Do AMD provide any benefit to the retail GPU segment.

[humbug]

I want to see AMD in a strong position, competitively, but i'm not buying their GPU's just because i prefer them as a company vs Nvidia, as i always have i will buy what ever is best for me, AMD are not a charity, neither am i.
AMD need to get to grips with their competitive problem without our sympathies.

 

[CAT-THE-FIFTH]

Are they? I'm glad of that, at the very least they should allow AIB's to double VRam if they so chose, my only concern is pricing, i'm ok with them charging a bit more for higher capacity VRam GPU's, but i don't want to se pricing go daft, there is no reason why they should cost a lot more.

To answer your question, yes.

The issue is if AMD pulls a dumb move like this,there is nothing stopping them from doing it a few months later. They did it with the GTX960 2GB,which had a 4GB version launched months later.

Someone at AMD has to have their head checked if they think they can sell a 6NM RX7600XT 16GB(with bog standard GDDR6) and try and think it can be positioned against an RTX4060TI.

The rumours indicate it will be around RTX3070TI 8GB level performance. Even if you clock the RX6650XT core to 3GHZ,and add RDNA3 IPC improvements into the mix,it might just about get to an RX6700XT or maybe an RX6750XT at 1080p. It will be slightly behind in rasterised performance and much slower in RT.At qHD and 4K,the lack of memory bandwidth and the 64MB less of Infinity Cache is going to cause problems.

An RTX4060TI 16GB would be faster in all metrics than an oveclocked RDNA3 RX6650XT 16GB.

Plus what level of nonsense improvement is this? Get an RX7700XT and you get the same performance as the RX6700XT,but some extra VRAM?

 

[humbug]

The issue is if AMD pulls a dumb move like this,there is nothing stopping them from doing it a few months later. They did it with the GTX960 2GB,which had a 4GB version launched months later.

Someone at AMD has to have their head checked if they think they can sell a 6NM RX7600XT 16GB(with bog standard GDDR6) and try and think it can be positioned against an RTX4060TI.

The rumours indicate it will be around RTX3070TI 8GB level performance. Even if you clock the RX6650XT core to 3GHZ,and add RDNA3 IPC improvements into the mix,it might just about get to an RX6700XT or maybe an RX6750XT at 1080p. It will be slightly behind in rasterised performance and much slower in RT.At qHD and 4K,the lack of memory bandwidth and the 64MB less of Infinity Cache is going to cause problems.

An RTX4060TI 16GB would be faster in all metrics than an oveclocked RDNA3 RX6650XT 16GB.

Yeah, AMD have a lot of Kudos points from Ryzen, there is a reason for that, being #### wasn't it, they have to make us like them and their products, if they get that right they will be rewarded.

 

[CAT-THE-FIFTH]

Yeah, AMD have a lot of Kudos points from Ryzen, there is a reason for that, being #### wasn't it, they have to make us like them and their products, if they get that right they will be rewarded.

AMD only got where it was with Ryzen,was by offering more for less. They seemed to haved dialed down on that in recent releases,ie,having useless stock CPU coolers,splitting B650 into B650/B650E,etc but still do offer better value in a number of segments. But even that only came after they managed to actually beat Intel,not only in performance but also efficiency.

Their dGPUs don't have the performance advantage,or even are more efficient. They need to price better than Nvidia to get sales for similar performance,or just outperform them more. This is what they had to do with Intel over three generations of Zen.

 

[humbug]

AMD only got where it was with Ryzen,was by offering more for less. They seemed to dial down on that in recent releases,ie,having useless stock CPU coolers,splitting B650 into B650/B650E,etc but still do offer better value in a number of segments. But even that only came after they managed to actually beat Intel,not only in performance but also efficiency.

Their dGPUs don't have the performance advantage,or even are more efficient. They need to price better than Nvidia to get sales for similar performance,or just outperform them more. This is what they had to do with Intel over three generations of Zen.

Again, yes agree with all of that, i get that on the Ryzen side AMD want to reap some of those rewards, i'm perfectly fine with that, but they have to find a balance in maintaining that appreciation for what they did by continuing somewhat in the same vain and bolstering their coffers, if they can find that balance its all good.

 

[CAT-THE-FIFTH]

Again, yes agree with all of that, i get that on the Ryzen side AMD want to reap some of those rewards, i'm perfectly fine with that, but they have to find a balance in maintaining that appreciation for what they did by continuing somewhat in the same vain and bolstering their coffers, if they can find that balance its all good.

I can understand Nvidia trying a fast one because they throw money at marketing and have the bigger market share. Not sure what AMD is attempting to do here? Sell more Nvidia cards?

 

[humbug]

I can understand Nvidia trying a fast one because they throw money at marketing and have the bigger market share. Not sure what AMD is attempting to do here? Sell more Nvidia cards?

I don't know, i do know a bunch of Intel's sacked lot ended up at AMD, hundreds if not thousands of them. Intel's marketing has been the worst for as long as i have been around.

 

[CAT-THE-FIFTH]

I don't know, i do know a bunch of Intel's sacked lot ended up at AMD, thousands of them. Intel's marketing has been the worst for as long as i have been around.

If they want to rebrand the RX6650XT RDNA3 clone as an RX7700XT,you know it will be around £400,or nearly £450 if the RX6700XT RRP of $480 is kept.

Unless this close to £300,what is the point?

 

[humbug]

If they want to rebrand the RX6650XT RDNA3 clone as an RX7700XT,you know it will be around £400,or nearly £450 if the RX6700XT RRP of $480 is kept.

Yeah and i will have no interest in it and probably something to say about that....

 

[CAT-THE-FIFTH]

Yeah and i will have no interest in it and probably something to say about that....

Its made on 6NM,so it would be as greedy an upsell as what Nvidia is doing this generation. So I hope they don't try this trick.

The moment Nvidia relents and allow higher VRAM models,that is end of the AMD dGPU division. They can't just add some more VRAM to a low end tier dGPU,and call it a day.

If Navi 33 actually had a 192 bit bus and stacked cache,I might look at it differently. The problem is the die is barely 200MM2 on a 6NM process node,so it looks like a straight shrink of Navi 23 unless all the published specs have been incorrect.

 

[KompuKare]

The hard part about using the Zen strategy for GPUs are:

1. Nvidia seldom mis-step (aside from pricing and segmentation)
2. GPUs are super parallel and performance & die size go hand in hand.

If - and this is a big if at this stage - AMD wanted to gain marketshare, they should use the savings from going chiplet to compete on price. But so far everything points to them not bothering.

Navi32 is almost too small for that strategy to work as the saving from chiplets vs monoliths mostly go to increased more packing costs. Plus chiplet will always use more power.

Navi31's ambitions were too modest. A 500mm² core plus the IO chips should have been able to take the crown. Low volume but halo is important, and AMD's current strategy of "let's price 5% less than Nvidia" hasn't worked and is actually is therefore very low volume anyhow.

 

[Zarax]

Its made on 6NM,so it would be as greedy an upsell as what Nvidia is doing this generation. So I hope they don't try this trick.

The moment Nvidia relents and allow higher VRAM models,that is end of the AMD dGPU division. They can't just add some more VRAM to a low end tier dGPU,and call it a day.

If Navi 33 actually had a 192 bit bus and stacked cache,I might look at it differently. The problem is the die is barely 200MM2 on a 6NM process node,so it looks like a straight shrink of Navi 23 unless all the published specs have been incorrect.

A die shrunk Navi 23 with minor improvements would make sense for the low end of the market, so 7600XT and below.
Everything else should be on proper MCM on at least 5nm, otherwise there would be very little improvement.

 

[humbug]

Its made on 6NM,so it would be as greedy an upsell as what Nvidia is doing this generation. So I hope they don't try this trick.

The moment Nvidia relents and allow higher VRAM models,that is end of the AMD dGPU division. They can't just add some more VRAM to a low end tier dGPU.

If Navi 33 actually had a 192 bit bus and stacked cache,I might look at it differently. The problem is the die is barely 200MM2 on a 6NM process node,so it looks like a straight shrink of Navi 23 unless all the published specs have been incorrect.

Whatever node its made on is not relevant to me, if its a good GPU its a good GPU, Zen 3 (Ryzen 5000) was made on exactly the same node as Zen 2 (Ryzen 3000) didn't matter at all it was a much better CPU, it was so good i bought one, paid over £400 for it, very unusual for me, but it was worth it, i still think that more than 2 years later and am looking forward to a couple more with it.

So if they want to save costs by making it on an older node, fine, do that, i'm good with it if its a good GPU for a reasonable price, that's all that matters.

 

[CAT-THE-FIFTH]

The hard part about using the Zen strategy for GPUs are:

1. Nvidia seldom mis-step (aside from pricing and segmentation)
2. GPUs are super parallel and performance & die size go hand in hand.

If - and this is a big if at this stage - wanted to gain marketshare, they should use the savings from going chiplet to compete on price. But so far everything points to them not bothering.

Navi32 is almost too small for that strategy to work as the saving from chiplets vs monoliths mostly go to increased more packing costs. Plus chiplet will always use more power.

Navi31's ambitions were too modest. A 500mm² core plus the IO chips should have been able to take the crown. Low volume but halo is important, and AMD's current strategy of "let's price 5% less than Nvidia" hasn't worked and is actually is therefore very low volume anyhow.

Yet,AMD seem to be fighting less aggressively against Nvidia,then against Intel. Who the heck thinks releasing the RX6700XT replacement,with a reheated RDNA3 RX6650XT with some more VRAM constitutes a strategy. It's an el-cheapo 6NM die.

All this will do is cost them,and their own board partners money. Better not to releasing junk like the RX7700XT if it indeed uses low end dGPUs like Navi 33.

A die shrunk Navi 23 with minor improvements would make sense for the low end of the market, so 7600XT and below.
Everything else should be on proper MCM on at least 5nm, otherwise there would be very little improvement.

The issue is that ROCm leak indicates the RX7700XT is using an RX6650XT clone.

Whatever node its made on is not relevant to me, if its a good GPU its a good GPU, Zen 3 (Ryzen 5000) was made on exactly the same node as Zen 2 (Ryzen 3000) didn't matter at all it was a much better CPU, it was so good i bought one, paid over £400 for it, very unusual for me, but it was worth it, i still think that more than 2 years later and an looking for ward to a couple more with it.

So if they want to save costs by making it on an older node, fine, do that, i'm good with it if its a good GPU for a reasonable price, that's all that matters.

The reality Navi 33 is a low tier dGPU compared to Navi 22. How is it going to overcome:
1.)20% less shaders.
2.)1/3 the Infinity cache amount. This is why the RX6700 series does much better at higher resolutions over the RX6600 series.
3.)2/3 the memory bandwidth.
4.)Half the PCI-E bandwidth.
5.)6NM is just a denser 7NM process node.
6.)It has a smaller die than Navi 23,indicating hardly a transistor jump.
7.)RT performance will probably not increase.

It isn't. It will be just clocked to the max,so will be way out of the peaked effiency range of the chip to catch an RX6700XT/RX6750XT. So you end up having to plonk a larger cooler on it. So an RX6700XT,with a bit more VRAM,and probably will hit more issues at qHD and 4K.I also have a PCI-E 3.0 system too,so AMD can keep low tier junk like that.

So if they go and do this moronic move,Nvidia will wait a few months. Maybe they will allow their AIB partners to make 16GB models,or launch an RTX4060 Super. Then the "RX7700XT" will look even worse. An AMD 700 series card loosing to the Nvidia 60 series cards.

It is better for AMD not to bother launching a Navi 33 based RX7700XT. It will be another Vega 64 level move. It will make Nvidia gain even more sales.

I would say £300ish for a junk GPU like that. You can get RX6700XT cards for £330~£350 now. Any higher and it deserves to rot on the shelves.

 

[humbug]

We don't that CAT, the 5800X was over 40% better at gaming than the 3800X, despite being on the same TSMC N7, it blew the 10900K out of the water....

 

[Zarax]

Can we really trust such a leak?
Unless AMD marketing is run by severely impaired chaps I severely doubt they will release such a stack with huge performance gulfs.

IMHO what they are going to aim in terms of performance is:
7800XT -> 6950XT
7700XT -> 6800XT
7600XT -> 6700XT

I kind of suspect the rumors about chiplets being underperforming due to hardware errors weren't far from reality and AMD used the time to retape N31 and fix N32, anything else they wouldn't bother VS using discounted RX 6000.

 

[CAT-THE-FIFTH]

We don't that CAT, the 5800X was over 40% better at gaming than the 3800X, despite being on the same TSMC N7, it blew the 10900K out of the water....

The RX7900 series is made on TSMC 5NM for the GCD. AMD had to increase memory bandwidth too on top of this,which is partly why they could reduce the Infinity Cache amount.

Navi 31 has 58 billion transistors and Navi 21 27 billion transistors. This is what contributed to the nearly 50% performance jump over Navi 21.

Navi 33 is made on TSMC 7NM,and instead of using the higher density to add more transistors,AMD just did a straight shrink,with a few extra RDNA3 changes. The transistor jump is tiny.
Memory bandwidth hasn't increased,even if the Infinity Cache is improved.The RX6700XT has 3X the amount of Infinity Cache.

All they are doing with this is hoping they can clock it high enough to match an RX6700 series dGPU,but with less memory bandwidth. So basically this will be OK at 1080p,but start to have issues at higher resolutions. Maybe putting 16GB on it will help,but not the severe reduction in memory bandwidth.

It also does not change the fact,they will be copying Nvidia. They are upselling a tiny 6NM die to £400+ now. So we get an RX6700XT 16GB for similar money after nearly 3 years.

At least Nvidia is using expensive TSMC 4NM and GDDR6X in many models. AD107 probably costs more to make than Navi 33.

Nvidia will the waltz in and soft reboot the range at the end of the year,and then the "RX7700XT" can get deep discounted.

Can we really trust such a leak?
Unless AMD marketing is run by severely impaired chaps I severely doubt they will release such a stack with huge performance gulfs.

IMHO what they are going to aim in terms of performance is:
7800XT -> 6950XT
7700XT -> 6800XT
7600XT -> 6700XT

I kind of suspect the rumors about chiplets being underperforming due to hardware errors weren't far from reality and AMD used the time to retape N31 and fix N32, anything else they wouldn't bother VS using discounted RX 6000.

It worries me as this is from ROCm,and usually these entries are there to prepare the software stack for later on. But AMD does boneheaded moves like this.

I seriously hope they unlaunch any Navi 33 based RX7700XT. It should be an RX7600XT 16GB at best.

 

[humbug]

I'll continue this later, i have a train to catch

 

[Zarax]

The RX7900 series is made on TSMC 5NM for the GCD. AMD had to increase memory bandwidth too on top of this,which is partly why they could reduce the Infinity Cache amount.

Navi 31 has 58 billion transistors and Navi 21 27 billion transistors. This is what contributed to the nearly 50% performance jump over Navi 21.

Navi 33 is made on TSMC 7NM,and instead of using the higher density to add more transistors,AMD just did a straight shrink,with a few extra RDNA3 changes. The transistor jump is tiny.
Memory bandwidth hasn't increased,even if the Infinity Cache is improved.The RX6700XT has 3X the amount of Infinity Cache.

All they are doing with this is hoping they can clock it high enough to match an RX6700 series dGPU,but with less memory bandwidth. So basically this will be OK at 1080p,but start to have issues at higher resolutions. Maybe putting 16GB on it will help,but not the severe reduction in memory bandwidth.

It also does not change the fact,they will be copying Nvidia. They are upselling a tiny 6NM die to £400+ now. So we get an RX6700XT 16GB for similar money after nearly 3 years.

At least Nvidia is using expensive TSMC 4NM and GDDR6X in many models. AD107 probably costs more to make than Navi 33.




It worries me as this is from ROCm,and usually these entries are there to prepare the software stack for later on. But AMD does boneheaded moves like this.

I seriously hope they unlaunch any Navi 33 based RX7700XT. It should be an RX7600XT 16GB at best.

Any chance we might be seeing a typo?

 

[CAT-THE-FIFTH]

Any chance we might be seeing a typo?

This is the same AMD which relaunched the RX690 as the RX5700XT,because Nvidia jacked up pricing.

I wish it is a typo,but AMD RTG has a long history of doing stuff like this and not reading the market well:
1.)Fury X at GTX980TI pricing(it was slower and had less VRAM)
2.)Vega64 pricing at GTX1080 level or worse(wasn't faster and drank power)
3.)RX5700XT rebrand(why I think some of the leaks ended up being wrong on pricing)
4.)RX6600/RX6600XT launch pricing meaning you barely got an improvement over the RX5600/RX5700 series
5.)The RX6500 series being worse than the RX5500 series
6.)RX7900XT pricing being rubbish,and making the RTX4070TI look better than it should

Then invariably AMD sales don't do so well,and they have to drop pricing. They did it with Zen4 too. Nvidia doesn't suffer as much from this as AMD does,and it seems repeatedly AMD dGPU launches leave room for easy Nvidia replies.

What Nvidia is doing by massively jacking up prices should make it far easier for AMD to beat Nvidia at most price levels,whilst maintaining decent pricing. AMD trying to be 5% slower for 10% less money,whilst being slower in RT,having less effective upscaling,etc isn't going to work.

I hope it is a typo,but I simply don't trust RTG hasn't entered another alternate reality zone which has happened before....with predictable results.

"BuT wE aRe NoT a BuDgEt BrAnD?",right?