Question Rome Launch 7 August , what about TR?

[Topweasel]

Is the siege mentality still required for AMD fans? Honestly they deliver good product at high end prices, there are is no free stuff in their price list.

I was just pointing out that noone is actually buying those 28 or 64C monsters, for average server sale there are always sweet spots and before Rome that spot was that Xeon.

P.S. before pointing out the "obvious" about pricing revolution, consider that noone is paying list prices, If my company isn't You can bet a farm that others get great deals too. Expect even greater deals from Intel now

I get what you are saying. But it's not true. Intel doesn't develop a whole new solution for Cascadelake or have a 14nm shortage because their HCC dies are the ones flying off the shelves. Intel's XCC dies are flying off the shelves. The I need a building just for my datacenter people and the cloud/VM guys are gobbling them up at a crazy rate. Remember those guys were so desperate they were buying server Skylake for almost a year before it was finished with functionality turned off just to increase the compute density.

These are more then just Halo products they are the heart of their server markets and Intel was desperate to keep up that they had to resort to half butted attempt to glue some dies together.

 

[Markfw]

Real CEO's (unlike armchair ones like you) know there's a lot more to an enterprise class server than just the CPU.

Actually real CEO's leave this decision to the CIO, because it's their job.

I worked in a major company with square miles(literally) of floor space for servers. And performance and power consumption are king. I knew the CIO personally, as I worked in IT.

So don't tell me I don;t know what I am talking about,.

 

[JoeRambo]

Not sure if I can reveal who I represent, but I've had my eyes on the 64 core rome for some time now. While we don't have the bulk purchasing power of the big cloud providers, our purchases are critical. Those 28 core Xeon are eye watering expensive but we still have to buy them. 64 core rome will allow us to dramatically decrease our purchasing cost while further reducing the footprint. It's a huge game changer for what we're doing if I can convince those above me we shouldnt just stay safe with Intel.

In my company, we were using 2x20C Xeon Golds and are now considering replacing (some of) them with 1S 48 or 64C Rome. It is just damn good CPU and while Intel 2S NUMA was pallatabble, even better for some of our loads is having no NUMA at all and very uniform and predictable latency for inter core communications.

I even had this very thing pointed out in a thread a few weeks ago:

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...tencies-epyc-advances-in-server-rooms.2567277

But that of course does not stop ADF guys from attacking anyone mentioning that XEON used to be best buy before Rome

But i know for sure from our middle man who deals with servers from companies starting from H and D, that golden spots of retail are not those 28C monsters and even our 20C CPUs are rather exception, price/perf curve has peaks somewhere else.

 

[Kedas]

if I can convince those above me we shouldnt just stay safe with Intel.

Who told you that intel CPU's are safe?
The only safe thing about it for now is that the next security problem will most likely also hit your competition since they have the same CPUs, but your chances are dropping and keep dropping over time. (more EPYC market share)
The right way of thinking should be go EPYC to go eco, faster and safer.

 

[Kedas]

I didn't saw any mention of a new chipset? it's up to the manufactures to support DDR4 3200 and PCIe4 in the bios update or not.

2nd Gen AMD EPYC™ processors used on motherboards designed for the 1st Gen AMD EPYC processor require a BIOS update from your server manufacturer. The EPYC 7742, 7642 and 7542 are 225w parts and require additional updates, contact your server manufacturer for support. For PCIe® 4 and DDR4-3200 memory support, please contact your server manufacturer. A motherboard designed for 2nd Gen EPYC processors is required to enable all available functionality. ROM-06a

For TR we do expect a new chipset I assume even if it's only to differentiate PCIe4 support.

I see 8 core EPYC2 so we may see an 8 core TR also?

 

[SPBHM]

very impressive performance from all the tests I've seen, also 64 cores with decent clocks and 225W TDP...
Intel is really behind for now in so many metrics.

the only bad thing that I can see are those latency numbers, thinking of TR which might be used for more consumer applications, maybe even some gaming that could be problematic, or maybe it's better with just 4 active dies?! or with a different I/O die!?

it will be interesting to see what happens with TR no doubt.

 

[Gideon]

The Boost algorithm looked really impressive. I can't find that info in Servethehome review anymore, but the top-of-the line model boosted to 3.2GHz all cores constantly in some integer workloads. With more stressing AVX2 workloads it went down to 2.4. Intel can't sustain speed anywhere near that on all cores

EDIT:
@Hitman928 has the relevant screenshots in this post

 

[beginner99]

Question as I'm not workign in server industry:

Do the list prices actually mean anything?
Or does Intel just offer higher rebates?

I mean for enterprise software (except probably oracle) list prices are a joke. If your dumb, you pay 50% of list price, mostly less.

 

[Gideon]

From the ServeTheHome review:

On the other hand, the dual AMD EPYC 7502 is faster and uses $2600 each list price SKUs while the Intel Xeon Platinum 8280 is a $10,007 list price SKU. AMD provides four more cores per socket which help performance, but they are doing so at an initial list price discount of around 74%. That is a big deal if you assume Intel Xeon Platinum list pricing is designed so server vendors can utilize large 60% discounts. AMD still has the better platform, but we think here the Xeon Platinum 8280 can be very competitive with an 80% discount off of list price.

It's true that Intel top-of-the-line units list prices aren't that relevant. Server vendors buying in bulk can get considerable discounts from Intel (60% mentioned in the article), but the thing is, nothing stops AMD from doing the same, if large contracts are on the line. If anything they are in a better position considering the margins they require and the need to get market share

I find it slightly aggravating that people act like discounts are some 23rd Century technology only Intel can offer, but AMD absolutely must sell at list price.

 

[Topweasel]

We also have to remember that a lot of the buying craze on Intel Server chips happened because a bunch of companies had to make up for a 20-30% reduction in performance as the initial Meltdown and Spectre fixes specifically hit the performance in functions that they primarily used. Even if the prices closed up due to discounts. There may be a strong push to move away from a single supplier. Specially when one offers better performance, IO, memory bandwidth/size, and hasn't experienced any performance crippling security patches.

 

[moinmoin]

From the ServeTheHome review:


It's true that Intel top-of-the-line units list prices aren't that relevant. Server vendors buying in bulk can get considerable discounts from Intel (60% mentioned in the article), but the thing is, nothing stops AMD from doing the same, if large contracts are on the line. If anything they are in a better position considering the margins they require and the need to get market share

I find it slightly aggravating that people act like discounts are some 23rd Century technology only Intel can offer, but AMD absolutely must sell at list price.

AMD should just publicly offer price matching at the same performance point for whenever Intel makes such discount offers.

(Who else had to laugh out loud at the "AMD still has the better platform, but we think here the Xeon Platinum 8280 can be very competitive with an 80% discount off of list price." quote from STH?)

 

[positivedoppler]

Who told you that intel CPU's are safe?
The only safe thing about it for now is that the next security problem will most likely also hit your competition since they have the same CPUs, but your chances are dropping and keep dropping over time. (more EPYC market share)
The right way of thinking should be go EPYC to go eco, faster and safer.

You don't have to convince me, but I have to convince many others. We are far more conservative vs normal users in terms of embracing new technology/vendor. With our 28 core Xeons, we don't even risk utilizing multi-threading and this has nothing to do with security. We are using VMWare and virtualization is something new to even me. I have to offer almost bullet proof guarantee that porting 28 core Xeons to 64 core AMD will have almost zero risk

 

[moinmoin]

You don't have to convince me, but I have to convince many others.

Do those above you get all the unfiltered news regarding security issues in CPUs, be it in Intel's products or others?

 

[Kenmitch]

I find it slightly aggravating that people act like discounts are some 23rd Century technology only Intel can offer, but AMD absolutely must sell at list price

List prices are just some made up figure that a company would be happy to sell their products at. You need to have a starting point for negotiations, but not high enough that you devalue your own product during the process. Imagine your pricing out a server and Intel wants to charge you 10k a cpu, you mention Rome and suddenly they offer them for 6k....Would you consider it a good deal?

 

[Atari2600]

AMD should just publicly offer price matching at the same performance point for whenever Intel makes such discount offers.

(Who else had to laugh out loud at the "AMD still has the better platform, but we think here the Xeon Platinum 8280 can be very competitive with an 80% discount off of list price." quote from STH?)

I had a few laughs during the STH the article. There were a few barbs flying.

 

[maddie]

List prices are just some made up figure that a company would be happy to sell their products at. You need to have a starting point for negotiations, but not high enough that you devalue your own product during the process. Imagine your pricing out a server and Intel wants to charge you 10k a cpu, you mention Rome and suddenly they offer them for 6k....Would you consider it a good deal?

I suppose cultural differences matter, but I would be pissed if a vendor is suddenly willing to accept 40% off the original quote. Why not before? Was I your cash cow?

 

[Guru]

AMD are having trouble keeping up with demand just for their desktop processors, now they have to keep up in demand for servers too and their mobile/tablet/laptop business is increasing as well, so more demand. While many of their products for mobile/tablet are still on older process node, they can't push desktop, server cpu's and RX 5000 series gpu's on 7nm, and have threadripper as well.

So based on logic and most recent news and facts, the soonest AMD is going to release a Threadripper product is Q1 2020.

 

[Yotsugi]

So based on logic and most recent news and facts, the soonest AMD is going to release a Threadripper product is Q1 2020.

No.
They will launch it circa Cascade HEDT launch window.
Spare no enemy behind!

 

[Kenmitch]

I suppose cultural differences matter, but I would be pissed if a vendor is suddenly willing to accept 40% off the original quote. Why not before? Was I your cash cow?

I think the same way over here. I doubt my response would sound as civil as yours tho.

 

[moinmoin]

AMD are having trouble keeping up with demand just for their desktop processors, now they have to keep up in demand for servers too and their mobile/tablet/laptop business is increasing as well, so more demand. While many of their products for mobile/tablet are still on older process node, they can't push desktop, server cpu's and RX 5000 series gpu's on 7nm, and have threadripper as well.

So based on logic and most recent news and facts, the soonest AMD is going to release a Threadripper product is Q1 2020.

The problem with supply is not due to TSMC lacking capacity, it's due to the production lead time being 2-3 months from initial order to finished product, and apparently AMD didn't predict the launch demand well.

 

[Markfw]

The problem with supply is not due to TSMC lacking capacity, it's due to the production lead time being 2-3 months from initial order to finished product, and apparently AMD didn't predict the launch demand well.

The 3600,l3600x 3700x and 3800x have been pretty well in supply, it seems as though the only OOS on a regular basis is the 3900x. And it looks like in 8 days, that could change.

 

[moinmoin]

The 3600,l3600x 3700x and 3800x have been pretty well in supply, it seems as though the only OOS on a regular basis is the 3900x. And it looks like in 8 days, that could change.

Indeed. So one shouldn't take that temporal scarcity of a specific model as an indication on how limited or not supply will be for Epyc 2 and when Threadripper 3 will release.

 

[Kedas]

Since the 3900X is the only one with 2 dies it could also be an assembly bottleneck and not a die supply bottleneck. (that doesn't take that long to fix)

Or they made "small" calculation error:
"How many core dies do we have for 3900X?"
"20 000 dies"
"Ok 20 000 3900X CPUs is enough, check"
oh wait that only 10 000 CPUs

 

[gk1951]

Markfw I have a quote that supports AMD's efforts:

Who said the quote Rome wasn't built in a day?
John Heywood

Rome Wasn't Built in a Day, But They Were Laying Bricks Every Hour. John Heywood was an English playwright who lived hundreds of years ago. Today, Heywood is known for his poems, proverbs, and plays. But more than any one work, it's his phrases that have made him famous.

AMD has been "laying the bricks" for quite sometime and now Rome appears in site.

Great for AMD, great for the server industry and great for future TR builders.

 

[Kedas]

I have to offer almost bullet proof guarantee that porting 28 core Xeons to 64 core AMD will have almost zero risk

It's never without risk but you have to convince them that not changing may be a bigger future risk/cost.

edit: just saw this

I'm just wondering how long before we can update CPU's like this guy 'floep'