Question AMD 2Q23 Earnings

[A///]

This time it's the dGPU market that they are putting a fork in. It's the surest sign it'll be fine in the long run.

RDNA3 is nothing to write home about. Neither is 90% of nvidia's junk lineup this generation. the 4090 is nice, but overpriced to feed jensen's big ego. amd's instinct cards are nice but you need to the software to back it up. nvidia has spent the better part of a decade worming its way into the field and has polished their stack over time. amd needs to dedicate dozens if not hundreds of engineers to their software stack on the pro side to get it going which in turn would move mountains for consumer products through trickle down. as it stands if you want a good amd dgpu you buy rdna 2 when rdna3 is vastly overpriced for what it is. you can argue new codecs for streaming all day but when the major networks don't support it there is no point. 2 years ago amd was not a thinking man's choice, it was the only choice. with rdna 2 being a winner in many tests against an overpriced and unavailable ampere, and intel chasing their tail down a mountainside. still are in between pat's calls to god.

 

[jpiniero]

AMD in the 10Q said that YoY the client shipments were down 42% and the ASP dropped 22%.

 

[Thunder 57]

still are in between pat's calls to god.

You mean his "review mirror"? That aged poorly.

 

[A///]

and last quarter they said they had to cut back orders. does amd differentiate client from dc? Both were affected last quarter and this compared to their highs not long ago. dc waning for either amd or team blue and hot would point towards dc and co not wanting to increase capex in needy times such as this or sufficinetly getting their own services saturated with new capable hardware.

ive been under the sun all day. time for a nearly frozen pinot gris spooned out of a glass and into my mouth.

 

[A///]

You mean his "review mirror"? That aged poorly.

rear view and no, the man is religious. before twitter cordoned off user tweets with a login redirect you could read his tweets where he cited bible verses. intel may have to part their backside like moses and the red sea if they want to endure any more punishment from the markets with lame product stacks.

 

[yuri69]

AMD keeping prices high to increase margins is showing losing potentially more marketshare than it is worth.

If we consider AMD console revenue grew while their overall revenue shrunk for gaming, AMD discrete graphic revenue may be less than 100 million dollars this quarter. That is not enough to sustain R and D and AMD needs volume. AMD has to accept people will never consider their brand as equal to Nvidia without actions to justify this. RDNA3 did not help AMD efforts(fell below expectations) and their lack of software engineers in the RTG group show they are the budget brand. Hate marketing will only get them so far and has a side effect of poisoning the market for everyone. Only AMD fans will overlook AMD software short coming and pay Nvidia prices for AMD hardware.

Neglected driver releases for RDNA2 post RDNA3 launch, late phoenix graphic drivers, failure on time some of their promised software features show that they are not spending enough on software engineers and the lack of sales is going to worsen this. I think this poor execution on the software side is the result of engineers being moved to datacenter side. But this gamble might not pay off.

I think MI300 is coming a bit too late. Q3 outlook doesn't look good and if Q4 is where we start seeing some Mi300 revenue as a result of El Capitan, I don't think the supply of Mi300 will be that strong this year. And if q1 of 2024 is when supply of MI300 becomes sufficient, AMD will soon have to contend with Blackwell. Nvidia has a tendency to make announcements near the end of Q1 for their next gen compute products. AMD need to release their data center products early since they have a software disadvantage, releasing it late mean companies are losing valuable time developing software to work on MI300.

This.

GCN/CDNA solutions have been lagging behind nV and the current time-to-market doesn't look like AMD is catching up any time soon. Not even mentioning the SW ecosystem side.

Discrete GFX business has been declining like past Polaris. The revenue keeps shrinking. Discrete GFX cards should be the primary cost cut target. It's a miracle we are going to see 3 chips based on RDNA3 considering the revenue.

AMD needs the GFX IP for integrated/consoles, but beyond that? It sounds wasteful to design 3 chips for a 5-10% market share.

Time will tell...

 

[IEC]

Crypto killed the GPU market for gamers. Thankfully, crypto mining is mostly dead for GPUs.

I'd imagine with the market proving they only want AMD GPUs to lower prices on nVidia GPUs (see poor sales data above), AMD will instead focus on datacenter/Instinct lineup instead of high end Radeon.

AI and LLM and the exponential increase in datacenter GPU demand will keep consumer GPU space uninteresting for a while.

 

[adroc_thurston]

It's a miracle we are going to see 3 chips based on RDNA3 considering the revenue

It's a truly wonderful thing given RDNA4 had 5 parts.

GCN/CDNA solutions have been lagging behind nV

That's not what NV thinks but I digress.

 

[adroc_thurston]

I'd imagine with the market proving they only want AMD GPUs to lower prices on nVidia GPUs (see poor sales data above), AMD will instead focus on datacenter/Instinct lineup instead of high end Radeon

It's a truly wonderful thing that they're focusing on both.

 

[soresu]

It's a truly wonderful thing given RDNA4 had 5 parts.

#5 was a Zen5/RDNA4 APU?

 

[A///]

AI and LLM and the exponential increase in datacenter GPU demand will keep consumer GPU space uninteresting for a while.

maybe it'll subsidize normal gpus for the typical consumer or it won't and we're all kidding ourselves heres.

 

[adroc_thurston]

#5 was a Zen5/RDNA4 APU?

No, 5 dGPUs.

 

[soresu]

No, 5 dGPUs.

Do you mean 5 SKUs of 2-3 dies or 5 different dies?

 

[adroc_thurston]

or 5 different dies?

This.

 

[jpiniero]

What's funny is that AMD probably made close to nothing in the quarter from dGPUs. Sorny was probably a billion by itself and the remaining 500ish million could easily be between Microsoft and the handhelds. The ROG Ally allegedly sold about a half million.

 

[adroc_thurston]

and the handhelds

That's under Client.

 

[jpiniero]

That's under Client.

I originally thought so too but the earnings documents made it sound like that is not the case.

 

[adroc_thurston]

I originally thought so too but the earnings documents made it sound like that is not the case.

Deck should be under Gaming, Z1 something under Client.

 

[A///]

No, 5 dGPUs.

had or has? I don't believe amd has the funds for the r n d for 5 different gpu dies. unless they got a major advance from clients so desperate for a dgpu for ai and nvidia couldn't give them an allotment. why pick the platform that has questionable support for machine learning and ai when team green mean machine has been doing this for over a decade?

 

[adroc_thurston]

had or has?

Had, parts were killed for TTM reasons.

I don't believe amd has the funds for the r n d for 5 different gpu dies.

ughhhh yes they do.

 

[yuri69]

Why on earth would they design 5 different dGPU dies in a single gen?

The last gen having 5 different dGPU dies was the 300-series in 2015. That one included some rebrands and "firmware refreshes".

Anyways, AMD's ultimate goal seems to be to disintegrate the GPU die to make it modular in order to decrease the number of different dies manufactured on the hi-perf process. The dual-die MI 250X is sold as a single-die SKU too. The 7900 GRE requires just 4 out of 6 IMC chips. The goal seems to be clear.

The Radeon brand has to be restarted completely. The current strategy of pricing-a-tad-lower-than-nV hasn't been working for ages. Why would anybody prefer the Radeon GPU brand which is a tiny bit cheaper *BUT* notoriously known for lacking SW support, missing features, nowadays bad perf/W, and got a far inferior branding?

Radeon has to admit they are #2 by lowering the price and profiling themselves as such.

 

[Thunder 57]

Why on earth would they design 5 different dGPU dies in a single gen?

The last gen having 5 different dGPU dies was the 300-series in 2015. That one included some rebrands and "firmware refreshes".

Anyways, AMD's ultimate goal seems to be to disintegrate the GPU die to make it modular in order to decrease the number of different dies manufactured on the hi-perf process. The dual-die MI 250X is sold as a single-die SKU too. The 7900 GRE requires just 4 out of 6 IMC chips. The goal seems to be clear.

The Radeon brand has to be restarted completely. The current strategy of pricing-a-tad-lower-than-nV hasn't been working for ages. Why would anybody prefer the Radeon GPU brand which is a tiny bit cheaper *BUT* notoriously known for lacking SW support, missing features, nowadays bad perf/W, and got a far inferior branding?

Radeon has to admit they are #2 by lowering the price and profiling themselves as such.

Ugh. Lowering the price will not work. Software support is almost always excellent. Missing features, what DLSS? RDNA2 had better perf/W than Nvidia at the time.

 

[yuri69]

Ugh. Lowering the price will not work. Software support is almost always excellent. Missing features, what DLSS? RDNA2 had better perf/W than Nvidia at the time.

Missing a true DLSS alt, various subpar implementations of FSR, unusable VR months after the release on RDNA3, long standing multi-monitor/video power consumption problems, often broken OpenCL driver releases, recent broken HW browser video acceleration, no OptiX acceleration, broken DaVinci resolve, ROCm is no alternative to the CUDA stack, long standing OpenGL Windows issues, etc.

RDNA2 was an exception. Although, nVidia used a subpar Samsung process back then.

 

[Ajay]

Had, parts were killed for TTM reasons.

Any clues? Not enough talent, overly ambitious, bad management?

ughhhh yes they do.

People are stuck in the past. AMD is a very different company compared to ten and even 5 years ago.

 

[Ajay]

RDNA2 was an exception. Although, nVidia used a subpar Samsung process back then.

NV threw more xtors at the problem, and more power, oh and better marketing. It worked, particularly helpful was the crypto boom.