Two CEOs together. ARM and NVDA.
This expanded my view on the impact of the acquisition.
possiblity of ARM acquistion by nvidia to be approved by china will be close to zero(ARM IP falling under US jurisdiction will be a very bad news for china) . We have to factor this uncertainity in when projecting future pipeline for nvidia
Yes. This is a long process spanning multiple jurisdictions.
There is history of china blocking.
Intel is out for an year atleast
Next to watch is the GPU wars. AMD launch coming 28th Oct.
NVDA is cancelling lineups (shipping only those numbers they are obligated to deliver).
This has allegedly angered the AIBs (card manufacturers) that were lining up their production line to manufacture these cards. I think there is some weight to this. It has also pulled the RTX 3070 date ahead to 29 oct. This is a day after amd launch on 28th. Prepping for dealing with AMD launch impact.
IMO, if AMD prices above NVDA (especially 3070), that would mean that they are very confident that NVDA cannot match their volumes. They will simply wipe the floor in that range.
intel Data center revenue down + reaffirmed delays in latest gen processors.
They have blamed it on covid + others are saying something called capacity digestion (what is this?). Let us see what AMD reports tomorrow.
AMD earnings out
They do not report split in revenues between ryzen/EPYC/semi-custom . But the commentary is that more cloud/HPC are coming for EPYC.
Oracle is building on every processor out there… amd/intel/nvidia a100/arm . Secular story! Right?
Note: Semi-custom is majorly gaming consoles - https://www.amd.com/en/products/semi-custom-solutions
AMD launch of radeon for this year is done.
They say they are matching NVDA.
Better perf when paired with AMD CPUs - Only AMD can do this. Because they do both CPU and GPU.
- Anyone who has see the presentation is going to halt buying rtx3070 until Nov 18th. Unless they have reasons other than perf to buy rtx30 series.
- They are still sticking to last year’s 7nm node for radeon. Mature process compared to samsung 8nm that NVDA uses. It is a big mystery about NVDA GPU delays. If it has something to do with new samsung 8nm process node.
- They are matching NVDA perf with 7nm node!
- Smart access + Rage mode shown for all graphs - not sure if including that or excluding that.
Some special attention to PC designers + emphasis on software stack (major pain point with radeon)
All of the perf numbers will need confirmation from Independant testers. on Nov 18.
3070 competition (6800) is priced at 679$. This is more than 3070. I really suspect, 3070 has production issues now and AMD knows it. Else they might slash it afterwards.
EDIT: I totally missed the 6700 & 6700XT. This could be the real competitor to 3070. This will be revealed in sometime when 6000 series hits the market. This could totally screw NVDA lineup. Very Very interesting. The ball is in NVDA court now. Each NVDA card could have 2 AMD alternatives with more VRAM.
On jan 2021, when we look at gpu sales numbers, we will know amd has arrived in GPU too.
Even NVDA acquisition of ARM was about Total Addressable Market.
INTEL - So yeah - nothing in CPU till next year. Not to mention, this one also is in 14nm. Basically more clock speed. They have problems with their ‘10nm’ too ?
"Intel says the chips will come to market in the first quarter of 2021, which means we’ll likely see them sometime in April of next year, ceding the lucrative holiday shopping season to AMD’s Ryzen 5000 processors. "
HW alone at 133B$. This is not just gpu.
Data relavant to dGPUs
Great interview with a server engineer with 20 years experience
- Biz application lifetime is about 25 years!! servers still run applications go back to windows 2000. These will not work in AMD at all.
- Server engineers overall have moved to 3 year cycle compared to 7 year cycle earlier in their agreement with Intel. Ideally server engineers want to move immediately. Lesser cycle gives them better choices.
- Current EPYC CPUs are being validated. Server folks validate new component before bringing it online. But covid has pushed that out.
- Older applications may not move to AMD.
- The cost of switching is huge - unless AMD can give 4x perf do not expect any switching from intel to AMD. The cost is the down time in servers. 10% 20% better perf is not enough.
- Risk is premium - Money is not the problem. The cost is the down time. In healthcare, a down server can cause death.
- Older applications will not be switched to AMD
- AMD cannot go above 50% - Intel is well entrenched and whatever applications current work with intel based servers
- Enterprise servers - At current application counts - Even if Intel kept selling the same thing for 10 years AMD cannot go above 50% market share!! Steady perf is premium. No one is going to move applications to a new platform now- supply chain + stability + security is important.
- Most AMD can do is 35%!!
- If AMD has to go above 50% - they have to rely on amazon/microsoft etc. That too even 10 years is optimistic. Not to mention need for new applications in enterprise market to bring this change. Even for ARM. All things same - 30-40 years.
- When he was running the show in a healthcare server - Moving from intel 32 bit xeon to 64 bit xeon took them 8 years.
- Intel security issues caused the move to zen3 but delayed due to covid. Otherwise, they would have waited for things to settle down. May be zen 6 in 2024!!
- AMD is very competitive in 2020 than before with their support + new hardware etc.
- Now that cycles are 3 years - Cost of purchase starts to matter now compared to Total Cost of Operations
- Enterprise buy Intel because they have to - Old applications. Just keep things running. 14nm works too. But the margins keep reducing for intel.
- NVDA CUDA has 10 year head start - Very difficult to switch to AMD - Platform is sticky.
- Radeons have a shot in supercomputing space.
- One thing going for AMD is openness. In general the industry prefers open standards compliant stack. NVDA tries to tie users to its stack + applications - A problem with NVDA GPU software. AMD is open friendly.
Much more. I am next going to see how much money does 35% mean.
Here are the bullet points from AMD, and we have segment-by-segment breakdowns below, including historical data:
AMD overall x86 CPU share was 22.4%, an increase of 4.1 share points quarter over quarter (QoQ) and 6.3 share points year over year (YoY): Highest share since Q4 2007
AMD desktop x86 share, excluding IoT, was 20.1%, an increase of 0.9 share points QoQ, and 2.1 share points YoY: AMD desktop share has grown over 12 consecutive quarters: Highest since Q4 2013
AMD notebook x86 share, excluding IoT, was 20.2%, an increase of 0.3 share points QoQ, and 5.5 share points YoY: AMD notebook share has grown over 12 consecutive quarters: This is a new record for AMD x86 notebook share, eclipsing the previous high of 19.9% set in Q2 2020
AMD client x86 share, excluding IoT, was 20.2%, an increase of 0.5 share points QoQ, and 4.3 share points YoY: Highest since Q2 2011
Ryzen 5000 has landed. Their perf numbers are validated by third party to be above what AMD showed in their presentation earlier. (note that AMD did hint about it when they said ‘you can set various world records of your choosing’. This has to do with memory speed. This processor is hungry for memory speed.)
Dislodged Intel from top performance comfortably. Dislodged even from single thread performance. We now know why they placed the price above intel’s this time.
This also indicates that AMD graphics numbers could be at least what they presented.
NVDA is re-adjusting there entire line up in wake of AMD launches. Buyers are already not getting their hands on NVDA GPUs. Not sure if scalpers have been the main reason for low supply of rtx to consumers. The early launch, the price fixing (rebate to AIBs) and subsequent maneuvering around AMD announcement tells us that NVDA was throttling supply to create demand scenario.
To sum it up: If AMD can supply enough cards this season, they will have a great season - I will compare this with zen1/zen2 times in CPU (if SW stack plays out fine). Not to mention TSMC 7nm vs samsung 8nm (NVDA) process yields. Do note that this is not Intel sitting like a duck in 2018-2019. NVDA is responding. But this could hurt its 60% margin.
A little off topic below. Kindle let me know. I will delete.
Hopefully those wanting to buy rtx cards in this forum waited for sometime. New lineup will have to competitive with AMD and hence will give great value/$.Very likely you will get 3080 Ti for $1000. Just compare with AMD per/price lineup and NVDA will at least try to match that. You can do your math on savings.
I honestly think 100% of CPU market will be ARM in 10 years time. Intel and AMD would be close to dead unless they can figure out something in time. Performance per power consumption matters a lot in servers & laptops and ARM wins heavily there.
Your twitter link is 2018. Ampere has made great strides from that point on. Go through this thread to know more.
10 years is a long time. AMD worked on its new architecture somewhere around 2012-2014 timeframe. And in 6-7 years, is kicking @$$. Server biz is not just about CPU/GPU. It is a platform business. Watch NVDA ARM takeover + AMD xilinx take over interviews to understand more.
We are seeing that the ZEN architecture is driving AMD resurgence in CPU space.
In HPC (High Performance Computing) space, two days back, AMD released its first CDNA Architecture based Compute GPU - MI100. Note that AMD has separate architecture for Gaming GPU and Compute GPU. This is unlike NVDA that uses the same architecture for both. This lets AMD focus on compute separately.
Another difference compared to NVDA is the infinity fabric (or say something that connects CPU and GPU in layman terms). AMD CPU and GPU can use the same memory. This incentivises use of both CPU and GPU in the same system from AMD for added benefits. This is an example of a platform (not just selling GPU or CPU alone).
NVDA is still the king in AI workloads though. And Mellanox+ARM will help NVDA in the future for selling platform instead of GPU alone. Read the following article to understand more
Not going into too much technical detail.
Early player Advantage + CUDA affinity + AI perf is with NVDA.
New architecture + CPU & GPU combine availability is with AMD.
Finally competition is moving to cash cow datacenter biz.
AMD Radeon cards are now available with third party testers. (This is not about datacenter but consumer market)
AMD may have just pulled a leveller in GPU vs NVDA. Again, there are multiple areas to look at when comparing and hence the initial reviews of testers is mixed (Example: Preference to one area, say 4k gaming, will tilt teh discussion in favour of NVDA). But overall, it is positive for AMD since they are giving out something that is indeed competing with NVDA RTX.
But what stood out for me was linux support. This is unexpected. It appears games in Windows VM runs under linux is spotless. This, to me, indicates that AMD has done their homework with the sw quality feedback. Looks like Dr Lisa Hsu, pulled up their SW team and told them to work on every feedback, windows or linux. If they are going to give out such support for linux, I expect the windows software quality/stability to be very good.
- Card availability is still a problem - although reports are that stocks were better than rtx - This will determine how good their profits are this quarter.
- We will have to wait for the fog to clear out and get clear verdict.
The fact that they did well for linux machines for games. And the word is that they have some 60 developers to do this. This tells me that AMD is still a company run by engineers (and not just business men). And we can expect better technology from these guys. But then, who is to say that supporting linux is not good for biz? Only that their approach is very open and towards open standards/source.