Hacker News Reader

Comment by ksec | Hacker News Reader

This is one of those Shut up and Take my Money Product announcement.

~20% Increase in IPC, ~10% Increase in Boost Clock Speed. It doesn't matter how Intel spin it, single thread performance will no longer be an Intel only selling point. 32MB L3 Cache is going to be very useful for certain types of Application.

Some of these were rumoured for quite some time. But having it confirmed is a completely different matter. And the pricing is still very good compared to what we used to get from Intel.

My only concern is stock availability. And not just at launch but over its life time. AMD has been very conservative with their production estimate. ( Again It is not TSMC's fault ) It wasn't long ago they were on the verge of bankruptcy, so it is understandable but at the same time I wish they took a little more risk.

It is also interesting there is no mention of EPYC 3. I am quite concern about their lack of progress in the Enterprise / Server segment.

View on HN

Isn't this how it's always been? New tech shows up in enthusiast/consumer segment first, bugs are ironed out, manufacturing ramps up & yields go up, then new server parts are announced?

Besides, isn't EPYC2 the best server CPU already? There's no time pressure on AMD, they're comfortably in the lead.

Regular EPYC CPUs have too high a TDP for smaller server installations, such as so-called edge servers. EPYC Embedded is still stuck on Zen 1. Intel hasn't upgraded their comparable line of mid-range server CPUs (e.g. Xeon D line), either, but AMD won't win on performance alone. Intel has a huge SKU lineup, much more volume, and a much richer vendor and platform ecosystem. If the vendor demand isn't there to push AMD on EPYC Embedded, Ryzen Emedded, and other market segments, AMD should build demand, otherwise they'll just recapitulate their rise & fall during the 2000s.

Cloudflare recently announced that they would be building out some data centers with EPYC CPUs. Do you believe that the situation has changed since February? They did a pretty exhaustive analysis [0] of price, performance, and power where they saw an advantage in switching away from Intel for that generation at least.

If the existing server hardware is already in a decent spot, then maybe they need to spend more resources on sales rather than making changes to keep it cutting edge.

[0] https://blog.cloudflare.com/an-epyc-trip-to-rome-amd-is-clou...

I think Cloudflare is a different kind of "edge" service than what EPYC Embedded targets, and I would think Cloudflare uses regular EPYC chips. The "edge" that EPYC Embedded, Xeon D, etc target is, I think, more about the hardware configurations (smaller enclosures, minimal number of drives and other devices) and the type of facility (usually not a colocation facility, so power and heat are of significantly more concern). But the workloads are still very much server-class.

EPYC Embedded chips are competitive with Intel's offerings (Xeon D, etc), but as I said Intel's ecosystem is richer--for example, more, better motherboards. It's not enough that AMD's chips are competitive with Intel's. AMD has a huge ecosystem handicap, and so either they need to improve the ecosystem or sell chips that are dramatically better than Intel's, and for EPYC Embedded neither is the case. Long term a better ecosystem is necessary for AMD to survive because a broader product and customer base brings consistent income and mindshare--staying power. I would hope and assume AMD is working closely with cloud vendors on their proprietary hardware, but the results are opaque; judging by traditional channels (Supermicro, etc), AMD hasn't even begun to close the gap.

I don’t think survival is at stake. AMD powers both XBox and PlayStation for the second generation in a row and that ought to help keeping them alive. It’s more a matter of whether they can capture enough of the market so they’re not merely “surviving”.

Fair enough. Survive was a poor choice. What I had in mind was surviving as a contender at the high-end, so we can continue to benefit from competition for server-class chips. Alpha, SPARC, and POWER all lost the high-end market (their only market) to Intel at a time when Intel was inferior. AMD previously failed because Intel surpassed them, but that's because AMD couldn't leverage their initial advantage to secure their markets and thus their ability to keep investing. Without volume and mindshare failure is inevitable. Providing the best high-end chips is insufficient to remain competitive long-term. The reasons for previous failures were complex (ISA, operating system, sales channels, etc), but ultimately it comes down to something like diversity--customer and product diversity provide buffers in terms of sales as well as changes in market direction. AMD's chips are indisputably better than Intel's right now, but even with Intel's mindboggingly massive fumbles they're barely sweating in terms of current and prospective revenue.

> Providing the best high-end chips is insufficient to remain competitive long-term

Reason Alpha SPARC and POWER failed was they did not support the de facto standard X86 instruction-set, I believe. So the fastest most power efficient chip is not enough unless it can also run the most common software

> most common software

Maybe it is different this time because well for one Intel and AMD are on the same architecture

but more importantly for arm or other platform, maybe this time is different because the servers mostly run on open source software. Given the potential revenue, I'd imagine it should be pretty straightforward to port common server software to any architecture if it isn't already?

In this case "most common software" meant proprietary and nonportable operating systems and applications. Would you buy a server that will run the software you want to use either never (Windows and Windows applications developed on Windows) or maybe, eventually, as long as the new platform is a success (only slightly more portable proprietary Unix variants)?

> Alpha, SPARC, and POWER all lost the high-end market (their only market) to Intel at a time when Intel was inferior

What's interesting is that the industry was so sure that Itanium / Windows NT was going to crush everything that many of them just gave up. Compaq and Silicon Graphics specifically tired to switch to Itainium, which was years late and never a huge success. They probably could have gotten another generation or two in. Might have saved SGI.

With the kind of marketshare ryzen has gotten on custom builds plus several % increase in server( which has large margins) I don't think amd has financial issues right now. They'll probably target gpus next, go for the low hanging fruit and then slowly iterate on the rest of their products.

And before that Cloudfare announced that they are migrating to Arm servers.

Turns out it just wasn't performant enough for their applications, per this follow up post:

https://blog.cloudflare.com/technical-details-of-why-cloudfl...

ARM just needs to be a little bit more competitive on performance per watt and a few other metrics...I think when Apple inevitably either proves ARM can be a very powerful platform via their migration to complete ARM based computing, we'll see a lot more movement and investment here, I'd say ~5 years is a good timeframe to start seeing serious long term shifts to ARM. If Apple fails to do this within that time frame, I think that may push it out longer. They're already pushing the limits of ARM as a platform with their custom chipsets

I think nvidia's long term strategy is going to be built around this, among other things, otherwise I believe they overpaid for ARM holdings. They'll need to make sure they can expand the market for ARM CPUs to drive ever more GPU sales in the future. They haven't been as successful getting their embedded GPUs into the ARM market if I recall correctly

AMD and Hector Ruiz sat on their temporary advantage and were in denial of Intel's investment in the Core architecture.

AMD is keeping the throttle down this time.

> but AMD won't win on performance alone.

Yup. Let's talk about price and TDP...

> Intel has a huge SKU lineup,

Why is that supposed to be an advantage?

> much more volume,

> and a much richer vendor and platform ecosystem

This is just marketing-speak.

richer vendor...ecosystem = more resellers

richer...platform ecosystem - what does that even mean?

> AMD should build demand

How?

And have you noticed how many people here saying they want to buy intel?

> otherwise they'll just recapitulate their rise & fall during the 2000s.

Who's in the stronger position now in terms of price and performance?

A huge sku lineup according to every marketing book every made is a terrible idea here. I hope they don't view it as a strength. I'd be slashing like hell.

AMD lacks AVX512 instruction set, which is a show stopper for many applications.

1. not many 2. avx2 with twice the cores is about as good as avx512 a lot of the time

Which applications, and why? Some computational ones will go significantly slower on the same number of cores if they could have kept avx512 fed (perhaps small data in cache), but most don't spend all their time in something like GEMM. The new UK "tier 1" HPC system is all EPYC.

Automatic vectorization for AVX512 would work on the simplest of cases and using intrinsics or writing inline assembly is beyond the scope of 99% of software projects.

AVX512 is garbage. It incurs a massive performance hit from both frequency and mode-switching penalties. Apart from a few niche HPC and ML applications which you'll never encounter, AVX512's most compelling use cases are to drive Intel's shady market fragmentation, and to create more bullshit FP benchmarks that Intel can claim to win.

https://www.extremetech.com/computing/312673-linus-torvalds-...

Now that we've covered both ends of hyperbole, it's maybe worth noting that a lot of CPU parallel tasks can be well accelerated by ISPC, which can make reasonably effective use of AVX512 (with the aforementioned clock speed caveat): https://www.mail-archive.com/ispc-users@googlegroups.com/msg...

Also, AVX512 is a much nicer (orthogonal) ISA than SSE or AVX2.

While there are some niche applications that need larger amounts of memory than GPUs can offer, it's worth noting that this speedup comes from making the CPU act more like a GPU, and they aren't as fast as GPUs acting like GPUs (which are essentially many 32-wide vector units, rather than AVX512's 16-wide).

Nah, to be fair, the most compelling application is: https://gamozolabs.github.io/fuzzing/2018/10/14/vectorized_e...

This is not true anymore: https://travisdowns.github.io/blog/2020/08/19/icl-avx512-fre...

That's an i5. When Cloudflare talked about the issue, they were using high performance Xeons.

I'm aware, but the ice lake xeons are starting to ship.

> a show stopper for many applications

It's really not.

There's a very small set of applications where the precision of AVX512, but more cores also speeds them up.

A much more important set of applications is those sped up with GPUs or other ML accelerators. Notably the high performance of these CPUs is useful with those too, because they are great at data pipeline crunching prior to the GPU part.

It's not a given at all, even if that's common. As a fresh counter-example, Nvidia released Ampere on TSMC 7nm for the enterprise many months before they released chips of the same architecture for consumer devices on Samsung's cheaper and significantly less dense 8nm node.

> There's no time pressure on AMD, they're comfortably in the lead.

Intel thought this way, and that's why AMD are where they are now.

And Microsoft to a lesser extent.

Sitting on huge piles of cash helps though.

My last AMD processor was the Athlon 64 x2 circa 2005, it's going to be nice going back.

I ran one of those until a couple years ago when I got the 2400G APU. What a nice upgrade. Even better parts available now.

I'm using a AMD Athlon II X2 260, released in 2010.

I'm still running an Ath 64 x2 6000+ for a web server, I think it's from about 2006. That with an old hard drive are running 120W, which would be nice to hack down.

I don't remember if I jumped straight to this from my Pentium Pro 200, but the role of this box started with that one.

My Athlon is in a case next to my left knee. I was almost scared off by the Intel hyper but not too sure of AMD spin. We're all casualties of marketing departments. In the end I bought AMD because it was cheaper than Intel.

Yeah I had one of those. It was great but so noisy.

Then Core2, then i5. I built my son a computer last year on Ryzen and it's great, but I don't see myself upgrading my i5 for a couple of years. My money will go on GPU upgrades (currently a 1070) instead I think.

I just upgraded an Athlon64 x2 to a Phenom II x4 because I could get the chip for $20 and a cooler for $30. Heck of a speed boost on that box, which was looking a little long in the tooth otherwise.

It's now beefy enough to run a modern OS and GUI and even do some light gaming on, but the mobo has serial and parallel and PATA and floppy ports so it serves as my media archiving mule too.

With a 125W TDP, you're paying in other ways.

A 2018 Celeron J4105I is a 10W part that's only fractionally slower than a 965BE but costs $6 a year to run instead of $60.

Oh for sure, the 4850e was purpose-built for low power when it was my daily driver. Now it only runs a few hours a week if that.

I should grab the meter and see if its idle power consumption changed much, though. Obviously it's not running hard most of the time...

I ran a Phenom II x4 DDR2 until this summer when I bought a Ryzen 3700X. It could not play Battlefield, Rainbow Six Siege, Squad, Far Cry, Pubg ( could almost play, loading stream time was the killer), Mordhau and the likes, but it could play a lot of games, so many that I didn't need to upgrade, it was the computer as a whole that was giving up.

I could play Wargame Red Dragon, Dark Souls 3, Dota2, Dishonored 2, Remnant from the ashes, Warhammer 1 total war, Disco Elysium, Dragon Age Inquisition, Fallout 3 New Vegas.

Athlon 64 was the bomb at that time

> 32MB L3 Cache is going to be very useful for certain types of Application.

"Going to be"? The Ryzen 3600 ($199) and above already have 32MB L3 cache, the 3900X ($499) and above already have 64MB. The big L3 cache is already a big selling point of Zen 2.

The difference they talked about is it is now uniform latency instead of split into two blocks of 16MB accessible at different latencies depending on which group of four cores you are on.

3990X has 256mb of L3.

Yeah but it's not as relevant to the discussion given it's in a different class of CPU not really aimed at average consumers. It's fair to mention it, though.

But only 16MB of L3 is accessible to each core in Zen 2 - no single core can utilize all 256MB of L3 cache. That's why Zen 3 cores being able to access the full 32MB of L3 cache that's on die is a big deal.

> 32MB L3 Cache is going to be very useful for certain types of Application.

Are there applications where more cache is actively detrimental?

No? Other than the next best use of that silicon/heat budget, to whatever degree the latter is relevant for cache. But not afaik from an architectural point of view.

General rule of thumb is that bigger caches are slower to access.

If AMD have traded a bit of access latency for their larger cache, then theoretically there will be a memory heavy application with a working set that fits in 16MB that will see a preformance hit.

Though, we don't know Zen 2 was running into area based speed limitations for it's L3 cache. It's entirely possible Zen 3's cache runs at the same speed.

Slower, but always faster than dram. So no, there are no downsides.

A larger but slower cache would be a downside for an application that did not need the extra capacity. Now it's cache accesses are slower with no gains anywhere else.

Of course if you don't have enough memory at a given tier and are overflowing in to the next adding more will always make things better, but once you hit that point adding more just adds cost at best and might actually make things worse if it's slower for the extra capacity.

There are a bunch of downsides to large caches. Everything from access time to chip layout to latency and even power consumption. Everything cant be cache and be awesome.. :P

Having more cache can potentially lower the speed of the cache, as the access time is limited by the time the longest path takes, the propagation delay.

So there's a tradeoff between cache size and cache speed, which is why there are separate L1, L2, and L3 caches of various sizes. So potentially the L3 cache in this architecture could be slower than the L3 cache in the 3000 series. It could also be the same speed if the size was limited for other reasons, such as yield.

AMD claims a significant improvement in memory latency though, which is concordant with their large gains in gaming workloads (a 20 % general-purpose-throughput-oriented IPC increase alone would never give you a 20 % FPS increase in games).

A larger cache size would improve memory latency assuming the working set can utilize the full 36mb, which I'm sure the 2 games that had a 20% uplift can.

It's purely speculation but I suspect the cache size was limited by yield concerns rather than timing constraints. It looks like the 5600X has 1mb less cache so they probably engineered a way to disable faulty sections of the cache on a 1mb granularity.

Edit: My speculation's wrong. The cache difference between the 5700X and the 5600X is due to core count differences. It's the sum of the various cache sizes, and I misread the slide.

> a 20 % general-purpose-throughput-oriented IPC increase alone would never give you a 20 % FPS increase in games

Is this true even for games that are CPU-bound? When I play MS Flight Simulator, enable the Dev toolbar, and look at the framerate monitor, it tells me that it's spending 20 ms of CPU time per frame, which causes my framerate to cap at 50 fps. A 20% increase in IPC would theoretically bring the frame time to 16.67 ms, giving me a cap of 60 fps.

There was a now-deleted comment about the CPU busy-waiting for the GPU, to which I had this reply:

Reviews/benchmarks of Flight Simulator by e.g. Gamers Nexus show that Flight Simulator is heavily CPU limited, running on a single CPU thread.

isn't the difference between L1, L2 and L3 also because of the functionality, not just the size+speed? L1 is data and code. L2 is data only, per core. L3 is data, synchronized between cores.

Yeah, technically there are 2 L1 caches; x86 is a 'Modified Harvard' architecture. The instruction cache also typically has to deal with caching micro-ops. I believe L2 and beyond store both instructions and data. There's also cache associativity, where the the same location in memory can be stores in one of N locations, which can differ per level. I think L1 caches are typically more associative because that takes extra silicon per byte. It looks like Zen 2 at least has an 8 way associative L1 cache.

With smaller caches and especially L1 you typically want higher associativity.

With 32KB L1 at 64B line size, you can only cache 512 lines. Grouping them in larger buckets means less spillage as hot lines randomly end up in the same bucket.

L2 and L3 are unified caches, they contain both data and code.

L1 code and L1 data caches are separate.

While this is true, in practice for the vast majority of applications this is a good tradeoff since the relative slowdown of L3 cache vs. the improvements in reductions of cache misses ends up being tiny:huge.

I'd expect the workloads that could suffer (all else equal) would be something like SIMD optimized matrix multiply where you're always able to prefetch the elements needed into cache effectively and memory access tends to be sequential. But those slight losses would likely be dwarfed by the improved core clocks, etc.

There probably are in which more cache is useless.

Exactly. All else being equal more memory is either neutral or better, but all else is not always equal. If the larger capacity memory is slower, then any applications that didn't need more memory just lost performance.

Not as such, but larger caches are slower (due to the speed of light) which can reduce performance. A few apps have seen performance regressions on Tiger Lake.

> but larger caches are slower (due to the speed of light)

Wouldn't it be more accurate to say due to gate delay because of an additional level in (de)muxing?

I assumed the slowdown is due to wire delay (a larger cache has longer wires) not gate delay but I could be wrong.

Cache access times are measured in nanoseconds. Light nanosecond ((c) Grace Hopper) is ~30cm

The performance is in the same ballpark for cost-performance as Intel 10th gen, and the 10th gen Intel platform will be compatible with Intel's 2021 chips, whereas if you get an AMD board now, this will be the last processor which it supports. The AMD boards are more expensive for comparable features. Intel 10th gen also has better all-core boosting, which might be important for upcoming games with better multicore support.

Intel current platform does not support PCI4 though, if that is important to you, you will need a new mobo anyways.

Also I don't think many people are going to upgrade from intel 10th gen to intel 11th gen, assuming gen 12 will have DDR5 (worth skipping a gen)

`buy the rumor, buy the news` -- amd

Reply on news.ycombinator.com