Radeon RX 6700, 6700 XT & 6750 XT users club

[RJARRRPCGP]

Nah, 7700 XT's performance competition is the non-XT version of 6800. 6750 XT is noticeably slower.

RX 6750 XT seems to have lower IPC, which explains the high core boost frequencies.

RX 6800 family can't clock as high, but has more performance-per-core-frequency. More FPS at only 2.5-2.6 GHz than RX 6750 XT @ 2.7 GHz

 

[AusWolf]

RX 6800 family can't clock as high, but has more performance-per-core-frequency.

Having more cores might have something to do with it.

 

[Lew Zealand]

RX 6750 XT seems to have lower IPC, which explains the high core boost frequencies.

RX 6800 family can't clock as high, but has more performance-per-core-frequency. More FPS at only 2.5-2.6 GHz than RX 6750 XT @ 2.7 GHz

The 6750 XT's "IPC" is identical to the rest of the series, instead it has a different problem:

Not enough C.

Look at the ratio of cores to memory bandwidth throughout the whole 6000 series, even just using memory bus width as a crude guide and you'll notice who got screwed: the 6700 XT and 6750 XT. They got 2560 cores but should have 3072 with that 192-bit bus, maybe 2880 at the very least. So AMD boosted the core frequency and by necessity also power consumption.

At least they fixed core count in the 7700 XT but then re-broke it with too big of a price hike. Well, I guess that makes the 7800 XT look good by comparison.

 

[AusWolf]

At least they fixed core count in the 7700 XT but then re-broke it with too big of a price hike. Well, I guess that makes the 7800 XT look good by comparison.

Unlike Nvidia, AMD's intention seems to be to sell cards based on fully enabled dies. Maybe their yields are so much better (which I find odd), or maybe it's just a different business strategy where Ti and Super (or xx50) refresh series aren't planned ahead.

 

[Macro Device]

That's not a problem performance-wise

I'd not be so sure. Chiplet design effectively increases latencies which damaged Ryzen 1 and 2 success a solid bit. Lessons are learnt, sure, but this is an uncharted territory in the world of GPUs. I suppose RDNA3 would be about 5 to 20 percent faster if it was for monolythic design. And about 25 to 50 percent more expensive for AMD as well.

 

[AusWolf]

I'd not be so sure. Chiplet design effectively increases latencies which damaged Ryzen 1 and 2 success a solid bit. Lessons are learnt, sure, but this is an uncharted territory in the world of GPUs. I suppose RDNA3 would be about 5 to 20 percent faster if it was for monolythic design. And about 25 to 50 percent more expensive for AMD as well.

Maybe... All I know is that my 7800 XT is about 1.5x faster than the 6750 XT I had before while having a similar chip power draw (212 W sustained vs. 200 W). How much of it is due to 5 nm, I don't know.

 

[Macro Device]

Maybe... All I know is that my 7800 XT is about 1.5x faster than the 6750 XT I had before while having a similar chip power draw (212 W sustained vs. 200 W). How much of it is due to 5 nm, I don't know.

Still don't get it why you upgraded to 7800 XT C:

Getting less than 200% uplift is, "why did I even bother?"

 

[AusWolf]

Still don't get it why you upgraded to 7800 XT C:

Getting less than 200% uplift is, "why did I even bother?"

Because I was curious of how RDNA 3 works. Most of my tech purchases aren't based on need, but on curiosity. It's a hobby on its own.

 

[Macro Device]

Because I was curious of how RDNA 3 works. Most of my tech purchases aren't based on need, but on curiosity. It's a hobby on its own.

Ah, like how I hammerdrilled a hole in my good ol' Celeron's IHS and was comparing before/after temperatures...

 

[AusWolf]

Ah, like how I hammerdrilled a hole in my good ol' Celeron's IHS and was comparing before/after temperatures...

Yep, kind of.

 

[Ruru]

I used the same RAM and video for the first Athlon build, a T-bird 900 MHz.

I have a T-bird 1000C with the legendary AXIA core, OCs over 1400MHz easily.

 

[RJARRRPCGP]

I have a T-bird 1000C with the legendary AXIA core, OCs over 1400MHz easily.

I suspect the 1000Cs are rare. I thought those were usually the 1400s or at least the 1200s.

I guess I lost the silicon lottery. "IRQL_NOT_LESS_OR_EQUAL" BSOD at 1050 MHz on my 900B. (with CPU-multi-only-OC)

I had a 1300B for only a short time in 2002, (July 5th to October 13th) because I did something real dumb. I accidentally cracked the core!

 

[Noci]

Because I was curious of how RDNA 3 works. Most of my tech purchases aren't based on need, but on curiosity. It's a hobby on its own.

Auswolf, be aware that this it is a dangerous hobby indeed, because curiosity killed the cat, uhh Wolf, let's just hope you're not very feline but more K9 \

And for BMD, look at the bright side, every improvement, no matter how small, is an improvement . If one did his/her research on beforehand, it will not disappoint just realize that everybody has a his/her perception, motivation and/or reason for the improvement.

Downside is we have to kick Auswolf out of this thread as he is now a RX 7000 series owner

 

[AusWolf]

Auswolf, be aware that this it is a dangerous hobby indeed, because curiosity killed the cat, uhh Wolf, let's just hope you're not very feline but more K9 \

What can go wrong? I have a (slightly) better GPU with a bit worse idle power consumption. No biggie.

Downside is we have to kick Auswolf out of this thread as he is now a RX 7000 series owner

Khm... I still have the 6750 XT as a spare.

 

[Ruru]

What can go wrong? I have a (slightly) better GPU with a bit worse idle power consumption. No biggie.


Khm... I still have the 6750 XT as a spare.

I will have a similar upgrade, I'll save for 7800 XT after new year and put the 6700 XT to my 2nd setup. The R5 3600 is already missing it after spending 10 months with it.

 

[Jism]

Be careful with that, it's discontinued for a reason. The one tube of MX-5 I had separated into a gummy and a watery phase even before I opened it. I ended up throwing it away and ordering some MX-4 instead.

That fluid you mention is just the oil thats supposed to mix up with the actual paste. It's really no issue. You can mix up the oil with the paste again. It is discontinued because some people don't know that over time the oil will seperate from the paste just like the oil in nutella pasta will be. You mix it back up and it's fine.

Wondering how effective this might be for an ASIC B card like the PowerColor Fighter?

What do you mean? If it's a reference design, then yes. Some cards have IC's to prevent more current being pulled. That's not out of protection but out of costs. They don't want you to OC or push beyond it's rated specs.

I have a T-bird 1000C with the legendary AXIA core, OCs over 1400MHz easily.

I remember those. Lovely chips. Just extremely fragile. One botched cooler installation or removal was enough to start cracking dies at the edges. Coolers back then vs now where not so good while those CPU's only consumed 50W to 65W on avg.

 

[AusWolf]

That fluid you mention is just the oil thats supposed to mix up with the actual paste. It's really no issue. You can mix up the oil with the paste again. It is discontinued because some people don't know that over time the oil will seperate from the paste just like the oil in nutella pasta will be. You mix it back up and it's fine.

Yet, I never had this issue with the MX-4, or any other paste in fact.

 

[Jism]

It's bin on my CPU for over a year, it's bin on my GPU, no issues really.

the hardened part is just without oil. Remix that up, repaste your chip and it's good.

 

[AusWolf]

the hardened part is just without oil. Remix that up, repaste your chip and it's good.

So one should squeeze the whole tube out into a dish and mix it up before using? Why was there no user's manual to indicate that? And why was it sold in a syringe in the first place?

Anyway, I'm glad it's been discontinued. It should have never been sold in such a state to begin with.

 

[Jism]

No,

You squeeze out the fluid with the harder stuff. You mix that up till a good paste. That's what you spread over your GPU or CPU. It's what i did and with zero issues. While using the stock cooler at approx 3500 RPM, i managed to obtain 77 degree core and 98 degree hotspot while consuming 260W of power. That's quite impressive - i only repasted it because of the temperature difference hotspot / GPU with stock paste.

 

[AusWolf]

No,

You squeeze out the fluid with the harder stuff.

How, when half of your syringe is the oily liquid stuff and the hard part is stuck on the bottom? You can sugar-coat it any way you want to, but it's still a problem, and it's still the main reason why the MX-5 was discontinued so soon. Thermal paste should not separate straight out of the factory, just like no other paste does. Ever. The fact that you took the effort to mix it up into a usable consistency doesn't mean it's normal.

 

[Ruru]

How, when half of your syringe is the oily liquid stuff and the hard part is stuck on the bottom? You can sugar-coat it any way you want to, but it's still a problem, and it's still the main reason why the MX-5 was discontinued so soon. Thermal paste should not separate straight out of the factory, just like no other paste does. Ever. The fact that you took the effort to mix it up into a usable consistency doesn't mean it's normal.

I read that MX-6 is practically what MX-5 was supposed to be, am I right?

 

[AusWolf]

I read that MX-6 is practically what MX-5 was supposed to be, am I right?

I wouldn't be surprised considering that the MX-5 was only sold for literally a couple of months (which is way shorter than the average lifespan of thermal paste products), and the MX-6 came straight after.

 

[Ruru]

I wouldn't be surprised considering that the MX-5 was only sold for literally a couple of months (which is way shorter than the average lifespan of thermal paste products), and the MX-6 came straight after.

Yeah. I had one syringe of MX-5 and it did its job though.

 

[Ijachok]

Hi! I've started trying to undervolt my XFX RX 6700. And I see on Youtube, that people increase the Power Limit. For undervolting. But I need to reduce consumption. And it seems to me logical, to limit it. What is the best thing to do with Power Limit?

I haven't touched the Power Limit yet. I gradually reduced the value in the Voltage field from 1200 mV to 1080 mV. The consumption is almost the same as without my intervention. It was 140, now it's 138 under load in Furmark and Metro Exodus. I reduced the Voltage to 1050 mV. The situation did not change. I see, that other people have decreased the chip's voltage to somewhere around these values.
And I'm interested in the value that I highlighted on the screen.

Do I need to get lower values than this one, highlighted in blue for noticeable results?

And a more general question, do I do any mistakes?