AMD 7800 / 7700 reviews

Page 5 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Tuna-Fish

Golden Member
Mar 4, 2011
1,346
1,525
136
My point earlier in the thread was about the silicon both Nvidia and AMD were using. I was pointing out the power consumption of Nvidia cards vs. AMD cards. I said that Nvidia was using the best silicon short of 3nm. AMD is a node behind Nvidia cards. Which shows why Nvidia cards are more efficient.

And as multiple people have told you in this thread, you are wrong, but you refuse to engage with their replies.

Your idea that NV has a node advantage here is simply misinformed. The GCDs on N31 and N32 use basically the same process as NV does. AMD is clearly doing much worse, but it is not because of the process.
 
  • Like
Reactions: Tlh97 and KompuKare

Hans Gruber

Platinum Member
Dec 23, 2006
2,131
1,088
136
It's been a disapointing three years or so, right when it was promising to really get good again as GPUs moved to modern node processes. It's like the stagnation of 5-7 years ago but they are doing it by choice rather than necessity... and the supply chain imploded AND then proved to both vendors how much we might actually be willing to pay. :/
Add to it the nonexistence of all those crypto 30 series Nvidia and AMD RDNA2 cards that were lightly or never used before the crypto market collapsed. It's like they are being told to hide those cards away or destroy them. You would think investors would like to get something for their GPU's other than a total loss.
 

biostud

Lifer
Feb 27, 2003
18,238
4,755
136
I didn't misread your comment. I went off on a bit of a tangent and I didn't want to totally hijack the thread. I am just a bit frustrated with the GPU market.

I play World of Tanks 95% of the time and get ~100-130fps @5120x1440 so, everything is fine :p (but otherwise totally agree)
 

Hans Gruber

Platinum Member
Dec 23, 2006
2,131
1,088
136
And as multiple people have told you in this thread, you are wrong, but you refuse to engage with their replies.

Your idea that NV has a node advantage here is simply misinformed. The GCDs on N31 and N32 use basically the same process as NV does. AMD is clearly doing much worse, but it is not because of the process.
I am not wrong. Nvidia is on the most enhanced/advanced 5nm node. AMD is on 5nm standard silicon because it costs less. AMD went with the less expensive 5nm version and mixed in 7nm (chipset/controller) which they call 6n with 5nm silicon. The reason the power efficiency is better on the 40 series Nvidia cards. The version of the 5nm silicon is newer and has more advanced power efficiency. The 30 series Nvidia cards high end 3080/3080ti and 3090/3090ti were built on Samsung silicon. Nvidia wanted the best TSMC silicon for the 40 series.

TSMC first started selling 5nm silicon? It was March 2019. It is now September 2023. During that time they enhanced, perfected, improved efficiency on the 5nm silicon several times. They have more advanced variants of the 5nm silicon. TSMC has all the benefits of each variant on the 5nm in their literature. Nvidia has the best variant which is more expensive than what AMD uses. Apple was on 5nm while AMD was on the older 7nm with Zen 2 and Zen 3. TSMC 7nm silicon was on the market mid 2017. Zen 2 was released in August 2019. Dr. Su was asked why AMD was using 7nm instead of 5nm. Being on the cutting edge technology is more expensive she said.

FYI, Intel ARC cards were supposed to get an alchemist+ refresh in 2023 that would have moved their process from N6 to N4 like what Nvidia currently uses. This would be about a year after the Nvidia 40 series was released. This would mean better performance for the ARC cards and lower power consumption. The cost is lower as time goes by because it's not longer cutting edge but still advanced.
 

PJVol

Senior member
May 25, 2020
533
446
106
Aside from stupid naming decision, I was curious about how RX 7800 looks compared to RX 6800.
Given both have the same number of functional units, except half as much of LLC, the rest looks like this:

7800 vs 6800
----------------------------------------
GFX clocks +17%
Vram clocks +22%
TDP +5%
xtors +5%
=== performance / power in games ====
+22% / +6%

So I don't even know, if this looks bad or really bad for the brand new toy? )
 
Last edited:

Elfear

Diamond Member
May 30, 2004
7,097
644
126
I'm really tempted by the Hellhound 7800XT. $500 and comes with Starfield which I've been wanting. Only a 50ish% performance increase over my 6700XT though so not sure it's worth it.
 

PJVol

Senior member
May 25, 2020
533
446
106
I'm really tempted by the Hellhound 7800XT
Take a look at the HWUB review, the last section of it. I may be wrong but it feels like something is shady with its clocks.
PS: start watching from 20:50
 
Last edited:

Tuna-Fish

Golden Member
Mar 4, 2011
1,346
1,525
136
I am not wrong. Nvidia is on the most enhanced/advanced 5nm node. AMD is on 5nm standard silicon because it costs less.
This is not true. They both customize their processes, all the big clients do. The only difference is that NV named theirs, while AMD uses the standard TSMC naming. Note that N4 is not the same as 4N. We wouldn't know about cost differences, but under electron microscope there are no meaningful differences between the processes.

7nm (chipset/controller) which they call 6n

No, they are using TSMC N6. Which is a N7-class process, but is a separate thing, and not named by AMD, but TSMC. (it's mainly a cost upgrade to N7, with minor density and power improvements.)
 

Hans Gruber

Platinum Member
Dec 23, 2006
2,131
1,088
136
This is not true. They both customize their processes, all the big clients do. The only difference is that NV named theirs, while AMD uses the standard TSMC naming. Note that N4 is not the same as 4N. We wouldn't know about cost differences, but under electron microscope there are no meaningful differences between the processes.



No, they are using TSMC N6. Which is a N7-class process, but is a separate thing, and not named by AMD, but TSMC. (it's mainly a cost upgrade to N7, with minor density and power improvements.)
This is a direct quote from TSMC.

Compared to N5, N4P will also deliver a 22% improvement in power efficiency as well as a 6% improvement in transistor density. In addition, N4P lowers process complexity and improves wafer cycle time by reducing the number of masks.

When you look at the power draw for the 4070 vs the 7800XT. The Nvidia silicon is at least 22% more power efficient than the N5 that AMD uses. How Nvidia manages that efficiency gain shows in the reviews of power draw. The quote above is from October 2021. Both processes are made from 5nm TSMC silicon. It gives an idea of the benefits of what N4 can do vs. standard N5.

I will add a wikipedia link showing all the enhanced variants of 5nm TSMC that you insist do not exist. Or that all TSMC 5nm silicon is the same throughout the product cycle.
 
Last edited:

gdansk

Platinum Member
Feb 8, 2011
2,078
2,559
136
This is a direct quote from TSMC.

Compared to N5, N4P will also deliver a 22% improvement in power efficiency as well as a 6% improvement in transistor density. In addition, N4P lowers process complexity and improves wafer cycle time by reducing the number of masks.

When you look at the power draw for the 4070 vs the 7800XT. The Nvidia silicon is at least 22% more power efficient than the N5 that AMD uses. How Nvidia manages that efficiency gain shows in the reviews of power draw. The quote above is from October 2021. Both processes are made from 5nm TSMC silicon. It gives an idea of the benefits of what N4 can do vs. standard N5.

I will add a wikipedia link showing all the enhanced variants of 5nm TSMC that you insist do not exist. Or that all TSMC 5nm silicon is the same throughout the product cycle.
It isn't the first time nor the last that AMD will be less efficient on the same process. But now with N32 and N31 AMD has the MCD approach wasting power too.

4N isn't N4 nor is it N4P. The difference between what AMD calls N5 and Nvidia calls 4N is not known. But look: the 7900 XTX vs RTX 4080 are closer in efficiency than the 7700 XT vs 7800 XTX. How could process ever be the key factor when that's the case?

Edit: sorry trying not to be argumentative
 
Last edited:

Hans Gruber

Platinum Member
Dec 23, 2006
2,131
1,088
136
It isn't entirely process-related. 4N isn't N4 nor is it N4P. You started saying RDNA is an improved 6nm but you shift the goalposts to somehow not be wrong... but you still manage to be wrong.

Moreover Nvidia clearly says Hopper is 4nm but never does so for Ada. Guess why? Because it isn't. Stop perpetuating marketing wank.
So people should just ignore the efficiency improvement on the 5nm TSMC node? For those wondering why the 4070 (4N) uses less power than the 7800XT(N5). Nvidia said when they switched from Samsung to TSMC. They wanted the very best TSMC silicon for the 40 series cards.

Nvidia quote
RTX 40 Series GPUs are also built at the absolute cutting edge, with a custom TSMC 4N process. The process and Ada architecture are ultra-efficient.Apr 21, 2023

I do not have the efficiency numbers for 4N. I gave an example of N4P above. 4N is newer and even more efficient than N4P. It looks like there are at least 6 different versions of TSMC 5nm silicon. They release new versions of the silicon with improved performance/efficiency throughout the lifecycle of the process node.

I said I needed clarification if the entire 7800XT product was made with N5 which is standard 5nm TSMC or if the new cards were 6N which is advanced 7nm silicon based. There was a question whether the 7800XT and 7700XT were on the same N5 silicon that the 7900xtx and 7900xt are on.
 

gdansk

Platinum Member
Feb 8, 2011
2,078
2,559
136
N32 GcD is 60 CU at 200mm² and N31 GCD is 96 CU at 305mm². Given that includes the media engine and other fixed size pieces that's almost perfect scaling. And they both achieve similar clock rates.
Why would it be a question if they are the same process?

If 4N is even better than N4P then why is the "4N" 4070 only 112% efficiency of the N5/N6 7800 XT? Shouldn't it be more like 125% improvement from N5 to "4N"? And even more because 1/3rd of N32 is made on N6 and has to bang bits across longer wires?

It just doesn't make sense if you think about it Hans. If there was a big difference between the two processes then AMD would be much further behind in efficiency, like the 7700 XT is, with all their products.

"4N" is mostly marketing wank. But what modifications Nvidia did obtain are less about getting performance and more about being able to make reticle limit dies with as few defects as possible. That's Nvidia's bread and butter unlike most TSMC customers who stick to reasonable or small size dies.
 
Last edited:
Jul 27, 2020
16,161
10,238
106
Interesting. I'm not up on what games will use AI in the future, so I'm not sure how excited to be. 😄
I'm giving away a million dollar gaming idea here: A game where you see an AI generated image and you have to generate something similar with a suitable prompt. That game's gonna get pretty tiring waiting on the GPU to generate the image :p
 
  • Haha
Reactions: Elfear

GodisanAtheist

Diamond Member
Nov 16, 2006
6,783
7,117
136
7800XT and 7700XT have the benefit of launching into the middle of the cycle, people had the performance numbers reasonably dialed in early on. AMD looks like they've pushed the arch well out of its comfort zone to hit performance targets but that's kinda what happens with these late launch mid-range parts (6700XT was similar).

Ultimately, all I want from these cards is to put more downward pressure on used 6700XT pricing. The 7700XT seems deliberately priced to avoid that scenario, letting 6750's happily occupy the $300-350 price point uncontested.
 
  • Like
Reactions: igor_kavinski

Mopetar

Diamond Member
Jan 31, 2011
7,831
5,980
136
Sapphire 7800XT Pulse sold out! What happened to launch day availability? /s

I don't know what the supply is like, but with the free game you're effectively paying $450 (plus what we the AIB manufacturer is charging over MSRP) for a new card that has pretty solid performance.

Unless you care about RT (or DLSS instead of FSR) or think it's worth $100-$150 then the 7800XT had great value compared to any other new cards in the market. It's 30+ % above the 4060 Ti that costs the same $500 based on TPUs numbers.

Even if you don't value the game code at all or don't think you could sell it, TPU still has the 7800XT as better performance per dollar than any other current generation card at MSRP, with the exception of the 4060 and 7600 which are in a completely different performance tier.

It may not be the most impressive card and a discounted 6800XT isn't meaningfully worse, but the value of almost everything else in the current generation of cards has been so bad that the 7800XT manages to look good by comparison. It's not surprising that people are buying. The only question is how much supply there was for them to buy.
 

beginner99

Diamond Member
Jun 2, 2009
5,210
1,580
136
AMD went with the less expensive 5nm version and mixed in 7nm (chipset/controller) which they call 6n with 5nm silicon. The reason the power efficiency is better on the 40 series Nvidia cards.
The reason is architecture. And the only reason ampere was so bad was because it was on the every bad Samsung "8nm" process which was probably closer to a 16 nm than 7nm that RDNA2 used. RDNA2 just looked so good because NV used the worst possible "advanced" process available.

The part from NV marketing about most advanced process is just that, marketing.
 

RnR_au

Golden Member
Jun 6, 2021
1,705
4,153
106
I said I needed clarification if the entire 7800XT product was made with N5 which is standard 5nm TSMC or if the new cards were 6N which is advanced 7nm silicon based. There was a question whether the 7800XT and 7700XT were on the same N5 silicon that the 7900xtx and 7900xt are on.
AMD is not using standard TSMC N5 for any of their products afaik. They are on record on saying that they will be using an optimised version.

Dr Su reinforced that technology roadmaps are all about making the right choices and the right junctures, and explicitly stated that our 5nm technology is highly optimized for high-performance computing – it’s not necessarily the same as some other 5nm technologies out there.
Source

Even if you think that the statement above is about Zen 4, it would be brain dead to think that AMD could be using an optimised process for their cpu's but not for their gpu's.
 

KompuKare

Golden Member
Jul 28, 2009
1,013
924
136
For a chiplet design both Navi 32's and Navi 31's perf/watt is actually pretty good.
Okay, nobody else has any chiplet GPUs which makes comparison hard but there was always going be a power penalty from going chiplet.

IMO the mistake AMD made was that if they were only making three RDNA3 designs, they should have been low end 6nm monolith, medium 5nm monolith, high 5nm chiplet. And at each tier, IMO they should have aimed a bit higher. Certainly with Navi 33, whereas Navi 32 would have been fine as a Navi 22 successor but they are trying it on as a Navi 21 successor.

Ah well, armchair strategists is fun, but at the end of the day if they can't make anything at a piece we consumers want to buy at, then they won't get any sales. High margins at low volumes can't be very profitable but what do we know?!

And coming 6 to 9 months late and undercutting Nvidia by $50 is not going to work. Well, in sure I will eventually want to play Starfield but being a Bethesda game, I'll wait until the modding community had fixed it.
 
  • Like
Reactions: Tlh97

R81Z3N1

Member
Jul 15, 2017
77
24
81
I think the Starfield promo was just a little too juicy, I am actually surprised by the sell out of the card. We will just have to wait see if the other shoe drops, if we see a major increase in Ebay sales due to folks wanting this card, then folks haven't learned much from the past.

Also going forward, know that a few reports in EU have shown high sales as well, which is a good sign for AMD. Might be a little pent up demand at play as well. Know that some YT folks joke about this card having staying power, like this is your next 5 year card.