Question AMD EPYC 7742 Benchmarks and Review Simply Peerless

[nicalandia]

AMD EPYC 7742 Benchmarks and Review Simply Peerless

Our AMD EPYC 7742 benchmarks and review show why this is a simply peerless 64 core CPU capable of going head-to-head with dual and quad Intel Xeon CPUs

www.servethehome.com

I agree, these and the new TR3 3990X are Peerless for $6,950 a pop that is a steal

 

[Schmide]

You thought that I didn't read the article because I said that $6950 was not a terrible price for the best chip in the entire tested lineup?

Rather than argue about it, let me educate you on the elements of my post.

Perhaps you didn't know that "not terrible" (similar usage as to "not bad", "not too shabby", "not gonna lie", etc) is a common (if ironic) phrasing for affirmation and positivity. Here's an article talking about it, it's the last point. And another about its etymology. Usage such as mine, indicating "actually quite good" when saying "not bad" or other "not (insert negative)" has been around for >100 years. So when someone replies to your "How's your day?" with a "Not too bad, yours?" you should probably not respond to them the way you responded to me. It's not all that nice, and it seems ignorant.

As for waiting for Markfw's benchmarks, there's something humans engage called "humor". Here's a Wikipedia link for you to read about it. There are literally millions of videos on YouTube freely accessible for you to see what it is. George Carlin might be a harsh entry, but Jim Gaffigan and Brian Regan are great places to start. Also, there are many comedy clubs where you can go have a drink and watch them live, which is really fun to do with a partner or loved one. And wait til I tell you how many comedy specials there are on Netflix and Hulu and Amazon Prime!

I'm not against humor. I am against using it to mischaracterize or marginalize something. If something is close to an order of magnitude better in pricing than previous market value, "not terrible" is off base.You have every right to disagree. I have every right to point it out.

Using run of the mill colloquialisms may be fine for passing conversation and even to a point here.

People put effort into this article as did the engineers that created the product. It's not a joke unlike this thread.

 

[amrnuke]

I'm not against humor. I am against using it to mischaracterize or marginalize something. If something is close to an order of magnitude better in pricing than previous market value, "not terrible" is off base.You have every right to disagree. I have every right to point it out.

Using run of the mill colloquialisms may be fine for passing conversation and even to a point here.

People put effort into this article as did the engineers that created the product. It's not a joke unlike this thread.

In thinking about your post, regarding "close to an order of magnitude better in pricing" than previous market value, I would be interested to know where you got this data. EPYC 7601 was $131/core, 7742 is $108/core. That's not even close to an order of magnitude (at least as commonly used, 10x - though it's not even 2x the value, so I'm not sure what definition you mean when you say "order of magnitude").

IMO, "excellent" would be a better scaling of cost per core from bottom to top of the EPYC lineup, since cost of silicon doesn't rise exponentially with the amount you put on a chip, e.g. for excellent value, since the 8C/16T 7262 2P (boost 3.4GHz) retails for $575, I would anticipate a 64C/128T 7742 2P (boost 3.4GHz) might retail at $4600, plus a margin for binning I/O chiplets, for power consumption, etc., if they need to, though this doesn't intrinsically increase the cost of the production of the chip. Yet it retails for $6950, which is a 50% upcharge per core over the 7262. That's why it's not excellent in my mind. However, I understand that it's industry-standard to charge more for putting more power in a single socket, a "convenience charge" if you will. And it's in-line with their scaling on Ryzen and TR, roughly. And the market will still eat it up, as above, because plugging this much power into a 2P system will be worthwhile for a lot of people.

Another thing, the 7262 has 128MB L3 for 8 cores (16.5MB/core L1-L3), while the 256MB L3 for 64 cores (4.5MB/core L1-L3) -- which to me is somewhat striking. Well, anyway, I'm not a computer engineer. I'm sure those decisions have a reason. But considering the substantial die space cost of L3$, it lends even more credence to my thoughts that the 7742 is even less excellent from a pure value standpoint.

So, in my opinion, it's not excellent pricing when they're upcharging because they can, not because they need to. There's little "silicon" reason for them to price it where they have. Just because they're not, to use a colloquialism, screwing the purchaser as hard, doesn't mean that they're not still screwing the purchaser. It's just better than what is currently on the market from Intel. However, my key point about pricing is that their per-core pricing is worse than most of the rest of AMD's lineup and for no legitimate reason other than the convenience of having 64 cores in one socket, which does have value, but not to the point of a 50% upcharge. Hence, I wouldn't call it great or excellent or terrible. It's... not terrible. "Actually terrible" would be Intel's per-core pricing.

As for this forum not being a place for joking, that's your opinion, but it's not a forum rule, and it's fun, and I like fun, so I'll keep on doing it.

Regarding colloquialisms, it's very interesting that you use one in your own post. Regardless, I like colloquialisms, so I'll keep using them too.

 

[Markfw]

In thinking about your post, regarding "close to an order of magnitude better in pricing" than previous market value, I would be interested to know where you got this data. EPYC 7601 was $131/core, 7742 is $108/core. That's not even close to an order of magnitude (at least as commonly used, 10x - though it's not even 2x the value, so I'm not sure what definition you mean when you say "order of magnitude").

IMO, "excellent" would be a better scaling of cost per core from bottom to top of the EPYC lineup, since cost of silicon doesn't rise exponentially with the amount you put on a chip, e.g. for excellent value, since the 8C/16T 7262 2P (boost 3.4GHz) retails for $575, I would anticipate a 64C/128T 7742 2P (boost 3.4GHz) might retail at $4600, plus a margin for binning I/O chiplets, for power consumption, etc., if they need to, though this doesn't intrinsically increase the cost of the production of the chip. Yet it retails for $6950, which is a 50% upcharge per core over the 7262. That's why it's not excellent in my mind. However, I understand that it's industry-standard to charge more for putting more power in a single socket, a "convenience charge" if you will. And it's in-line with their scaling on Ryzen and TR, roughly. And the market will still eat it up, as above, because plugging this much power into a 2P system will be worthwhile for a lot of people.

Another thing, the 7262 has 128MB L3 for 8 cores (16.5MB/core L1-L3), while the 256MB L3 for 64 cores (4.5MB/core L1-L3) -- which to me is somewhat striking. Well, anyway, I'm not a computer engineer. I'm sure those decisions have a reason. But considering the substantial die space cost of L3$, it lends even more credence to my thoughts that the 7742 is even less excellent from a pure value standpoint.

So, in my opinion, it's not excellent pricing when they're upcharging because they can, not because they need to. There's little "silicon" reason for them to price it where they have. Just because they're not, to use a colloquialism, screwing the purchaser as hard, doesn't mean that they're not still screwing the purchaser. It's just better than what is currently on the market from Intel. However, my key point about pricing is that their per-core pricing is worse than most of the rest of AMD's lineup and for no legitimate reason other than the convenience of having 64 cores in one socket, which does have value, but not to the point of a 50% upcharge. Hence, I wouldn't call it great or excellent or terrible. It's... not terrible. "Actually terrible" would be Intel's per-core pricing.

As for this forum not being a place for joking, that's your opinion, but it's not a forum rule, and it's fun, and I like fun, so I'll keep on doing it.

Regarding colloquialisms, it's very interesting that you use one in your own post. Regardless, I like colloquialisms, so I'll keep using them too.

BTW, the only place I have seen the 7742 for sale by someone I would buy from, is newegg, and its not $6,950, its $7,544, which given that its the fastest and most cores on the planet (for anything someone could buy, like not a custom chip for 10 million) is not a bad price. As I have said elsewhere, if I was a little better off at the moment (I just bought a $55k truck), I would get a pair now. I just got 2 7601 retail chips for $27 per core, or $1,744. They are half the horsepower, but 1/4th the price, so not a waste.

 

[Schmide]

Order of magnitude it is not.

Previous pricing to current pricing. Intel was charging 10k+ an 8280, more if you wanted to use more memory. So straight out of the bat you have a more than doubling cores for 2/3 price. Intel does have some instruction advantages so add a precentage that way. Add to that the ability to put more cores in less space and draw less power. Although each of these is not a doubling of value, it is some multiplier that contributes to the total cost of ownership. The extra I/O adds density.

AMD preview showcased a slide with base 44% total cost of ownership with 45% less servers and 83% more performance. Since then intel adjusted their prices and AMD did as well.

 

[ClockHound]

BTW, the only place I have seen the 7742 for sale by someone I would buy from, is newegg, and its not $6,950, its $7,544, which given that its the fastest and most cores on the planet (for anything someone could buy, like not a custom chip for 10 million) is not a bad price. As I have said elsewhere, if I was a little better off at the moment (I just bought a $55k truck), I would get a pair now. I just got 2 7601 retail chips for $27 per core, or $1,744. They are half the horsepower, but 1/4th the price, so not a waste.

Detroit may love you, but this forum is stunned. More than a little upset at your priorities. You could have spent $54k on your computer and less than $1k on the truck. Here's a perfectly serviceable '95 F-250 with a $200 starting bid. Sure it might breakdown a lot, but then you just stay home and watch your pair of 7742s do stuff really quick.

 

[Markfw]

Detroit may love you, but this forum is stunned. More than a little upset at your priorities. You could have spent $54k on your computer and less than $1k on the truck. Here's a perfectly serviceable '95 F-250 with a $200 starting bid. Sure it might breakdown a lot, but then you just stay home and watch your pair of 7742s do stuff really quick.

I know you were being funny with the post, but seriously, with my disabilities, I needed a truck with all the electronic gizmos to help me drive, or I wouldn;t be able to drive. I am deaf, so I can't use the cell phone if it breaks down, and I am dizzy, so I have trouble driving at all.

But seriously, I will get them at some point.

 

[Thunder 57]

Did you read the replies to the article?

2nd or 3rd reply had someone who had an idiot boss who wouldn't even consider AMD. "nothing but compatibility problems" or words to that effect.

IMO that person's boss should be fired for incompetence.

As much as I hate to say it, that "OK boomer" meme may be appropriate there. Think about it though, by the time us young folk become the boss man, we can spout crap like "Intel!? Nothing but security risks" even if they had the better product.

 

[Topweasel]

In thinking about your post, regarding "close to an order of magnitude better in pricing" than previous market value, I would be interested to know where you got this data. EPYC 7601 was $131/core, 7742 is $108/core. That's not even close to an order of magnitude (at least as commonly used, 10x - though it's not even 2x the value, so I'm not sure what definition you mean when you say "order of magnitude").

IMO, "excellent" would be a better scaling of cost per core from bottom to top of the EPYC lineup, since cost of silicon doesn't rise exponentially with the amount you put on a chip, e.g. for excellent value, since the 8C/16T 7262 2P (boost 3.4GHz) retails for $575, I would anticipate a 64C/128T 7742 2P (boost 3.4GHz) might retail at $4600, plus a margin for binning I/O chiplets, for power consumption, etc., if they need to, though this doesn't intrinsically increase the cost of the production of the chip. Yet it retails for $6950, which is a 50% upcharge per core over the 7262. That's why it's not excellent in my mind. However, I understand that it's industry-standard to charge more for putting more power in a single socket, a "convenience charge" if you will. And it's in-line with their scaling on Ryzen and TR, roughly. And the market will still eat it up, as above, because plugging this much power into a 2P system will be worthwhile for a lot of people.

Another thing, the 7262 has 128MB L3 for 8 cores (16.5MB/core L1-L3), while the 256MB L3 for 64 cores (4.5MB/core L1-L3) -- which to me is somewhat striking. Well, anyway, I'm not a computer engineer. I'm sure those decisions have a reason. But considering the substantial die space cost of L3$, it lends even more credence to my thoughts that the 7742 is even less excellent from a pure value standpoint.

So, in my opinion, it's not excellent pricing when they're upcharging because they can, not because they need to. There's little "silicon" reason for them to price it where they have. Just because they're not, to use a colloquialism, screwing the purchaser as hard, doesn't mean that they're not still screwing the purchaser. It's just better than what is currently on the market from Intel. However, my key point about pricing is that their per-core pricing is worse than most of the rest of AMD's lineup and for no legitimate reason other than the convenience of having 64 cores in one socket, which does have value, but not to the point of a 50% upcharge. Hence, I wouldn't call it great or excellent or terrible. It's... not terrible. "Actually terrible" would be Intel's per-core pricing.

As for this forum not being a place for joking, that's your opinion, but it's not a forum rule, and it's fun, and I like fun, so I'll keep on doing it.

Regarding colloquialisms, it's very interesting that you use one in your own post. Regardless, I like colloquialisms, so I'll keep using them too.

I don't know about that. Your assessment doesn't place any value development, support, actual packaging, demand, or for being top dog. Those have to be worth something. Intel's demand was/is pretty high and they just rallied up the costs higher and higher. AMD is offering a CPU on a newer more expensive process with basically 2.5 the CPU power at 2/3rds the price, and the assessment is that they aren't being as price gouging? I get that threadripper pushes the boundaries of acceptable consumer CPU pricing and is surely them using their superiority to push ASP and margins up (though it was pretty obvious that based on Ryzen pricing and historical Threadripper pricing it was going to be ~$50 per core, and EPYC ~$100). But this is an extremely high level chip, being sold at an incredibly competitive price. There needs to be a balance out there between our quest for cheapest cost, and a companies desire for higher profitability. I really don't think the 7742 shifts that much toward the later. They could sell it for 2x that and it would still be a good value and honestly the density alone would be worth it to larger datacenters.

 

[Atari2600]

As much as I hate to say it, that "OK boomer" meme may be appropriate there. Think about it though, by the time us young folk become the boss man, we can spout crap like "Intel!? Nothing but security risks" even if they had the better product.

To me - its the danger of a boss that stops actually working and just manages.

If I'd my way, every manager in a company would have to spend at least 1 day a week doing end work.

In my line, that would mean managers must roll up their sleeves and write code, or do stress analyses, or design parts, or systems design etc etc etc. As soon as someone becomes solely a manager, their finger starts to come off the pulse.

I wonder when is the last time that boss actually deployed a server or a software stack - I'd say its probably in excess of 5 years.

 

[lobz]

To me - its the danger of a boss that stops actually working and just manages.

If I'd my way, every manager in a company would have to spend at least 1 day a week doing end work.

In my line, that would mean managers must roll up their sleeves and write code, or do stress analyses, or design parts, or systems design etc etc etc. As soon as someone becomes solely a manager, their finger starts to come off the pulse.

I wonder when is the last time that boss actually deployed a server or a software stack - I'd say its probably in excess of 5 years.

Believe it or not, there is a real need for managers in a bigger company, but I absolutely think they sould be smart enough to listen to actual professionals when making the decisions where their expertise my not be as apt.

 

[lobz]

Believe it or not, there is a real need for managers in a bigger company, but I absolutely think they sould be smart enough to listen to actual professionals when making the decisions where their expertise may not be as apt.