is whack, but less-so with this mod).

FWIW: How long until people gather a list of which models this works on? I give it a day and you'll know exactly what model(s). If people do stupid things they shouldn't attempt them, but most will figure it out.
 
Last edited:
There's a $1500 5080 at my local BB, I can hook you up IYW. If you act fast, they also have a $1000 5070ti.
Thanks but no thanks. No way I'm buying a mid-range card for over a thousand even if it's available locally. And, y'know, logistics is a thing and this "$1000" 5070 Ti becomes a solid $1300 GPU that I have to wait weeks to arrive to my Russian town. That's too extreme. I'd agree to get a 4090 on these conditions but not a 5070 Ti.

I also don't need that much from a GPU since I play at 1080p. 4070 Ti is perfectly enough for my tasks. Will wait till it's priced sanely.
Cheaper than nVIDIA
Maybe but they ain't moving that needle fast enough. Like, a 4070 Ti Super can be obtained for less money and there's very few examples where 9070 XT meaningfully outperforms it. If more than zero of course.
That is despite 4070 Ti Super never becoming cheaper in the first place.
 
Last edited:
I have flashed Radeon 8500 to 9100 to get more functions like ATI TruForm and run Serious Sam, lucky days.
 
how do we unlock the extra shaders (CUs and Stream processors)?

Who said it's possible or not possible? Do not silicon these days get a laser cut? How do you want to unlock something which is cut off electrically?
I assume also the bios has some sort of mechanism to not be tempered with.

That hack just poorly works because it's from two ASUS cards with similar silicion.

Did people do not get main point: The graphic card is not stable in certain usual operation modes.
 
It most certainly does, X800 Pro (12 pixel pipes) flashed to X800 XT for full 16 pixel pipes and clocks to boot. I remember those days well!

HD6950 flashed to 6970 brings back fond memories also.
Similar I think I had the X850 pro vivo, but it had the faster ram that could be flashed to the X850 XT PE or something. I remember at the store the boxes had a clear window so I went through them all looking for the correct Samsung memory chips and scored 1 card out of the stack.........only to never flash it lol
oh there it is
 
As I have explained many, many, times...nVIDIA does not need all that bandwidth

I'm going to go out on a limb and bet on the extreme opposite: the 5090 is probably still bandwidth starved in some scenarios, even with the almost 2TB/s throughput available. Nvidia's no longer using GDDR6 on high performance GPUs, Ada used GDDR6X which did 24 Gbps easily on pretty much all SKUs - the 4080 had memory rated for that, and the 4080S actually shipped at 24 Gbps. They were still horribly bandwidth starved, despite it all.
 
Interesting - the extra shaders aren't needed - not in 3DMark tests anyway.

What is this - a bandwidth bottleneck?
 
Like the 5600xt, except that took a year or two before MSI leaked an unlocked bios, for micron memory version anyway.

Good to see value in the market, of course that card only cost 250 in the first place. Mid range has inflated a bit...
 
While not a GPU, i remember being able to enable the extra core on my Phenom X3 720. Always Good Fun!
Phenom X3 Heka unlocked to X4 Deneb. Then another X4 Deneb (OEM) unlocked to X6 Thuban. Two successful core unlocks on one motherboard, in the second case the core increase reached 50%! That was a happy AM3 platform :) (mid-range Gigabyte)
 
Phenom X3 Heka unlocked to X4 Deneb. Then another X4 Deneb (OEM) unlocked to X6 Thuban. Two successful core unlocks on one motherboard, in the second case the core increase reached 50%! That was a happy AM3 platform :) (mid-range Gigabyte)

Deneb didn't unlock, it was a complete quad-core processor. Zosma (960T) did, because Zosma was just Thuban with 2 cores turned off. The whole thing with the code names at the time was silly. AMD had basically 2 types of Phenom II dies, Deneb and Thuban, with 4 and 6 cores respectively. The rest were derivatives of these. Athlon II branded CPUs additionally had the L3 cache entirely disabled.
 
To this day, nothing beats the Athlon XP pencil mod to unlock multiplier in order to overclock. A pencil!
 
Here's nice video on how to and some results. About 25% performance increase by just flashing a XT BIOS and another 5% from undervolting in Cyberpunk although testing is very limited in the video. I kinda find hard to believe this much difference though.

 
Back
Top